AMD Zambezi news, info, fans !

The official term for a "module" is a compute unit.
So please use it to avoid confusion.

:)

Quote:

---

Originally Posted by bamtan2

good info, thank you. I think that sounds legit. it competes well but falls behind in games, where intel is usually the leader anyway.

now the final question becomes: can the normal overclock of a bulldozer make it faster than a normal overclock of a sandy bridge. probably 99% of 2500k can run at 4.2ghz. if 99% of bulldozer can run 5ghz or more on air, there might be a chance for bulldozer yet...

---

99% of 2500K can run 4.7 Ghz on air. Normal OC on water is 4.8 to 5.2 Ghz. (Those running 5.2 really pushing unsafe volts IMO)

If BD can run 5.2-5.3 I'll give it a fight in price to performance for overclockers.
I predict no more than 5.0-5.1 on air, 5.2-5.3 for those of us on water.

The whole goal here really is for AMD to beat what Gulfy can do in single thread, and based on game performance on AMD slides alone, it might be possible. After all, that was Intel's first 32nm chip, right? ;)

IPC doesn't seem to have increased, it actually seems to have decreased slightly...
But a 5% decrease in IPC with a 20% increase in clock speed is well worth it in my book.

Quote:

---

Originally Posted by BeepBeep2

IPC doesn't seem to have increased, it actually seems to have decreased slightly...
But a 5% decrease in IPC with a 20% increase in clock speed is well worth it in my book.

---

5%? No its something in the 80-90% K10.5 IPC range. It means 10-20% decreased.

due to the shared resources IPC can be measured in 2 ways, and its important to know them both. in theory SB has decresed IPC if we try to compare a core + SMT vs 2 cores, and i think people are doing it here too. if gaming is using just a few threads, then IPC might look really good, but if crunching is using all 8, it might look really bad. no point arguing over IPC right now until we get all benchmarks released.

Quote:

---

Originally Posted by Manicdan

the problem with a 2 module only chips is that you cant unlock it, for some of us that would have been the main selling point.
for AMD though it would be a great price/mm2 rather than turning perfectly fine chips into half working chips simply due to demand. and to be honest up to 4 cores (or threads, however people want to perceive it) is still all thats really desired from most people, and anything more just isnt worth the extra price.

i do hope that because they are built starting with just 2 modules, it can have some with a good yield that OCs like crazy and thus offering 4.5ghz or more in stock forms.

---

im still not convinced they will have a native quad part,even the leaked slides have all three FX4100 models showing 4Mb L2 and 8Mb L3 which means they are native 8 cores with 2 disabled Compute Units(just for The Stilt).but who knows?

Quote:

---

Originally Posted by BeepBeep2

But a 5% decrease in IPC with a 20% increase in clock speed is well worth it in my book.

---

You get something wrong sir, 8150 have no more than 3.6Ghz while all core loaded, the video said that. 1100T has 3.3Ghz.

Quote:

---

Originally Posted by undone

You get something wrong sir, 8150 have no more than 3.6Ghz while all core loaded, the video said that. 1100T has 3.3Ghz.

---

4.2 Ghz max overclock on Thuban vs 5.0+ max overclock on FX-8150

I said absolutely NOTHING about stock speeds.

Quote:

---

Originally Posted by rog

5%? No its something in the 80-90% K10.5 IPC range. It means 10-20% decreased.

---

Are you another one of those people that looked at leaked Cinebench results and divided by 8?

Quote:

---

Originally Posted by Manicdan

due to the shared resources IPC can be measured in 2 ways, and its important to know them both. in theory SB has decresed IPC if we try to compare a core + SMT vs 2 cores, and i think people are doing it here too. if gaming is using just a few threads, then IPC might look really good, but if crunching is using all 8, it might look really bad. no point arguing over IPC right now until we get all benchmarks released.

---

chew* stated something about SuperPi, a single threaded program gaining a significant amount when set to more than one thread...we will see where this progresses upon release/NDA release in my opinion.

The question is, does it work? You're sitting on a gold mine here, Fruehe. If it works, I'll order a dozen!

(sorry, just had to)

I'm understanding the increased use of the term "throughput" now. Now to determine what types of code will gain "throughput"....

Quote:

---

Originally Posted by xVeinx

The question is, does it work? You're sitting on a gold mine here, Fruehe. If it works, I'll order a dozen!

(sorry, just had to)

I'm understanding the increased use of the term "throughput" now. Now to determine what types of code will gain "throughput"....

---

Database, virtualisation, some HPC programs.

Quote:

---

Originally Posted by xVeinx

I'm understanding the increased use of the term "throughput" now. Now to determine what types of code will gain "throughput"....

---

Code that uses 8 threads and up ;-)

Quote:

---

Originally Posted by BeepBeep2

99% of 2500K can run 4.7 Ghz on air.

---

I mean for regular use. safe settings, and a large sample size of chips tested. my research done on other forums indicates 1.45v and PLL is sketchy, and 1.35v no PLL is safe max. under these settings, I don't believe 99% of 2500K do 4.7ghz. I am willing to be corrected by a large experiment.

I hope hardware reviewers get this comparison sorted out in time for their bulldozer reviews. I don't want to read one review saying bulldozer is bad because it isn't fast enough to beat the 5.0ghz that 99% of 2500k reach, while another review says bulldozer rules because it beats the 4.0ghz that 99% of 2500k reach.

How much are the bulldozer 8c 6200 series likely to cost? If they aren't too much more than the consumer version 2p/4p would be pretty interesting. Since they go up to 3.7ghz in turbo mode you would be able to get a better balance of single and multi threaded performance.

I also think that, starting out all BD's will be made with 4 modules. AMD did this with the PH & PH II so that their yields where higher. If the 4 module FX 8xxx has a busted part, they can resell it as a FX 6xxx or a FX 4xxx. Remember when the first PH II x4's came out, we had PH II x3 with a disabled core, that we could unlock. Now that they have a mature process at 45 nm, they are now making true x2 and x3 parts that do not have disabled cores to unlock.



MaddMutt

MaddMutt: You have a point (just I don't think they have a native PhII X3 :) ), and given bad yields, I don't think we will see native less-than-8-cores from this Bulldozer iteration. Remember that it doesn't have to had broken parts to sell it as less-than-8-cores, usually they sell it as that when it doesn't fit in a given TDP class with more cores enabled.

Quote:

---

Originally Posted by The Stilt

The official term for a "module" is a compute unit.
So please use it to avoid confusion.

---

Hmm, shouldn't the module also be official, as AMD itself called it that way in many slides and other places?
Why all need to learn another new term?

Quote:

---

Originally Posted by BeepBeep2

chew* stated something about SuperPi, a single threaded program gaining a significant amount when set to more than one thread...

---

How do you set it to more than one thread? Do you mean running more than one instance?

Quote:

---

Originally Posted by dess

MaddMutt: You have a point (just I don't think they have a native PhII X3 :) ), and given bad yields, I don't think we will see native less-than-8-cores from this Bulldozer iteration. Remember that it doesn't have to had broken parts to sell it as less-than-8-cores, usually they sell it as that when it doesn't fit in a given TDP class with more cores enabled.


Hmm, shouldn't the module also be official, as AMD itself called it that way in many slides and other places?
Why all need to learn another new term?


How do you set it to more than one thread? Do you mean running more than one instance?

---

Affinity in task manager I assume. It should be set to all cores by default so no worries.

Quote:

---

Originally Posted by El Gappo

Affinity in task manager I assume. It should be set to all cores by default so no worries.

---

Why would it be better if it jumps between the cores all the time? Shouldn't it be better if it's stuck to one core, while the rest can go rest, so that turbo can kick in?
Or, is the turbo mechanism fast enough to make it irrelevant? Then, why it matters at all?

I think, if FX at 4.2 GHz in single is whoirse than Thuban at 3700MHz, here is some bug in chip (maybe too smal L1D or noit good optimized comunication between caches? Latency are bigger than K10, but not too much for this diferences...)

Quote:

---

Originally Posted by radaja

cool that is good news they should be lower TDP right,i thought there was only the x4 with two modules disabled like the FX4100 with 4Mb L2 and 8Mb L3?

---

Correct, lately all roadmaps have shown the quad-cores with 8 MB L3 cache which indicates it's the same die so we will see what happens now. I know there was supposed to be a separate die for the Quads to begin with... Hex-cores would still be derived from Octa-core dies.

Quote:

---

Originally Posted by Manicdan

the problem with a 2 module only chips is that you cant unlock it, for some of us that would have been the main selling point.
for AMD though it would be a great price/mm2 rather than turning perfectly fine chips into half working chips simply due to demand. and to be honest up to 4 cores (or threads, however people want to perceive it) is still all thats really desired from most people, and anything more just isnt worth the extra price.

i do hope that because they are built starting with just 2 modules, it can have some with a good yield that OCs like crazy and thus offering 4.5ghz or more in stock forms.

---

This is a much lesser problem, for AMD. For all we know, Bulldozer won't be able to unlock :)

Quote:

---

Originally Posted by FlanK3r

I think, if FX at 4.2 GHz in single is whoirse than Thuban at 3700MHz, here is some bug in chip (maybe too smal L1D or noit good optimized comunication between caches? Latency are bigger than K10, but not too much for this diferences...)

---

It that performance wuold be true it would be slower than the x6 overall....

Given the max clockspeeds drops more compared to the X6 and it has a module overhead in multithreaded applications.... So i doubt your 'estimate' is correct.

overall not, overall in all tets is FX better than 1100t. I told only about pure single threads

Quote:

---

Originally Posted by FlanK3r

overall not, overall in all tets is FX better than 1100t. I told only about pure single threads

---

you mean fx 8150p?
It would be a fail if the 8150p < 1100t.

anyone remember this (december 2010) : zambezi performs 50% better than 1100t.

and we can read above "render performance is based on Cinebench R11.5"

Attachment 120656

and now?
watching the slides of donaninhaber we see in cinebench that

Attachment 120657

somoene lied..

Quote:

---

Originally Posted by liberato87

you mean fx 8150p?
It would be a fail if the 8150p < 1100t.

anyone remember this (december 2010) : zambezi performs 50% better than 1100t.
and we can read above "render performance is based on Cinebench R11.5"

Attachment 120656

and now?
watching the slides of donaninhaber we see in cinebench that

Attachment 120657

somoene lied..

---

It doesn't really look that far off TBQH. If we check the difference between the 1100T and the 8150, it shows the 8150 as performing 33% faster. With the 1100T lagging slightly behind the reference i7 950, it's hard to draw a solid conclusion from that first pic anyways. We also don't know how accurate that second slide is either. The subtext indicates that the numbers are AMD's own official results (and intel's own on the 3960x), but from where and with what system configuration?

--Matt

Interlagos November. So Zambezi at the same time?

Quote:

---

Meanwhile, AMD's server processor codenamed Interlagos will also have difficulty shipping on schedule and is expected to be delayed to November.

---

http://www.digitimes.com/news/a20110930PD207.html

Quote:

---

Originally Posted by undone

Interlagos November. So Zambezi at the same time?



http://www.digitimes.com/news/a20110930PD207.html

---

If zambezi doesn't come out until november I will genuinely cry :p

AMD lied saying that BD is an 8 core. In fact even now is lying. It is not.
Second, if BD is further delayed for an actual launch, on 12 octomber we should have at least a "paper launch".