AMD's Bobcat and Bulldozer

[Motiv]

who really needs to oc a dual socket board ??? seriously... besides for e-peen value

averagesystems.org is that way >>>

You may of noticed this forum has people who's idea of fun, is to OC a cpu until sparks are flying off the electricity socket; hollywood film style.

Berating people for doing something you yourself term as "e-peen value", only undermines your own argument, to a point of nonsense.

Can I ask why you are on a forum for overclockers and "extreme" systems, if you feel it is no more than a "e-peen" thing?

[AliG]

< This guy. When I was running a dual Istanbul setup between last July and earlier this year, I really wanted to overclock it. I did, to a limited degree, on my first (nVidia) motherboard.

Why? Because I do some of everything. When I'm compiling FPGA programming files during development, that's unfortunately a single-threaded process that takes a long time. At 2.2 GHz, you're looking at a minute and a half to two minutes every time you want to test a configuration. It's not like Visual Studio where you hit F5 to check your changes and the thing instantly starts up. When I'm relaxing by playing games, that's mildly threaded. Most games don't do their best with more cores and lower clock frequencies, though a few like Bioshock do fine. For the most part though, my gaming experience really tanked when I went from a Phenom II X4 940 @ 3.6 GHz to the dual Istanbul @ 2.2-2.5 GHz. And finally I do some video work which tends to take advantage of more cores quite nicely...some times.

It's never fast enough though. A Thuban overclocked to 4GHz, which isn't terribly difficult, can turn in throughput very close to what I used to get out of my dual Istanbul station. That was the point I decided to switch. If I had been able to overclock that platform, the question of changing or not wouldn't have been so easy to answer.

Even if you had been able to overclock it to 3ghz, even a 4ghz Deneb would likely be more useful to you as I can't see anything on that list that scales above 4 cores.

[AliG]

averagesystems.org is that way >>>

You may of noticed this forum has people who's idea of fun, is to OC a cpu until sparks are flying off the electricity socket; hollywood film style.

Berating people for doing something you yourself term as "e-peen value", only undermines your own argument, to a point of nonsense.

Can I ask why you are on a forum for overclockers and "extreme" systems, if you feel it is no more than a "e-peen" thing?

Because you don't have to spend a fortune to be extreme

There's more to extremesystems about having the most expensive setups. In fact the most impressive ones (outside of the guys who have access to the best products+cooling) are generally the builds that were low in cost but heavily modded to have top level performance. I think that's the true nature of XS; just like extenze it's not about having the biggest one in the world, but rather maximizing what you got.

Ohh and I totally deserve bonus points for linking XS to extenze in a positive way

[Sn0wm@n]

averagesystems.org is that way >>>

You may of noticed this forum has people who's idea of fun, is to OC a cpu until sparks are flying off the electricity socket; hollywood film style.

Berating people for doing something you yourself term as "e-peen value", only undermines your own argument, to a point of nonsense.

Can I ask why you are on a forum for overclockers and "extreme" systems, if you feel it is no more than a "e-peen" thing?

well then... why dont you go buy a 4 socket mobo and use 4 opteron ... xs has a group buy for those ... more power for your use that no oc'd dually could give you ... and i heard they are cheap too .... so one more argument as to why you would oc a server mobo besides e-peen ...

Because you don't have to spend a fortune to be extreme

There's more to extremesystems about having the most expensive setups. In fact the most impressive ones (outside of the guys who have access to the best products+cooling) are generally the builds that were low in cost but heavily modded to have top level performance. I think that's the true nature of XS; just like extenze it's not about having the biggest one in the world, but rather maximizing what you got.

Ohh and I totally deserve bonus points for linking XS to extenze in a positive way

exactly ... overclocking's true nature was to take cheapo cpu's and make em give a run for their money to big top bin cpu's .....

[m0n4g3]

i'm pretty sure i overclock to squeeze the most out of any rig i get.... wether it be top bin, medium bin, or cheapo bin......

*shrugs* i might just be different though!

[MrMojoZ]

Considering the points an overclocked 6 core Westmere puts out in WCG, I would certainly want clockable dual BD boards. I don't think the market exists to support selling them, but I sure as hell would like to own one.

[scorpidragon]

Considering the points an overclocked 6 core Westmere puts out in WCG, I would certainly want clockable dual BD boards. I don't think the market exists to support selling them, but I sure as hell would like to own one.

I'd like to see AMDs motherboard manufacturers release a Dual BD board with overclocking features too!

It looks like AMD is going to introduce HT4 into the chipsets and DDR4 is gonna hit the market sometime next year.

Also AMD needs to get rid of the north and south bridges and replace both bridges with a new central bridge that has a PCIe controller built in to it. Doing this would increase the chipsets performance as you no longer have a south bridge to slow things down.

CPU socket --> HT4 link --> Central Bridge.

[demonkevy666]

I'd like to see AMDs motherboard manufacturers release a Dual BD board with overclocking features too!

It looks like AMD is going to introduce HT4 into the chipsets and DDR4 is gonna hit the market sometime next year.

Also AMD needs to get rid of the north and south bridges and replace both bridges with a new central bridge that has a PCIe controller built in to it. Doing this would increase the chipsets performance as you no longer have a south bridge to slow things down.

CPU socket --> HT4 link --> Central Bridge.

nice idea

but right now the NB is linked to the SB by a A Link III which a pci-express link.

the NB is link to cpu by HTT link. The NB link the pci-express links to the HTT also.

[scorpidragon]

nice idea

but right now the NB is linked to the SB by a A Link III which a pci-express link.

the NB is link to cpu by HTT link. The NB link the pci-express links to the HTT also.

By removing the PCIe link from NB --> SB link, you will be able to add more or faster PCIe slots to the motherboard.

[tifosi]

The big problem with AMDs dual server boards are that they are not overclockable!

If the MC dual boards ever came out as an overclockable workstation board, then I would by two or three of them!

If MC or BD CPUs came out unlocked (like Intel's) runs at around 3.0 GHz, then I would buy four to six CPUs!

If AMD released BD as an Phenom FX CPU for G34 or whatever socket that comes out, then I would buy four to six of those CPUs!

Did you really miss this thread?

http://www.xtremesystems.org/forums/...d.php?t=233565

This bloke blew the pants off of the opty :P

[scorpidragon]

Did you really miss this thread?

http://www.xtremesystems.org/forums/...d.php?t=233565

This bloke blew the pants off of the opty :P

Not really interested in that thread.

The reason as to why I am not interested in G34 server boards are that they cannot be overclocked!

[nn_step]

Since I feel like injecting more data.

P4:

note the integer dispatch limitation. [Answer 2]



Yohan:

note the integer dispatch limitation. [Answer 2]




Conroe: [Answer 3]

note the integer dispatch limitation.




Nehalem:

note the integer dispatch limitation. [Answer 3]



note that Barcelona has the EXACT SAME integer dispatch limitation. [Answer 3]



While I was gone, I wrote a couple basic micro-benchmarks:

1) A Integer compute limited micro-benchmark [Makes OoO and prediction useless; since it is impossible to mispredict and no hazards exist]
Which quickly found several interesting things:
a) performance was determined entirely by the following formula: time = 100,000 / (clock speed * number of integer instructions per clock executed)
b) When work shifted to more complex integer instruction mix [less adds and more XNORs] AMD began to have a drastic performance advantage
c) When work shifted to more complex floating point instruction mix [less adds and more shuffles] Intel began to have a drastic performance advantage

2) In a compute low, data rich program [b-trees and triple indirection vs arrays and zippers]
Several interesting things were found:
a) Performance was determined ENTIRELY by average latency in nano-seconds for data fetch [CPU clock speed, integer unit count, and etc didn't matter]
b) Caches showed improved performance until reaching 1 MB [Then showed rapidly dropping diminishing returns]
c) prediction seemed to be extremely closely tied to less than a half dozen assumptions [BOTH for AMD and Intel] Ironically both AMD and Intel, share only 3 of them. Any code outside of those assumptions had an average latency of main memory. [Extreme cache thrashing]

3) Finally in the real world; I took the following common applications and modified them to properly monitor them.
They generated the following average IPC under Ideal conditions:
a) Firefox [0.3 - 0.5 basic browsing; 2.7 when rendering video]
b) Open Office [0.01 - 0.7]
c) Apache [0.9 - 1.2]
d) XFREE86 [0.02 - 0.04]
e) GNOME [0.0001 - 0.0002]
f) KDE [0.0001 - 0.0002]


Now, I may be incorrect but what I have drawn from the preceding experiential evidence, is that with the exception of video rendering; The user will see absolutely no reduction in performance by AMD reducing the number of Integer Units.

[richierich]

Now, I may be incorrect but what I have drawn from the preceding experiential evidence, is that with the exception of video rendering; The user will see absolutely no reduction in performance by AMD reducing the number of Integer Units.

Very nice.

[Particle]

Even if you had been able to overclock it to 3ghz, even a 4ghz Deneb would likely be more useful to you as I can't see anything on that list that scales above 4 cores.

That's not so. The video work I did scaled up to 90-100% with 12 cores. At 2.2GHz it was still a bit faster encoder than my 4 GHz Thuban. Deneb would have no chance, assuming it could even get that high. They don't usually make it unlike Thuban. My contribution to WCG would have also benefited, though I didn't remember to mention that as something I do.

3GHz would have been "good enough" for my poorly threaded needs like gaming for a longer time, which is why that is significant in my story.

[scorpidragon]

What I would like to see from AMD is for them to fit a board with a HTX type slot so that a socket (with 2x So-dimm) card could be plugged into the board! Like adding a 18-watt Bobcat CPU to that card would make an ideal co-processor for high-end/Budget server computer systems!

[behrouz]

now i asked Mr. John :

Hi John
i am very confusing.is it Correct that BD is K10 ? if yes then isn’t the K10.5 Shanghai?

John :
Bulldozer is a brand new architecture, it is not based on previous cores.


Which Mean : He Says K10 is NOT Bulldozer.
But here :

http://www.semiaccurate.com/forums/s...58&postcount=6

Such a bad misunderstanding of Fudo...

With "K8" AMD people refer to anything from Opteron to Magny Cours. K10 is Bulldozer according to several CVs of AMD engineers. Mitch Alsup just recently said the same in comp.arch.

http://www.semiaccurate.com/forums/s...2&postcount=12

@Skywalker:
Look into LinkedIn profiles. K10 = Bulldozer. And Mitch Alsup wrote: "(Ahem) K10 is BullDozer, K8 is Opteron and follow-ons. "

Remember Hans de Vries' Barcelona core image with markup? He called it K8L. K10 is just so widely used since Barcelona that it's nearly impossible to revert that.

And anyone else:
Please stop the discussion about the K8 being not enough vs. K10. Llano will have the most advanced core based on the same architecture used in Barcelona, Shanghai, Istanbul, MC, Lisbon, Thuban..

So It's should be K11

[Movieman]

who really needs to oc a dual socket board ??? seriously... besides for e-peen value

Me..Four systems running here and none are single socket.
Why? Here's some reasons:
1) Boards are better quality
2) Electrical usage for a given job done is better.

3) Space: The duals outperform the singles by a wide margin and take essentially the same space
4) Buy top boards,they don't break and I mean DON'T.
Whens the last time you saw a single socket board run 3+ years at 100% load with out a boot other than from a electrical outage or twice a year cleaning?
I see that all the time.
5) Time effectiveness: Toss any muti threaded task on a dualie and it's done much faster than any single socket machine can do.

Ok,thats settled, back on topic now.

Oh, I forgot one thing:
6) E-PEEN!

[FlanK3r]

behrouz: Because Phenom II is not really K10.5, but this name is from people and AMD accpet it. But oficilan is Phenoms II Family 10h (K9??), Bulldozer will family 15 think

[scorpidragon]

Well I'm going to bypass BD and AM3+ and I will wait and see what the socket for Llano is going to be and if it is going to be around 1155 to 1275 pins, then that is the board that I will buy! But I however have a feeling that the socket will only be 1035?

The reason why is that Llano has an onboard PCIe controller and BD does not!

It is also about time that AMD integrated the North and South bridges into one central bridge too!

[Sn0wm@n]

Not really interested in that thread.

The reason as to why I am not interested in G34 server boards are that they cannot be overclocked!

opterons have dual link ht so they can talk between themselves .. so you will need to look into server motherboard ... nothing else .. sorry but what you are dreaming about has to come from the server world ...

Me..Four systems running here and none are single socket.
Why? Here's some reasons:
1) Boards are better quality
2) Electrical usage for a given job done is better.

3) Space: The duals outperform the singles by a wide margin and take essentially the same space
4) Buy top boards,they don't break and I mean DON'T.
Whens the last time you saw a single socket board run 3+ years at 100% load with out a boot other than from a electrical outage or twice a year cleaning?
I see that all the time.
5) Time effectiveness: Toss any muti threaded task on a dualie and it's done much faster than any single socket machine can do.

Ok,thats settled, back on topic now.

Oh, I forgot one thing:
6) E-PEEN!

how much did all your dually system cost you ????

and yes i know people like to oc stuff up .. but only people on xs ... most people use their stock clocks system without any problem ... .. BUT THOSE PEOPLE WORK and dont game ....

and to tell you the truth my single socket am2 opteron never did have to shutdown for a whole 13 months .. and i did a reboot only because a hdd crashed ....


e-peen is all that matters to people oc'ing dual socket boards ....

[scorpidragon]

opterons have dual link ht so they can talk between themselves .. so you will need to look into server motherboard ... nothing else .. sorry but what you are dreaming about has to come from the server world ...

Yeah I'd say so. But I have heard that AMD maybe bring out a new 1P/2P board that will target high-end users?

Looks like I will need 12 threaded CPU then.

So that means I'll need the following then:
Phenom FX 12 core 3.2GHz CPU (with MCM Thuban) for socket G34.
16GB to 32GB DDR3 RAM.
SATA6 x4/x8 PCIe RAID card.
USB 3.0 x4 PCIe card.
Radeon HD5850/6850 PCIe card.

[Chrono Detector]

Like I said before, AMD really needs to make a socket that has more than 940 pins, and support more memory instead of dual memory channel. They've been stuck on 939/940 pins for years now.

[scorpidragon]

Like I said before, AMD really needs to make a socket that has more than 940 pins, and support more memory instead of dual memory channel. They've been stuck on 939/940 pins for years now.

For 8 core CPU 754 pins.
Dual PCIe controller 188 pins.
Quad channel memory 374 pins
Total: 1306 pins

For 16 core CPU 1508 pins.
Dual PCIe controller 188 pins.
Quad channel memory 374 pins
Total: 2070 pins

So it looks like AMD has been stuck on 754 pins for a long time too.

[JF-AMD]

AMD will not have a 2P client board.

http://bit.ly/chNf9B

There is no market size that will justify it.

[Particle]

e-peen is all that matters to people oc'ing dual socket boards ....

If I've said it once, I've said it probably four or five times: This is simply not true. Believe it or not, some of us aren't 100% work or 100% play.