AMD FX "Bulldozer" Review - (4) !exclusive! Excuse for 1-Threaded Perf.

I think AMD should have given out more samples to us here at XS and this lack of support could have been fixed way before launch. We are almost doing all the R&D for them right now anyways. This was just a way to sloppy and a rushed release. I would of rather had a proper release with support over a half a$$ one with all this negative publicity hurting the product.

Quote:

---

Originally Posted by The Stilt

Indeed there should be a update coming for Windows which optimizes Turbo Core functionality on Zambezi.
Currently Windows (7 atleast) is throwing the load from core to core which sometimes neutralizes the effect of the Turbo Core feature.
This is because the load is not being run on the currently boosted core(s).

---

So you mean one that huddles threads as much as possible to enable max. turbo, right? But we want the opposite of it as the findings proves it's more beneficial to populate only one integer cluster per CU... So, we need only a rather little patch really that enables the SMT-aware scheduling Win7 already know, as some say.

Quote:

---

But guys... please...

When talking about Zambezi please use the correct terms to avoid any further confusion.

A Zambezi node consists of: Four compute units and eight cores.
Each compute unit contains two cores.

In some of the slides a compute unit was called as a module, however thats not the official term.

---

Well, in the patent papers they call the former a core and the latter an integer cluster. Not so surprisingly, I may add.
Also, some people just refuse to call the latter a core, anyway, because it's rather marketing than technics.
I would use compute unit or CU for the former and integer cluster for the latter.
(Although, I find the "compute unit" a little laboured and awkward.)

Quote:

---

Originally Posted by dess

So you mean one that huddles threads as much as possible to enable max. turbo, right? But we want the opposite of it as the findings proves it's more beneficial to populate only one integer cluster per CU... So, we need only a rather little patch really that enables the SMT-aware scheduling Win7 already know, as some say.

---

I´ll give you an example what currently happens:

A FX-8150 (Turbo Core enabled) running at stock settings, SuperPI (a single threaded software) is being executed:

Because there is load only on one thread (in theory) the Turbo Core feature boosts couple cores up to 4.2GHz while the rest are operating at 3.6 - 3.9GHz frequency. What currently happens is that Windows is unable to put the load (SuperPI) on the boosted core (4.2GHz) but throwing the load between the cores.
And executing such program (or any program actually) is naturally faster when it is executed on a core operating at 4.2GHz rather on one which is operating at 3.6-3.9GHz.

I know that, and this idiotic behaviour needs to be addressed, as well, of course. Just I'm not sure it's going to happen with Win7, as well. In the short term the lesser patch would also be fine as it would boost performance of lightly threaded apps, like games, without even max. turbo.

Quote:

---

Originally Posted by The Stilt

And executing such program (or any program actually) is naturally faster when it is executed on a core operating at 4.2GHz rather on one which is operating at 3.6-3.9GHz.

---

Not really any program. It's considerable faster to execute a program with 4 threads on 4 CU's, one cluster/CU, even at stock frequency (but usually all-cores turbo can work), than on 2 CU's, two cluster/CU, at max. turbo. Just see the findigs across the topic!

Quote:

---

Originally Posted by dess

That's fine (except your wording is inaccurate). Somehow we should trick it to use this method for BD, as well...



It depends on if the penalty of forcing those closely related threads to communicate through L3 (instead of L2) is more or less than the gain on the lack of sharing resources. It seems most applications only benefits from it:

Attachment 121261

So, there could be a little patch that simply enables scheduling a' la SMT in Win7, that it already supports (if true)...

Quoted from the article:


And so the default behaviour will be separation (contrary to what JF said all along)? Would be just stupid if not... Of course, power consumption is higher because more modules are active, but here we can see also that with turbo enabled the the energy efficiency is really the same...

Well, unless there is a fix coming (HW or SW or both) that largely improves on the penalty of sharing resoruces. Just because the current numbers are much worse (anywhere between 95% to 160%, with one case of 180%) than what they've propagated (180% across the board), and so one can think there is some flaw somewhere here, as well. (And there is indeed the case of L1D trashing, that they claim to be responsible for only 3% decrease.)


What diagram? Do you mean this? Which part of it?


Do you mean, if we disable every other "core" in the BIOS? Then no, you will get this:
Core (Module) 0 - one cluster
Core (Module) 1 - one cluster
Core (Module) 2 - one cluster
Core (Module) 3 - one cluster

ps. perhaps the title of the thread should be changed to "Thread separation vs. turbo", or something like that, to be more meaningful.

---

Sorry, I should have typed out:
Core 0 - one cluster
Core 1 - disabled
Core 2 - one cluster
Core 3 - disabled
Core 4 - one cluster
Core 5 - disabled
Core 6 - one cluster
Core 7 - disabled

...but I thought everyone would be smart enough to get the picture.

The Stilt is also correct about module vs CU, compute/computational unit is the correct term.

Quote:

---

Originally Posted by BeepBeep2

Sorry, I should have typed out:
Core 0 - one cluster
Core 1 - disabled
Core 2 - one cluster
Core 3 - disabled
Core 4 - one cluster
Core 5 - disabled
Core 6 - one cluster
Core 7 - disabled

...but I thought everyone would be smart enough to get the picture.

---

I've thought you've called a CU a core (just like others before you), not just because you were listing four of it two times, but that you used terms like "shared" and "one cluster" next to it. Why calling something two names, anyway?

Now, care to elaborate for the stupid like me what it all means and what diagram:

Quote:

---

AMD's diagram
Core 0 - shared
Core 1 - one cluster
Core 2 - shared
Core 3 - one cluster (uses all resources for 1 thread)

---

Quote:

---

The Stilt is also correct about module vs CU, compute/computational unit is the correct term.

---

Effectively everybody uses the term Module here, but fine, let it be "Compute Unit"... (No, not Computational Unit, while it means the same, this term is not used in this form.)

Quote:

---

Originally Posted by dess

I've thought you've called a CU a core (just like others before you), not just because you were listing four of it two times, but that you used terms like "shared" and "one cluster" next to it. Why calling something two names, anyway?

Now, care to elaborate for the stupid like me what it all means and what diagram:




Effectively everybody uses the term Module here, but fine, let it be "Compute Unit"... (No, not Computational Unit, while it means the same, this term is not used in this form.)

---

http://media.bestofmicro.com/fx-8150...-310584-13.jpg
That image shows Thread 1 (a/b) running sharing a CU while Thread 2 has it's own CU, does it not?
By disabling every other core in BIOS, Thread 1 runs on it's own CU, Thread 2 is disabled.
That is what I was describing. There is no Thread 2 a/b in the diagram despite the whole article talking about how AMD wants threads to share modules and benefit from higher p-states.


I would also like to add that though I never called you stupid, I find your post quite rude and very demeaning.

Quote:

---

Originally Posted by BeepBeep2

That image shows Thread 1 (a/b) running sharing a CU while Thread 2 has it's own CU, does it not?

---

Yes.

Quote:

---

By disabling every other core in BIOS, Thread 1 runs on it's own CU, Thread 2 is disabled.

---

No. That way all of Thread 1a, Thread 1b and Thread 2 would run on a separate CU.
(You can't "disable" a sw thread this way, anyway. If there are less cores enabled than threads to run, then those will simply share the available cores more.)

Quote:

---

There is no Thread 2 a/b in the diagram despite the whole article talking about how AMD wants threads to share modules and benefit from higher p-states.

---

Why should there be Thread 2b? SW threads are given. The diagram represents a situation when there are three threads to run. Two of them closely reladed (Thread 1a and 1b), i.e. working on the same dataset. The 3rd being a separate one, Thread 2. And then there are two cases, regarding core/CU utilization, one being sub-optimal and the other being optimal (shown above). The latter being so because the related threads share a CU (and so they can share data in L2 cache), while the separate one can have a whole CU to itself, and the unneeded CU's can go to sleep, enabling higher turbo mode for the first two CU.

Now, according to the findings, running all these three threads simply on separate CU's (so every other cores [clusters] disabled) would in most cases still be better than allowing them to share the CU's(*) in order to limit the number of used CU's, and so have higher turbo. Because with unsharing you usually win more than with turbo... (This shouldn't be the case, if everything worked as planned, or at least as marketed, but still, it is.)

* At least if done in the wrong way (Thread 2 and Thread 1a/1b in one CU). But it's possible it's true even if it's done in the "optimal" way. It needs further tests to tell.

Quote:

---

I would also like to add that though I never called you stupid, I find your post quite rude and very demeaning.

---

Well, if I was to you I would have asked for forgive being confusing, instead of what you wrote there.

i think this is going to get a very mixed opinion if your thinking about client vs server aplications

2 threads sharing a CU is going to be better for servers since that second thread is getting a pretty generous bonus for less additional power consumption.

for client side we have to keep in mind what happens when the cpu isnt running at 100% load. those same few threads sharing CUs would only be worth it if turbo was able to compensate. when the speedup was 1.8x a 10% bonus to clocks would work out fine. but since it seems we only get about 1.5x, we need turbo to increase speed by 25% for it work out. and that is not going to happen nomatter how much the process matures. so for client apps that use about 4 threads or less the optimal answer seems to be to split everything up (unless your trying to somehow save power, which is quite rare for desktops to think about ~10% efficiency.

i am curious about the whole 1a and 1b thread that want to share resources. i assume this means they use the same L2 to their advantage, but im curious if software can determine if they want to do that, and then can windows give them the right thread assignments to accomplish it? and even if they did, how much of a bonus would there be, if the end result is just a little less L2 being used then it could never be better than independent CUs when looking at it from just performance. but if the data sharing increases code efficiency then there could be alot more perf. but i dont really know much about this kind of stuff

Quote:

---

Originally Posted by The Stilt

Indeed there should be a update coming for Windows which optimizes Turbo Core functionality on Zambezi.
Currently Windows (7 atleast) is throwing the load from core to core which sometimes neutralizes the effect of the Turbo Core feature.
This is because the load is not being run on the currently boosted core(s).

But guys... please...

When talking about Zambezi please use the correct terms to avoid any further confusion.

A Zambezi node consists of: Four compute units and eight cores.
Each compute unit contains two cores.

In some of the slides a compute unit was called as a module, however thats not the official term.

---

This is AMD's patent for BD.

It points to the case that in a node there are 4 cores and 8 compute units or clusters. It also details how clusters share resources.

Disabling in bios "cores" ( actually clusters 150B ) 1 3 5 7 effectively disables resource sharing.

Attachment 121294

Quote:

---

Disabling in bios "cores" ( actually clusters 150B ) 1 3 5 7 effectively disables resource sharing.

---

Can I supose that the 256 bit FPU remain fully active ? so the fp performance will not be directly afected.

Quote:

---

Originally Posted by Manicdan

i am curious about the whole 1a and 1b thread that want to share resources. i assume this means they use the same L2 to their advantage, but im curious if software can determine if they want to do that, and then can windows give them the right thread assignments to accomplish it? and even if they did, how much of a bonus would there be, if the end result is just a little less L2 being used then it could never be better than independent CUs when looking at it from just performance. but if the data sharing increases code efficiency then there could be alot more perf. but i dont really know much about this kind of stuff

---

I wonder how Win8 judges which approach to choose "real-time". Perhaps it simply couples child-threads with their mother thread? That's not always good.

Anyway, if they're not going to implement it in Win7, as well, I hope at least they will enable the SMT-aware approach for BD (if Win7 really supports it already), that's still much better than the default one.

Quote:

---

Originally Posted by zhadoom

Can I supose that the 256 bit FPU remain fully active ?

---

Most probably, but worth a test.

Quote:

---

so the fp performance will not be directly afected.

---

Actually, it raises FP performance, as well, for less-threaded apps.

Quote:

---

Originally Posted by dess

I wonder how Win8 judges which approach to choose "real-time". Perhaps it simply couples child-threads with their mother thread? That's not always good.

Anyway, if they're not going to implement it in Win7, as well, I hope at least they will enable the SMT-aware approach for BD (if Win7 really supports it already), that's still much better than the default one.


Most probably, but worth a test.


Actually, it raises FP performance, as well, for less-threaded apps.

---

"real time" is mean to cpu's you can't alt Tab out of a game if you it on AOD i just tried on my fable 3 and GTA:IV i was stuck in game until i quit
it can stop key board and mouse movement

Quote:

---

Originally Posted by demonkevy666

"real time" is mean to cpu's you can't alt Tab out of a game if you it on AOD i just tried on my fable 3 and GTA:IV i was stuck in game until i quit
it can stop key board and mouse movement

---

I didn't mean the task priority option in Windows, I've used the term in a sense something happens just in time. Sorry if I wasn't clear enough.

Well the plot thickens, 23 chips later I came to a conclusion, while this sounds good on paper it's exactly why AMD won't release a 4core disabled this way.

Feel free to test this yourself but after testing i found that you will take a significant clock hit, so much that it defeats the gained effeciency.......

what do you mean exactly with clock hit? i dont think that you will get lower max clocks? how can that be?

or maybe u mean more clock needed that it eats the efficiency?

Quote:

---

Originally Posted by chew*

Well the plot thickens, 23 chips later I came to a conclusion, while this sounds good on paper it's exactly why AMD won't release a 4core disabled this way.

Feel free to test this yourself but after testing i found that you will take a significant clock hit, so much that it defeats the gained effeciency.......

---

So are you basically saying that one int "core" in a module does not matter on power usage almost at all?

Quote:

---

Originally Posted by Oese

what do you mean exactly with clock hit? i dont think that you will get lower max clocks? how can that be?

or maybe u mean more clock needed that it eats the efficiency?

---

No, he means that you get less clocks with 4 than 8.

Quote:

---

Originally Posted by chew*

Well the plot thickens, 23 chips later I came to a conclusion, while this sounds good on paper it's exactly why AMD won't release a 4core disabled this way.

Feel free to test this yourself but after testing i found that you will take a significant clock hit, so much that it defeats the gained effeciency.......

---

If I understand what you try to said this will reduce the events where the max turbo ( 4.2GHz ) happens. This major effect will be in 2 thread apps because will use two cores of different compute units ( modules or something like ... ).

Quote:

---

Originally Posted by chew*

Well the plot thickens, 23 chips later I came to a conclusion, while this sounds good on paper it's exactly why AMD won't release a 4core disabled this way.

Feel free to test this yourself but after testing i found that you will take a significant clock hit, so much that it defeats the gained effeciency.......

---

If I understand what you try to said this will reduce the events where the max turbo ( 4.2GHz ) happens. This major effect will be in 2 thread apps because will use two cores of different compute units ( modules or something like ... ).

Quote:

---

Originally Posted by BeepBeep2

No, he means that you get less clocks with 4 than 8.

---

I tought he was comparing to 2CU FX-4100, basically with 2CU you get so much more clocks that it greatly outweights gain on 4CU with second int core disabled.

If only turbo would be reduced, i would not mind if i could disable it whatsoever.

from what i hear, turbo is very unpredictable and bugged, sometimes cores clock below stock even when fully loaded and c&q off (which doesnt seeem to work at least on some boards..)

max oc anyhow should be same or higher with cores disabled? if not, that would be very strange..

Quote:

---

Originally Posted by chew*

Well the plot thickens, 23 chips later I came to a conclusion, while this sounds good on paper it's exactly why AMD won't release a 4core disabled this way.

---

We were mostly talking about the case of lightly threaded apps on a 81xx, and it definitely works there.

Quote:

---

Feel free to test this yourself but after testing i found that you will take a significant clock hit, so much that it defeats the gained effeciency.......

---

According to findings here and some other places, there is roughly a 10% performance advantage with 4M/4C at 3.6 GHz, compared to 2M/4C at 4.2 GHz.
In other words, even if the 4M/4C part would work only at 3.6 GHz, it would usually still be faster than a current 4170.
Of course, it more or less depends on the given application. What were your results?

---

[snip]

Quote:

---

Originally Posted by dess

You've got it wrong, dude. :) 100/90=1.111, but it's not 11%, but 90%! So, you should get the reciprocals, and that will show you the percentages.

Chess 11800/8813=1.3389 -> 74.7%
Wprime 13.814/9.531=1.4494 -> 69%
Winrar 4467/3027=1.4757 -> 67.7%
3d06 5803/4134=1.4037 -> 71.4%
3dvantage 19215/12102=1.5878 -> 63%
3d11 6340/4289=1.4782 -> 67.6%
CB R10 20552/15033=1.3671 -> 73.1%
CB R11.5 6/3.8=1.5789 -> 63.3%
Blender 9.76/7.16=1.3631 -> 73.3%
X264 37.23/25.18=1.4786 -> 67.6%

(These are the 4CU/8C vs. 4CU/4C numbers!)

---

i really dont follow your math,
if 4CU/8T gets 100, and 4CU/4T gets 90, thats a speed up of 11% through CMT. saying that things are 90% slower by turning off alternating cores just makes a confusing statement, even if true.

i think his numbers were right, CMT is giving us 34-59% speedup

as for chews point, i think from a stock chip perspective of 4CU/4T vs 2CU/4T, i think is a little wrong.
there was a perf/power consumption test and with turbo they all came out near identical, ill try and find the chart.

EDIT: this chart
http://www.hardware.fr/medias/photos...IMG0033838.gif

Manicdan: You're right, my mistake. I was playing with numbers that was about relative performance before, and when I seen this list it escaped my attention that these are speed-up numbers, so rightfully the opposites. I quess I need some sleep. :) (Edited out that part of my message.)

We are talking well over a 300 mhz variance clock hit with a 4cu/4core versus a hypothetical 4170 when it comes to stability for 24/7 useage.

Preety much we are almost talking deneb/thuban type 24/7 speeds making a 4cu/4core a moot point, if thats the case just drop a 955 in a 790 board.

you are referring to a theoretical stock part that needs to meet TDP, right?

how is the overclocking affected? I still cant believe without specific confirmation that oc-ability goes down with deactivated clusters, maybe in comparison to a 4170 it remains lower due to the more frontend logic and thus heat involved, and thats why it doesnt pay off?

well finally, these tests indeed show that perf/watt is better when sharing ressources, performance doesnt go down as much as the wattage decreases, so 4170 > 4C/4CU.

Question for me is, in lightly threaded workloads and with some working scheduler, I think we dont have these high power requirements? I mean the old thing, max power usage is compared with medium-threaded, hindered performance, whilst this architecture doesnt scale anywhere lineally in that aspect, so this doesnt make sense...

Could it be that the "deactivation" allows the threads to use the resources of the CU more efficiently (turning off a cluster may not turn off its resources, just the "thread" showing up as a core to windows) thus generating slightly more power usage?

Quote:

---

Originally Posted by STEvil

Could it be that the "deactivation" allows the threads to use the resources of the CU more efficiently (turning off a cluster may not turn off its resources, just the "thread" showing up as a core to windows) thus generating slightly more power usage?

---

My take on it is this, BD is a highly unbalanced chip so sure you might get 2 core 4 clusters to do 4.8 24/7 but the chance of running 4 core 4 cluster at 4.8 is slim to none. To do that we would need full cluster control and be able to deactivate cluster a and use cluster B if we wanted which currently is not possible.

I tested down to 4.5 4core 4 cluster and it was still failing prime versus 2 core 4 cluster doing 4.8 no issues..........could also be a bug in bios somewhere but atm that is wishfull thinking.

Air cooling? so the "8core" variants dont overclock high+stable anyway? 6100 ihave seen 4.8ghz 5ghz turbo already..

Quote:

---

Originally Posted by Oese

Air cooling? so the "8core" variants dont overclock high+stable anyway? 6100 ihave seen 4.8ghz 5ghz turbo already..

---

Yes air cooling.

I may be testing a tad diff than most.

Key settings are turbo off, HPC mode enabled, APM disabled and in power section PWM thermalling disabled.

Those settings will keep the CPU at full throttle 100% with no performance loss.

That said seeing 4.8 with 5 turbo without knowing all settings may not be what you think it is.

Also one of the key elements is when i shoot off these figures we are talking prime 95.

I can bench pifast at 5.2+........

Also heat is not an issue in my testing currently, with a heatsink swap i'm barely hitting 50C loaded so enabling other units is not heat related failure.

Quote:

---

Originally Posted by chew*

My take on it is this, BD is a highly unbalanced chip so sure you might get 2 core 4 clusters to do 4.8 24/7 but the chance of running 4 core 4 cluster at 4.8 is slim to none. To do that we would need full cluster control and be able to deactivate cluster a and use cluster B if we wanted which currently is not possible.

I tested down to 4.5 4core 4 cluster and it was still failing prime versus 2 core 4 cluster doing 4.8 no issues..........could also be a bug in bios somewhere but atm that is wishfull thinking.

---

You mean ability to disable a specific compute unit?
Or "cherry picking" the individual cores which OC best?

The boot strap CU can be changed by bios programming so in case the CU0 is bad it can be technically disabled.

Chew, alright thank you! what is this hpc option about anyway? :D

Quote:

---

Originally Posted by The Stilt

You mean ability to disable a specific compute unit?
Or "cherry picking" the individual cores which OC best?

The boot strap CU can be changed by bios programming so in case the CU0 is bad it can be technically disabled.

---

You got the right idea :up:

Currently although bios calls it core 1-8 even though reality is 0-7.

Anyway if you disable cu 3 cu 4 is auto disabled. What would be a real treat is to be able to disable cu 3 yet be able to use cu 4 instead and so on. I can't gurantee that that will alleviate what I am seeing but never know until you try.

Quote:

---

Originally Posted by Oese

Alright thank you! what is this hpc option about anyway? :D

---

Best way to explain it, vantage scores enabled can jump 500 points cpu score wise.

Quote:

---

Originally Posted by chew*

You got the right idea :up:

Currently although bios calls it core 1-8 even though reality is 0-7.

Anyway if you disable cu 3 cu 4 is auto disabled. What would be a real treat is to be able to disable cu 3 yet be able to use cu 4 instead and so on. I can't gurantee that that will alleviate what I am seeing but never know until you try.

---

Try C5F 9911 bios.
I asked Asus to add control for individual cores.
It works perfectly and you can do the 'cherry picking' (i.e disable CU2 and leave the rest active).
The only limitations are: You cannot mix compute units working in single & dual core mode or change the boot strap core (so hope your CU0 is good) ;)

Quote:

---

Originally Posted by The Stilt

Try C5F 9911 bios.
I asked Asus to add control for individual cores.
It works perfectly and you can do the 'cherry picking' (i.e disable CU2 and leave the rest active).
The only limitations are: You cannot mix compute units working in single & dual core mode or change the boot strap core (so hope your CU0 is good) ;)

---

I have it, thats what i was testing with.

Problem is if core 2 = disabled then core 3 must be disabled, and so on.

I know core 1 is always enabled and can not be disabled, thats not so big a deal though as is being able to control the rest.

I had 0062 (where we had asus first implement this ;) ) back when we did Lhe which is why we had this added but it gained us nothing and ended up setting record on 0057 instead.

It's why in the BD info thread i suggested that not everyone had acess to all pertinent info and or bios to gain the max potential out of BD ;)

Basically what i would like to try is this. If we speak on what the bios terms are 1-8.

1 is auto enabled and can not be disabled, thats fine. currently i can only do this 1/3/5/7.

What i want to try is this. 1/4/6/8 or any combo pair if possible.

in my current setup 1-4 is fine but 5 and 7 are weak so want to try 6/8

Yeah, thats a design limitation as far as I know.
The "slave" cores (2, 4, 6, 8) cannot work without the master core enabled, however I am not sure about this.

So let the hunt begin ;)

Quote:

---

Originally Posted by The Stilt

Yeah, thats a design limitation as far as I know.
The "slave" cores (2, 4, 6, 8) cannot work without the master core enabled, however I am not sure about this.

So let the hunt begin ;)

---

As of now i concede the hunt to someone else.

If i have time i will test through chips one last time but 23 chips is exhausting.

As for hunting through 23 chips i have 5 candidates that should do 5 gig prime 95 once on water with cores 1-4. They all do 4.8 sub 1.4v currently, some 4.9.

I guess I will have to retest them all for core 1/3/5/7 but it was looking grim even at 4.7 that way.

So far I still say 1.30 is the sweet spot vid ;)

I tested 4 recent chips and clocks increased greatly as i went up in vid from 1.25 to 1.30

The 1.25 chips could not even post at 4.8 sub 1.4v

Quote:

---

Originally Posted by chew*

We are talking well over a 300 mhz variance clock hit with a 4cu/4core versus a hypothetical 4170 when it comes to stability for 24/7 useage.

---

300 MHz is 6% of 5 GHz, while the gain in performance on 4CU/4C is up to 26% C2C! How is it not worth it?

Quote:

---

Originally Posted by chew*

Problem is if core 2 = disabled then core 3 must be disabled, and so on.

---

Do you mean you cannot enable core 3, even if you want, or it's just affecting OC badly?

Quote:

---

Originally Posted by STEvil

Could it be that the "deactivation" allows the threads to use the resources of the CU more efficiently (turning off a cluster may not turn off its resources, just the "thread" showing up as a core to windows) thus generating slightly more power usage?

---

I'm not sure how you mean it, but it definitely consumes less than all cores enabled (C2C):

Attachment 121333
From here.

Quote:

---

Originally Posted by dess

Do you mean you cannot enable core 3, even if you want, or it's just affecting OC badly?

---

Not without core 2 being active (0-7) because core 3 is a "slave" core.

Heh against my better judgement I fired up rig one more time just to see if i missed something.

I found the wild card so to speak.

When disabling cores 2/4/6/8 ht must match NB.

When running 1/2/3/4 this appears to be a non issue.

It definitely needs further investigation.

Now the delta is not half as bad.

1/2/3/4 = 1.375 set for 4.8 gig

1/3/5/7 = 1.40 set for 4.7 gig

This was what i thought, there must be some quirks involved, physically there's no real cause for this imo..

Quote:

---

Originally Posted by Oese

This was what i thought, there must be some quirks involved, physically there's no real cause for this imo..

---

Well also you must keep in mind currently I am thrashing IMC purposely with mismatched 16g and 15gig usage in prime 95 as well + 64 bit win7 for a worst case scenario.

Aha alright :)

Chew what's the highest NB you gotten on air? And ln2?

Ok not quite what i was hoping for, have to test more benches....

So stability or a rough idea at least needed for 1/3/5/7, same identical settings will run 4.8 with 1/2/3/4.

Attachment 121334

Bottom line resource hit or not some apps are going to favor speed.......whether this only pertains to single threaded needs more testing. 3dmark 01 comes to mind.

Attachment 121335

Attachment 121336

Attachment 121337

Attachment 121338

Attachment 121339

Attachment 121340

Only the multi threaded bench gained with less clocks here.

Quote:

---

Originally Posted by Sgt.McRuff

Chew what's the highest NB you gotten on air? And ln2?

---

With 16g ram 2800 was best. Ln2 4500 ish through vantage.

i think i see what chew is getting at

if we overclock to the max with 2 CUs and 4 threads, vs 4 CUs and 4 threads, you can OC higher because heat wont be your issue nearly as quickly. we saw that in low threaded apps 5% gains are expected from not sharing CUs (in gaming for example), while he can get more than 5% higher clocks which will offset that value.

so i think the best thing is to just leave all cores there, OC high for full load, but then use turbo to OC lower threads even higher if possible.
so with a normal OCing air cooler, like ~4.7ghz on all CUs with a moderate voltage, but 5.0-5.2ghz for 1-2 CUs
i would like to know if using turbo with overclocking is as easy as it was with thuban, any insight on this guys?

Hey chew* .. you also want to try this Chess test on your FX-4110 or/and on your other FX's you have!

You can easy select from 1 core to 32 cores ,ofcourse i want to put your best results in my list.
I use always chess engines to overclock my system and to run stable 24/7 .. just give it a try ;)

You find explenation and software below in my signature.. setup need max. 3sec. if you have done the settings first time.

It can go even faster if you use Skype to help you with the setup.

JP.

Now I really have mixed feelings about running like this. It almost feels like a lose lose situation.

One of the key things you need to note here is not the overall scores or the cpu portions of the tests because it's quite obvious as was already observed without resource sharing multithreaded improves.......the issue is higher clocks benefit game tests even when resource sharing.

4600 results = 1/3/5/7
4800 results = 1/2/3/4

Attachment 121342

Attachment 121343

Attachment 121344

Attachment 121345

Attachment 121346

Attachment 121347

Attachment 121348

Attachment 121349

hmm here 1/3/5/7 is faster at lower clocks...

now the question is, which setting is consuming less energy? In (one of) these tests?

Still, it was shown that in the end ressource sharing provides better perf/watt, given same clocks. Now, with less clocks on the non-sharing variant, idk if this would change? Or, due to higher voltage, its even more the case?

My thinking is still

1.) BD 8-core doesnt consume that much in situations where not all cores are loaded, or cores are hindering each other (whilst this would be a case where some hindrance might consume power itself, like cache misses and chache traffic due to this), therefore one should use bulldozer as it is and dont worry too much about consumption, its only bad in full-load-conditions and doesnt affect perf/watt in medium loaded situations

2.) BD running as 4-core 1/3/5/7 is more power efficient, due to less clocks for same Performance and disabled (not-needed-for-gaming) additional cores, whilst it is better then some 1/2/3/4 BD

3.) For Gaming workloads, anything then 4100 BD has too many cores and too much overhead, so go for this.

Background is, i'd like to buy a 8-core, cause this has the most opportunities and i dont like disabled chips (case is different if those can be enabled however :D ), and somewhere i think for megatasking 8 cores would be nice, if i could have good perf/watt in lower threaded/loaded situations.

The Win8 scheduler seems to provide this, either in 1/2/3/4 or in 1/3/5/7 configurations.

Again the question, chew*: Those 8-cores clock higher when you go for 1/2/3/4 compared to all cores enabled, or do they clock lower then all cores enabled with 1/3/5/7 ??

Quote:

---

Originally Posted by Manicdan

if we overclock to the max with 2 CUs and 4 threads, vs 4 CUs and 4 threads, you can OC higher because heat wont be your issue nearly as quickly. we saw that in low threaded apps 5% gains are expected from not sharing CUs (in gaming for example), while he can get more than 5% higher clocks which will offset that value.

---

Well, it seems it's true only for 3DM2001... ;)

Quote:

---

Originally Posted by dess

Well, it seems it's true only for 3DM2001... ;)

---

theres proof of that from a review showing MOST games they tested, not just chews test of synthetic gaming benchmarks.

Quote:

---

Originally Posted by Manicdan

theres proof of that from a review showing MOST games they tested, not just chews test of synthetic gaming benchmarks.

---

Could you please provide a link for that?

Some games at least gain 10% this way. I'm not even sure if it's including (higher) turbo for the shared mode, so it could be more C2C.

http://media.bestofmicro.com/N/G/310...wow%202560.png

Attachment 121353

Sure, what about GT1 in vantage ( the only GT that matters is GT1 for cpu since gt2 is all about GPU and the diff for that can be bench run variance ), or another GT in 3d 11........

I can always add 05 and 03 gametests to the mix, they are indicative of 01......not surprising since they only benefit from 2 cores.

Bottom line this is not the holy grail fix due to the fact that there is a clock hit.

I haven't gotten to actual games yet but i can already see a trend, if not full loading 4 cores the advantage goes to clock speed..........

This test is far more real world due to fact that in real world clocks will vary betwen the overclocks.......running at same speed is pointless for a compare since headroom quite obviously varies enough to matter.

If the findings bother people it's not a big deal, this is stuff i'm looking at for my own daily rig, I can keep the info to myself.

I'll give 8 cores a shot later but i was pegging over 60C....even at my rather conservative settings.

Quote:

---

Originally Posted by dess

Could you please provide a link for that?

Some games at least gain >10% this way. I'm not even sure if it's including (higher) turbo for the shared mode, so it could be more C2C.

http://media.bestofmicro.com/N/G/310...wow%202560.png

Attachment 121353

---

i think your mixing up statements and there were quite a few

A: using 1 thread per CU helps gaming performance, TRUE
B: using 2 threads per CU and overclocking farther can in many cases give more perf if your not on a super strong WC system, also TRUE

heres a link with details for 'A'
http://www.hardware.fr/articles/842-...acite-cmt.html
4/6 games got ~5% faster when using 1 thread per CU (but at the same clocks), the other 2 games were identical, aka no penalty.

the windows 8 bonus i hope is not just about making the turbo work better by ensuring threads stick to modules so the others can turn off and reduce TDP, since thats something we should have working now with very simple fixes rather than needed to purchase a brand new OS.

Thanx for all the interesting info/tests. AMD chips, as usual, are always fun to play with.

I currently have a 1055T in my Sabertooth 990. You guys recon it's worth getting DB?

I really hope they come out with a patch for Windows 7 that includes some of the scheduler stuff Windows 8 has once it is all perfected. I like my Windows 7, I dont want to have to change to Win8 just for better performance on this CPU.

Quote:

---

Originally Posted by Manicdan

i think your mixing up statements and there were quite a few

---

I don't think I'm mixing up anyting.

Quote:

---

A: using 1 thread per CU helps gaming performance, TRUE
B: using 2 threads per CU and overclocking farther can in many cases give more perf if your not on a super strong WC system, also TRUE

---

C: Depending on the workload, sometimes it's A, sometimes it's B that gives you the higher performance...

Quote:

---

heres a link with details for 'A'
http://www.hardware.fr/articles/842-...acite-cmt.html

---

Thanks, but I've inserted charts here from this article already.

Quote:

---

4/6 games got ~5% faster when using 1 thread per CU (but at the same clocks), the other 2 games were identical, aka no penalty.

---

Nice, but it's just a few games.
Edit: I think you're reading the chart wrongly. All games gained some (2-12% [and 40% in case of Houdini 2, although it's not of typical game load]) with 4CU/2C, compared to 2CU/4C.

Quote:

---

the windows 8 bonus i hope is not just about making the turbo work better by ensuring threads stick to modules so the others can turn off and reduce TDP

---

This diagram (and the explanation here) suggests that the default mode for unrelated threads is to execute them without sharing, if possible.

Quote:

---

since thats something we should have working now with very simple fixes rather than needed to purchase a brand new OS.

---

What fixes do you mean?

Regarding utilizing cores with priority on unsharing (edit: I mean automatically), and the need for Win8: I've already pointed out it should be easy to do it on Win7 by simply enabling SMT-aware scheduling (if it's true it's already supports it in case of Intel CPU's with HT). Although, in some cases you will prefer to disable it.

Quote:

---

Originally Posted by dess

I don't think I'm mixing up anyting.
C: Depending on the workload, sometimes it's A, sometimes it's B that gives you the higher performance...
Thanks, but I've inserted charts here from this article already.
Nice, but it's just a few games.

---

you just proved your confusing people by mixing statements

A is about not sharing resources, B is about using extra TDP to overclock/turbo higher, it can only be both if you leave the chip stock and try to play with schedulers to see the results. if you have 2 chips with the same clocks, then A will be better. but if you have 2 chips with the same TDP then B will be better. you cant say that C is either because they are opposite sides of the same coin.

and just a few games is the only games weve seen tested. so for now its 66% of games out there do notice an increase by not sharing resources IF CLOCKS ARE LEFT ALONE. the test with WoW by Toms Hardware mentioned no details about how that 10% was gained, and it looks to be based around 'B' with using higher turbo settings. but they dont tell us anything.

Quote:

---

Originally Posted by Manicdan

i think your mixing up statements and there were quite a few

A: using 1 thread per CU helps gaming performance, TRUE
B: using 2 threads per CU and overclocking farther can in many cases give more perf if your not on a super strong WC system, also TRUE

heres a link with details for 'A'
http://www.hardware.fr/articles/842-...acite-cmt.html
4/6 games got ~5% faster when using 1 thread per CU (but at the same clocks), the other 2 games were identical, aka no penalty.

the windows 8 bonus i hope is not just about making the turbo work better by ensuring threads stick to modules so the others can turn off and reduce TDP, since thats something we should have working now with very simple fixes rather than needed to purchase a brand new OS.

---

This is so fascinating, that i had to register at this forum. I spent a lot of time this weekend, analyzing the french and the results from the beginning of this thread. And it seems that the performance gain of deactivating cores in the BIOS, is higher than that what the guys from France did. They see 2M/4C in sum 15 % behind 4M/4C. In the first post here it was reached 22 %. Just have a look at Fritz Chess. Here 8813/6335=139 % and they tested 8130/6417=127%. So if you try to reach 4M/4C(not gaming) from a base of 4 GHz, you have to clock the 2M/4C at 5 GHz. The power consumption would be significantly higher. This may change, when a native designed 2 computing unit chip for FX-4X00 is released.

The best effort in gaming may be deactivating 1 CU in bios, so you get a FX-6X00. And then take a piece of software, like "core affinity resident", for automaticaly binding the games to the right 3M/3T cores, simulating the windows 8 scheduler. Overclocking should be better than with 4M/4T and you get the advantage of 1 thread/CU.

I think that overclocking is not the best idea to gain more efficient performance, when you get a power draw of >500 watts. If you take a look at the effects of northbridge oc in games:
http://www.rage3d.com/reviews/cpu/am.../index.php?p=9
than i favour 4CU/4C or 3CU/3C with nb oc. That should only fill the saved watts between power draw of FX-8150 in standard configuration and 4CU/4C or 3CU/3C:
http://www.hardware.fr/medias/photos...IMG0033837.gif
so you get from 84 to 110 watts without turbo. You can see that the turbo in 2CU/4C mode is horrible inefficient. It raises the power draw from 60 -> 95 watt = 60 % and gains performance of 11 %. It is horrible.

So pleeeeeaaase, benchmark the 4CU/4C with overclocked nb and 3,6 or 4GHz!

The guys here:
http://www.hardocp.com/article/2011/...mance_review/5
got 7-10 % better results with DDR-1333 instead of all the reviewers with 1866. Memory bandwith is bad, but only WinRar is not amused about that. Drystone, Lame, WPrime, ... all with better results through all of the benchmarks.

If you imagine 5-10 % 4CU/4C + 5-10% nb oc + 5-10 % memory miracle = 15-30 %. Its worth a look to bring them together in benchmarking. :yepp:

Quote:

---

Originally Posted by Manicdan

you just proved your confusing people by mixing statements

---

No, I don't. I was saying in some cases it's unsharing, in other cases it's turbo that gives higher gains. It's not contradicion.

Quote:

---

A is about not sharing resources, B is about using extra TDP to overclock/turbo higher, it can only be both if you leave the chip stock and try to play with schedulers to see the results. if you have 2 chips with the same clocks, then A will be better. but if you have 2 chips with the same TDP then B will be better. you cant say that C is either because they are opposite sides of the same coin.

---

I wasn't speaking about doing both (using 4CU/4C mode and max. turbo) at the same time. Although, disabling every other hw thread lowers power consumption, thus making some room for lifting the clock, as well. Also, you can have max. turbo with 2 CU's and executing a dual-threaded program with 1 thread/CU.

Quote:

---

but if you have 2 chips with the same TDP then B will be better.

---

Prove it.

Quote:

---

and just a few games is the only games weve seen tested. so for now its 66% of games out there do notice an increase by not sharing resources IF CLOCKS ARE LEFT ALONE.

---

Seems you've missed my edit: I think you're reading the chart wrongly. All games gained some (2-12% [and 40% in case of Houdini 2, although it's not of typical game load]) with 4CU/2C, compared to 2CU/4C.

Quote:

---

the test with WoW by Toms Hardware mentioned no details about how that 10% was gained, and it looks to be based around 'B' with using higher turbo settings. but they dont tell us anything.

---

There are three indications (the above results with the games, the diagram and the explanation for it) that it's rather A. Or perhaps both (with 2 CU's).
Wouldn't they scale more similarly with turbo alone?

Quote:

---

Originally Posted by dess

No, I don't. I was saying in some cases it's unsharing, in other cases it's turbo that gives higher gains. It's not contradicion.


I wasn't speaking about doing both (using 4CU/4C mode and max. turbo) at the same time. Although, disabling every other hw thread lowers power consumption, thus making some room for lifting the clock, as well. Also, you can have max. turbo with 2 CU's and executing a dual-threaded program with 1 thread/CU.


Prove it.



Seems you've missed my edit: I think you're reading the chart wrongly. All games gained some (2-12% [and 40% in case of Houdini 2, although it's not of typical game load]) with 4CU/2C, compared to 2CU/4C.


There are three indications (the above results with the games, the diagram and the explanation for it) that it's rather A. Or perhaps both (with 2 CU's).
Wouldn't they scale more similarly with turbo alone?

---

chew is proving 'B' and im taking his answer as being accurate.
for the games i was looking at 4/4 being faster than 4/8, i completely ignored 2/4 because thats comparing a 4100 to an imaginary 4/4 chip. the perspective i was trying to show with point 'A' was the INCREASE in performance by chopping up an 8100's cores. (i have no idea what houdini 2 is, and based on details it looked like a multi-threaded program rather than a game)

i still belive the WoW example by TH was done by turbo alone. that game uses less than 3 full cores, but has 3 distinct threads, and the 2 weaker ones in total are weaker than the first one alone. so its possible they got it down to one CU and just let it max out at 4.2ghz. but if left untouched it might have used up 3 CUs and stuck between 3.6 and 3.9ghz. that game is actually rather important to me since thuban isnt that great at it, and it is very cpu limited, but also very limited to single threaded perf. i really wish TH did more talking about the results and not just the theories. showing us which threads were parked for the game, or what the clock speed was in their test. all we know is that Win8 helps, but it can be quite a few combinations to make it work out best. i wonder if Win8 will be smart enough to test whether using a single CU for the game and max out turbo will be better than using 2 CUs where it MIGHT still be able to near max out, but also give the heaviest thread its own resources.

Quote:

---

Originally Posted by Manicdan

chew is proving 'B' and im taking his answer as being accurate.

---

I think you should study his results more closer...
Which answer?

Quote:

---

for the games i was looking at 4/4 being faster than 4/8, i completely ignored 2/4 because thats comparing a 4100 to an imaginary 4/4 chip.

---

What's the point in doing so in case of games with 4 threads or less?

Max. turbo is enabled only if there are 2 active CU's. So, if you want to know if it's either max. turbo or unsharing that gives you more performance with 2-4 threaded games, it's indeed 4CU/4C vs. 2CU/4C that you should take for a comparison...

Quote:

---

the perspective i was trying to show with point 'A' was the INCREASE in performance by chopping up an 8100's cores.

---

The whole topic is all about this (for programs with 2-4 threads), why do you think you need to try to show this?

Quote:

---

(i have no idea what houdini 2 is, and based on details it looked like a multi-threaded program rather than a game)

---

AFAIK it's a well-threaded chess game.

Quote:

---

i still belive the WoW example by TH was done by turbo alone. that game uses less than 3 full cores, but has 3 distinct threads, and the 2 weaker ones in total are weaker than the first one alone. so its possible they got it down to one CU and just let it max out at 4.2ghz.

---

You can get max. turbo with 2 CU's.

There are two probable scenarios:
- Execute the more demanding thread on the first CU alone, the two weaker ones on the second CU, get max. turbo activated, and let the other CU's go to sleep.
- Execute all three threads alone in 3 CU's, at all-cores turbo.

Now, I don't think it runs even on Win7 at stock 3.6 GHz all the time... Given it's only 3 threads, it can run partially even at max. turbo there, only there is some penalty because of changing cores and so switching frequencies, etc. At least all-cores turbo can be active most of the time. Thus, 10% gain Win7 vs. Win8 can't come alone from turbo.

OK dirt 2 all same settings GPU wise.

4 threads on all systems.

Screenshot to show cpu usage of bench.

Attachment 121367

Sandy

Attachment 121365

4600 1/3/5/7

Attachment 121368

4800 1/2/3/4

Attachment 121369

Now to find a game that uses 2 threads.

Tested the 2 core theory out or "lightly threaded cpu app" now i just need a game to prove my point ;)

4600 1/3/5/7
Attachment 121370

4800 1/2/3/4
Attachment 121371

The real kicker is heaven doesn't gain to much from CPU so this is definitely a significant gain from clock speed.

Quote:

---

Originally Posted by chew*

Tested the 2 core theory out or "lightly threaded cpu app" now i just need a game to prove my point ;)

4600 1/3/5/7
Attachment 121370

4800 1/2/3/4
Attachment 121371

The real kicker is heaven doesn't gain to much from CPU so this is definitely a significant gain from clock speed.

---

Try Crysis Warhead or Crysis. I prove CrWh on Intel i5 760@3,3 GHz and HD6950, 640x480 lowest and got:

2 cores 238 average
3 cores 243 average
4 cores 243 average

3,4 or less doesnt matter, but 2 cores got well busy during benchmark run.

HL2 lost coast is enough to prove my point.

4600 1/3/5/7
Attachment 121375

4800 1/2/3/4
Attachment 121376

And to show this game is impacted by clocks

Attachment 121377

Both the sandy rig and the 4800 BD run are score limited due to a 300fps cap.....

Quote:

---

Originally Posted by chew*

4600 1/3/5/7

Attachment 121368

4800 1/2/3/4

Attachment 121369

---

So it looks like in Dirt 2 that 4 modules with 1 thread each at a slower speed does much better than 2 modules using both cores at a higher speed

Quote:

---

Originally Posted by EniGmA1987

So it looks like in Dirt 2 that 4 modules with 1 thread each at a slower speed does much better than 2 modules using both cores at a higher speed

---

Yep basically I came to the conclusion that any game that uses 4 cores heavily ( like 75% or better ) will benefit from 4core 4 cluster configs........problem is the clock hit it takes.

Preety much any game that is a console port (75% games made for pc? :rolleyes: ) that will only 50% load 4 cores or load 2 cores heavily will benefit from a 2core 4 cluster config due to the higher clocks attainable in that configuration.

I've tested like 23 chips like this, all of them take a clock hit in the 4core 4 cluster config, some much worse than others.

Quote:

---

Originally Posted by dess

snip

---

the idea of comparing 4/8 with 4/4 instead of with 2/4 is to isolate things to a single detail and test the crap out of that detail to understand it. if you add in clock rates or TDP then your looking at a scenario closer to reality, but not with scientific precision.

what chew is doing is great, but he would need to do about a dozen more tests with each scenario to cover everything. so instead he built a few real life examples and showed when either is better. i would just like to see if he can set up a turbo on the 4/4 to let it get a little higher clock rate when its a low threaded game. he might only get an extra 100mhz, but it might be enough to make it better in many more cases. i havnt seen enough examples of how turbo can be set up with overclocks and how well its working once set up.

Quote:

---

Originally Posted by Manicdan

the idea of comparing 4/8 with 4/4 instead of with 2/4 is to isolate things to a single detail and test the crap out of that detail to understand it. if you add in clock rates or TDP then your looking at a scenario closer to reality, but not with scientific precision.

what chew is doing is great, but he would need to do about a dozen more tests with each scenario to cover everything. so instead he built a few real life examples and showed when either is better. i would just like to see if he can set up a turbo on the 4/4 to let it get a little higher clock rate when its a low threaded game. he might only get an extra 100mhz, but it might be enough to make it better in many more cases. i havnt seen enough examples of how turbo can be set up with overclocks and how well its working once set up.

---

I think it would just be easier to set up 2 profiles in bios, load whatever one will benefit you most for the app your running. Thats all i'm doing now.

As far as 8 core operation I tried 4.6 on this air cooler at this voltage, no dice.

Quote:

---

Originally Posted by chew*

Yep basically I came to the conclusion that any game that uses 4 cores heavily ( like 75% or better ) will benefit from 4core 4 cluster configs........problem is the clock hit it takes.

---

Why is it a problem for them if those are still faster this way than with the higher OC?

Quote:

---

Originally Posted by Manicdan

the idea of comparing 4/8 with 4/4 instead of with 2/4 is to isolate things to a single detail and test the crap out of that detail to understand it. if you add in clock rates or TDP then your looking at a scenario closer to reality, but not with scientific precision.

---

Yes, this way you understand that an up to 4-threaded app won't scale any further with more cores than 4. What a revelation... ;)
And you will learn nothing about CMT scaling (in case of up to 4-threaded apps, again).

Quote:

---

what chew is doing is great, but he would need to do about a dozen more tests with each scenario to cover everything. so instead he built a few real life examples and showed when either is better. i would just like to see if he can set up a turbo on the 4/4 to let it get a little higher clock rate when its a low threaded game. he might only get an extra 100mhz, but it might be enough to make it better in many more cases.

---

Do you mean, to see if an even higher clock is possible in 4CU/4C mode, but with 2CU's gone to sleep? Could be.

Quote:

---

i havnt seen enough examples of how turbo can be set up with overclocks and how well its working once set up.

---

I think it can be done by modifying the VID and FID values that are assigned to the various power states.

Quote:

---

Originally Posted by chew*

I think it would just be easier to set up 2 profiles in bios, load whatever one will benefit you most for the app your running. Thats all i'm doing now.

---

How great it would be if it all worked full automatic... :) Could it made work under Windows with some driver? Or only MS could do it?
Well, in times like this Linux would come handy...

Quote:

---

As far as 8 core operation I tried 4.6 on this air cooler at this voltage, no dice.

---

What if you enable only two cores in two CU's...?

Quote:

---

Originally Posted by informal

Jack, hardware.fr already tried something along those lines with 4m/4t and 2m/4t ,both with Turbo on. In 1st case maximum turbo for all 4 "threads" is 3.9Ghz since all modules are running. In second case it's 4.2Ghz across 2 modules(4 threads). The % difference in Turbo clock(~7%) is not nearly enough to make up for sharing losses as can be seen here:
http://www.hardware.fr/articles/842-...windows-8.html
Attachment 121226

4m/4t is 26% faster(!) than 4m/2t at fixed 3.6Ghz and 15% faster when both are running their maximum Turbo modes allowed. Now comes the power draw story.
If you look at the power draw you will see the faster config is 20% more power hungry and I suspect this is the reason why AMD didn't configure the core priorities in that way. I think when PD arrives,power draw will go down sufficiently in order to schedule the threads the faster way and still get good power numbers. Still,with present BD core, for 20% more power you gain 26% more performance this way,not a bad tradeoff. If GloFo would get their act together and make possible for AMD to produce 3.6Ghz 5 module PD core with this thread affinity capability,this thing could very well be significantly more powerful than Thuban ,even in ST at fixed clock and noticeably more powerful than BD in both ST and MT with both Turbo on and off.

By the way,great thread DGLee :)

---

This almost supports the ex engineers whine about AMD not personally crafting the initial chips.

maybe clock hit depends on sth like bootstrap cu need 2 cores?

Like enable first cu fully, rest only one cu/c

sth like this ;)

Quote:

---

Originally Posted by Oese

Like enable first cu fully, rest only one cu/c
sth like this ;)

---

Thats mixing and it is not possible ;)

Is it not possible just because the bios isnt there yet or because it actually isnt supported to disable one core on the rest from the processor itself?

Quote:

---

Originally Posted by EniGmA1987

Is it not possible just because the bios isnt there yet or because it actually isnt supported to disable one core on the rest from the processor itself?

---

It is a design limitation as far as I know.
You can run the compute units either in single or dual core mode but not at the same time.

Quote:

---

Originally Posted by Oese

maybe clock hit depends on sth like bootstrap cu need 2 cores?

Like enable first cu fully, rest only one cu/c

sth like this ;)

---

maybe they need to seperate l3 into 4 segments of two cores each. maybe the atx form factor is reaching its limits and intel need to redesign the damn thing for 8+ cores

Quote:

---

Originally Posted by The Stilt

It is a design limitation as far as I know.
You can run the compute units either in single or dual core mode but not at the same time.

---

but is it a platform limitation? lets face it its the same freakin platform as the 90's bs intel atx. maybe the entire platform needs to evolve.

great work chew, this may not be what everyone was hoping to hear but it will surely be of assistance to people reading up to see whther BD fits their needs or not! :clap:

So on to the juicy stuff, your findings are consistent with a classic resource sharing bottleneck, or so it seems..
To sum it up:
- when a CPU heavy task fully loads both clusters in a CU, the shared resources have a hard time feeding or dealing with both clusters so you notice a bottleneck and performance per thread drops accordingly -> ergo, 4c/4cu wins
- when only one cluster is loaded, or both clusters are being used, but one of them (or both) is/are not being fully loaded, shared resources can keep up and so performance per thread takes a lesser/no impact -> ergo, clockspeed wins

This is fully understandable and seems perfectly logical

Next subject, max clocks: :up:
I'm still confused by something here, please see which interpretation is the correct one (or if both apply):
- 4c/2cu has lower power/thermal footprint, ergo for the same voltage, same cooling, it will clock higher?
or
- Luck of the draw means that you're likely to get different OC potential on all modules, and therefore you might (probably) end up with having to choose a cluster from a module with worse potential when taking the 4c/4cu approach, therefore limiting overall clocks a bit?

Also, for this last part, do you think it plausible that with better air or water cooling (for instance, i have 360 rad custom water loop) and maybe some extra volts one could push those limits higher and reach higher clocks? Perhaps 4.8 ~ 5Ghz across all cores? or is this a pipe dream for this stepping?

Last question:
Have you any idea of the actual power consumption difference between running 4c/2cu vs 4c/4cu vs 8c/4cu @ same clocks all around, for instance, 4.6Ghz? I am considering an upgrade but at this point im afraid my TX650 would not cut it powering a BD 8c OC plus a 6970 or 6870 CF?

My personal opinion on this:
Best case scenario (and the most balanced option for varied daily usage) would be for us to be able to leave them all on, and have AOD or something clock each loaded core higher.. While at the same time having some way to make the os treat one of the clusters in each cu as a "Hyperthreading-like" affair... However im guessing this would require a hefty power draw and adequate high end cooling to control..

Crappy math:
So taking into account an announced (and somewhat accurate, from what we can see) performance penalty of roughly 20% when fully loading both clusters (ex: Civilization or starcraft) we can speculate that it would take roughly 5 ~ 5.2Ghz clocks all around for a "FX-4110" to match performance of a PII x4 @ 4 ~4.2 Ghz. Whereas for less intense games with a 2main, 2 secondary threads would require only around 4.3 to 4.4Ghz.

This would be an avantage to the octa-cores that they could "shine" in both scenarios, we would just need a way to make this hassle free and be able to pair this up with NOT having to reboot or manually change these settings whenever we want to game, or encode, or whatever :P

once again, thank you guys for your awesome work! and also everyone sorry for the long post, feel free to ignore if you wish :p: but i just wanted to see if I had all the facts straight from this "investigation" :D

Please, do correct my conclusions if you seem fit as I have a feeling these are some of the points other people could benefit from aswell :D

Quote:

---

Originally Posted by Manicdan

the idea of comparing 4/8 with 4/4 instead of with 2/4 is to isolate things to a single detail and test the crap out of that detail to understand it.

---

The answer could be simple. If you use 2CU/4C, there are 2 floating point units available, in 4CU/4C mode 4. So a task using 4 cores, gets access to 4 vs. 2 fpus. The performance gain in 50% int and 50 % fpu scenarios, is up to 25 %. Xbox ported games, mostly use 2 cores. So they get access to 2 fpus in both scenarios and you see performance gain of about 5 % only, from doubled cache per core.

Console ports are not the problem, there is always enough power for a FX6 or FX8 to reach >60 fps. The problem is Dirt, F1, Civ5, Crysis 2,... . Northbridge overclocking and 4CU/4C can be the difference between playable or not.

i just installed win 8 beta, on a thuban system though. It seems, its stable enough to work for some gaming, and i guess this scheduler is more aware of what BD needs. I am looking forward to tests coming in at the other thread covering this..

Quote:

---

Originally Posted by omninmo

Last question:
Have you any idea of the actual power consumption difference between running 4c/2cu vs 4c/4cu vs 8c/4cu @ same clocks all around, for instance, 4.6Ghz? I am considering an upgrade but at this point im afraid my TX650 would not cut it powering a BD 8c OC plus a 6970 or 6870 CF?

---

I'm over 650W total load with a oc'ed 1090t and two 6950's @ 6970. So I'd say single GPU only, with that PSU.

Quote:

---

Originally Posted by omninmo

great work chew, this may not be what everyone was hoping to hear but it will surely be of assistance to people reading up to see whther BD fits their needs or not! :clap:


Have you any idea of the actual power consumption difference between running 4c/2cu vs 4c/4cu vs 8c/4cu @ same clocks all around, for instance, 4.6Ghz? I am considering an upgrade but at this point im afraid my TX650 would not cut it powering a BD 8c OC plus a 6970 or 6870 CF?

---

Take a look here at the picture "Fritz Chess benchmark 4.3 (watts atx 12v)":
http://www.hardware.fr/articles/842-...windows-8.html

2/4 = 60 watts (100%)
4/4 = 84 watts (140 %)
2/4 turbo = 95 watts (158 %)
8/8 = 101 watts (168 %)

totally power consumption in your scenario (oc and with HD6990) is about 500 watts in peak:
http://www.rage3d.com/reviews/cpu/am.../index.php?p=7

To reach 4/4 @stock in dirt with 2/4, you nearly have to double the power draw. In other games (console ports) its useless, because fps are mostly >60 fps and there is no gain in performance, gpu is limiting.

There are two different methods to get what you want:

1. let the FX-8 or FX-6 work in stock mode and bind the game/the app manually through the task manager to 1357 or 0246 or use a tool like "amd overdrive" or "core affinity resident" to build a profile; you do not need the expensive asus board and simulate the windows 8 task manager in a smarter way than inefficient cu parking+turbo
2. get the FX-8 or FX-6 into 4/4 or 3/3 mode to build X4 oder X3; asus is needed; thats it but you have to go into bios every time

:cool:

Oh, and by the way folks... but this might be a decent solution to allow us to manage our threads the way we like them and give us the extra flexibility of deciding if we want 4c/2cu, 4c/4cu or 8c/4cu on an app to app basis WITHOUT much hassle and without having to reboot! :D

http://img.tomshardware.com/us/2004/...taskassign.zip

just set your turbo OCs accordingly and setup each game in the way it will benefit you the most? :D win/win?

if someone could please test this and see if it plays nice with BD and let the rest of us now, would be grateful!

EDIT: hmmm seems people are already aware of utilities like this, my bad xD

http://www.hardware.fr/articles/842-...windows-8.html
1. They see 2M/4C in sum 15 % behind 4M/4C and fritz chess 8130/6417=127%

2. In the first post here it was reached 122 % in all tests and 8813/6335=139 % in fritz chess; fritz chess is 108 % of 1.

So 1. should be task manager method and 2. deactivating cores in bios.

Interesting detail:

you need 6CU/6C clock by clock to beat an FX-8150 through all benchmarks

awesome, chew!.. Only thing missing is a current deneb OC'd under same cooling etc (that i'd like to see)

Quote:

---

Originally Posted by son14

The answer could be simple.

---

This answer is not for that question, because he was speaking about 4CU/8C vs. 4CU/4C, which is irrelevant in case of games with 4 threads or less.
So, let's we go back to 4CU/4C vs. 2CU/4C.

Quote:

---

If you use 2CU/4C, there are 2 floating point units available, in 4CU/4C mode 4. So a task using 4 cores, gets access to 4 vs. 2 fpus. The performance gain in 50% int and 50 % fpu scenarios, is up to 25 %. Xbox ported games, mostly use 2 cores. So they get access to 2 fpus in both scenarios and you see performance gain of about 5 % only, from doubled cache per core.

---

You're searcing for the answer in the right direction, floating-point wise. Although, let's not forget this FPU is 2-way SMT, with double the resources of K10's FPU: it's 2xFMAC vs. 1xFADD+1xFMUL, FMAC being 1xFADD+1xFMUL combined. The thing is that all of this computing power can only be fully utilized with FMA code, because you can't have an FADD and FMUL independently started in the same cycle on an FMAC unit, only if it's an FMA instruction.

Thus, for legacy code 4CU/4C mode means double the resources utilizable all the time (not shared with the second thread), per hw thread, so it can have independent FADD's and FMUL's started in the same cycle on the two FMAC units. (Extra bonus is that it can have also 2xFADD or 2xFMUL, not just 1xFADD+1xFMUL, like on K10.) In 2CU/4C mode, it can have usually only 1xFADD OR 1xFMUL started in the same cycle (mostly one FMAC unit available because of the second thread engage the other one), per hw thread.

(And so, I think there won't be such a significant gain in performance for FMA heavy code, going 4CU/4C from 2CU/4C. EDIT: Or, even if there be so, both cases will be signicicantly faster than the same algorithm with legacy code.)

Quote:

---

...So a task using 4 cores...

---

(I think it's not the best wording, because less than quad-threaded apps can also be "using 4 cores" upon the constant core-variation of Windows. So, it's about how many threads an application have.)

Quote:

---

The problem is Dirt, F1, Civ5, Crysis 2,... . Northbridge overclocking and 4CU/4C can be the difference between playable or not.

---

Definitely worth a testing.

Quote:

---

Originally Posted by The Stilt

It is a design limitation as far as I know.
You can run the compute units either in single or dual core mode but not at the same time.

---

So, you're saying if we enable the 2nd core here, then we can't disable ony other cores, alone in a CU? :(

Quote:

---

Originally Posted by omninmo

Oh, and by the way folks... but this might be a decent solution to allow us to manage our threads the way we like them and give us the extra flexibility of deciding if we want 4c/2cu, 4c/4cu or 8c/4cu on an app to app basis WITHOUT much hassle and without having to reboot! :D

http://img.tomshardware.com/us/2004/...taskassign.zip

just set your turbo OCs accordingly and setup each game in the way it will benefit you the most? :D win/win?

if someone could please test this and see if it plays nice with BD and let the rest of us now, would be grateful!

---

Wow, I've forgot about this little utility. Perhaps it could be enhanced and adapted to BD, so all of it could work fully automatic! Would be even more useful! :yepp:

Quote:

---

Originally Posted by dess

So, you're saying if we enable the 2nd core here, then we can't disable ony other cores, alone in a CU? :(

---

If core 2 is enabled the CU0 would be running in dual core mode.
This means all of the remaining CU must be running in dual too.
You can disable more cores while core 2 is enabled but then you would need to turn a whole compute unit off.

I think the limitation is because each core needs to have a pair.
Atleast the schematics show all of the individual cores connected. I might be wrong thou.

Quote:

---

Originally Posted by dess

So, you're saying if we enable the 2nd core here, then we can't disable ony other cores, alone in a CU? :(

---

If core 2 is enabled the CU0 would be running in dual core mode.
This means all of the remaining CU must be running in dual too.
You can disable more cores while core 2 is enabled but then you would need to turn a whole compute unit off.

I think the limitation is because each core needs to have a pair.
Atleast the schematics show all of the individual cores connected. I might be wrong thou.

Quote:

---

Originally Posted by son14

Take a look here at the picture "Fritz Chess benchmark 4.3 (watts atx 12v)":
http://www.hardware.fr/articles/842-...windows-8.html

2/4 = 60 watts (100%)
4/4 = 84 watts (140 %)
2/4 turbo = 95 watts (158 %)
8/8 = 101 watts (168 %)

totally power consumption in your scenario (oc and with HD6990) is about 500 watts in peak:
http://www.rage3d.com/reviews/cpu/am.../index.php?p=7

To reach 4/4 @stock in dirt with 2/4, you nearly have to double the power draw. In other games (console ports) its useless, because fps are mostly >60 fps and there is no gain in performance, gpu is limiting.

There are two different methods to get what you want:

1. let the FX-8 or FX-6 work in stock mode and bind the game/the app manually through the task manager to 1357 or 0246 or use a tool like "amd overdrive" or "core affinity resident" to build a profile; you do not need the expensive asus board and simulate the windows 8 task manager in a smarter way than inefficient cu parking+turbo
2. get the FX-8 or FX-6 into 4/4 or 3/3 mode to build X4 oder X3; asus is needed; thats it but you have to go into bios every time

:cool:

---

the GPU they used alone used 375 watts with out OC ? :confused: 545 - 375 = 125 watts :wth:

Quote:

---

Originally Posted by demonkevy666

the GPU they used alone used 375 watts with out OC ? :confused: 545 - 375 = 125 watts :wth:

---

CPU PWM efficiency is around 80% + 80-90% for the PSU, so no.

interesting thread here

Please, please, please tell me all (or at least some) of this testing is with 64-bit applications on a 64-bit OS.

The application I am working on, the 64-bit version is 40% faster than the 32-bit version. There are twice as many general purpose registers in 64-bit mode. My application makes heavy use of 64-bit integers and no floating point. It is like Fritz except it is a breadth first search instead of a depth first search. It will use all the memory that you can throw at it and thrash it hard.

It wouldn't be that hard to add a dialog box to let you set the threads to run on the desired cores.

My next machine was going to be a dual socket C32 machine, I need the memory slots more than I need processing cores. But after reading Anands "Rendering and HPC Benchmark Session Using Our Best Servers", I have concerns about the memory performance of BD in dual socket boards.

@chew*,thanks for all your testing on this matter,you sir are awesome for going out of your way to help the community:up:

Quote:

---

Originally Posted by dess

What if you enable only two cores in two CU's...?

---

It would probably do well as a 6 core, the third core pair on this chip appears to be the really sore thumb on this chip.

Thats what always drops out in prime 95.

I gamed till like 5am this morning at 4.6 4 core, tbh I can not tell the diff between this rig or my sandy rig.

Both run very smooth with no glitches.

I did have some hitches with a kingston SSD though, not agreeing with the ch5 to well, might be firmware.

It's fine with a normal sata drive.

Quote:

---

Originally Posted by spurious

Please, please, please tell me all (or at least some) of this testing is with 64-bit applications on a 64-bit OS.

---

All my tests were done with 16g of ram installed, that should answer that question for you ;)

I'd like to apologize also, my cpu-z is outdated, this is a retail chip older versions call it ES, just installed new version

Once again thanx for all the usual posts and info. So refreshing not having to hear about BD bashing.
Kudos to Chew and the other xtreme addicts!

Quote:

---

Originally Posted by chew*

It would probably do well as a 6 core, the third core pair on this chip appears to be the really sore thumb on this chip.

Thats what always drops out in prime 95.

I gamed till like 5am this morning at 4.6 4 core, tbh I can not tell the diff between this rig or my sandy rig.

Both run very smooth with no glitches.

I did have some hitches with a kingston SSD though, not agreeing with the ch5 to well, might be firmware.

It's fine with a normal sata drive.

---

Thanks Chew for all your testing and feedback. It nice to see you don't feel any real world difference between BD and SB. Looking forward playing with my BD once it arrives. Got my UD7 today