Bulldozers first screens

[Postmodum]

Man... what a math class :P thumbs up fur the clarification and enrichment of math culture in the forum ( i mean it! )

[AKM]

http://www.chiphell.com/thread-190177-1-1.html

Crap >30%(ignoring useless Super pi) slower at same clocks with double the cores. I hope these aren't true otherwise I see no reason why anyone in his right mind would buy one.

[Manicdan]

how come those benchmarks look very similar to results from a PII 920?

from anand bench:
PII 920 gets 3244 pts single
11440 pts multithreaded

an 8 core chip with perf less than something from 2 years ago. lol

[informal]

Ridiculous scores.
I already posted about it on another forum.

If you look at the chiphell topic ,the OP is confused.He lists the ES which is clearly 6 core by the code,but then posts a screenshot of an 8 core ES from different week and with different revision marker.

Then comes the C10 result of 2340pts for single core,which is on the level of ~2.58Ghz Phenom II.You can see 32bit C10 results here:
http://www.pcstats.com/articleview.c...d=2511&page=10. Scaling of 4.85x is close to 6 core scaling number but CPUz now shows alleged 8 core Zambezi ES.

Similar goes for Super pi in which alleged Zambezi has 12% less performance than 2.8Ghz Phenom II:
http://www.pcstats.com/articleview.c...id=2511&page=8

Lastly ,we have a lowly CPU score from 3dmark06,4900pts,which is a tad lower than what 6 core 1055T gets as can be seen here.Also note that 945 QuadCore Phenom II at 3 Ghz has 14% lower score than this supposed 8 core 2.8Ghz Zambezi. Just ridiculous.


So we have a 10%-30% decrease,according to this "info",when going to Bulldozer.Mind you ,this is almost all fp intensive workload we see and somehow Bulldozer is sucking in it . We've already seen AMD slide which lists 8 core XX Ghz Zambezi as having around 1.88x the score of 1100T Thuban in C11.5. So we can easily dismiss this whole "test" as either fake one or done on some barely running hardware.

[jabway]

Those results don't look suspicious at all.

[Shadov]

Ridiculous scores.
So we have a 10%-30% decrease,according to this "info",when going to Bulldozer.Mind you ,this is almost all fp intensive workload we see and somehow Bulldozer is sucking in it . We've already seen AMD slide which lists 8 core XX Ghz Zambezi as having around 1.88x the score of 1100T Thuban in C11.5. So we can easily dismiss this whole "test" as either fake one or done on some barely running hardware.

Informal don't be surprised terrace posted a lot and looking at your signature he kinda succeeded.

Thanks God he was stopped at -30% performance.

[MrMojoZ]

Ooops.

You are proving my point right? Please tell me you meant to agree with me, otherwise you are saying that fanboy extremism is good and don't understand the benchmarks you posted.

[madcho]

should be fake

[informal]

From SA forum,new image:


3dmark06 ~3034pts,lower than Phenom I 9850
C11.5 ~2.86pts, 2.5x lower than 1100T Thuban
Super Pi.... ah who cares anymore lol
The chip this Chinese guy has is evidently a seriously borked ES 6 core Zambezi @ 2.6ghz.

PS now the scores are going further down with each new picture,we are in K7 range now^^.This is clearly terrace testing that ES

[FlanK3r]

haha:-D Right, this is not possible...2.86 R11.5 with hexacore-rofl...Score as some 2.8 GHz Athlon X3

[Olivon]

You are proving my point right? Please tell me you meant to agree with me, otherwise you are saying that fanboy extremism is good and don't understand the benchmarks you posted.

When you look at overclocked results, Thuban is far.

Don't forget power consumption and efficiency too.

[Aten-Ra]

Many people came to a conclusion,primarily from C11.5 numbers,that Zambezi will have very strong FP units.A total score of almost 11 pts tells us that per one single FMAC(in shared MT mode) we have roughly 30% more computational power than one whole Thuban core,roughly at the same clock.When Zambezi doesn't see load across all cores,FMACs can be "paired up" (per module) so that in single or poorly threaded fp workloads we have a lot higher fp resources,on top of those 30%. Now on top of all this comes new Turbo boost in poorly threaded workloads.All things summed up,at least in legacy (or FMA tuned) fp code,Zambezi will be a big step up from today's cores,yes even versus SB.

The two (2) 128-bit FMACs per Module will only be paired to one 256-bit FMAC when we will have AVX.

If we only have one FP instruction per Module, then one 128-bit FMAC can have all the resources (FP Sheduller etc).

[bamtan2]

80% of 100 is 80,or 25%(1.25x) less.

Aren't percentages sometimes so wonderfully confusing

100 is 25% more than 80.
80 is 20% less than 100.

Much of the confusion here comes in from what the english language leaves implicit. "x%" implies a lot more than it says. In the above statement, "25%" really means "25% of the next number". So "25% more than 80" really means "25% of 80 more than 80". So blame unclear english for the confusion.

More details:

In the english language, the "is" means equals, and the "x more than y" means an operation, or equation. ("5 more than 10 is 15" means "5+10=15")

In the case of "x%", the "%" is like a unit, meaning "divided by 100 (and then multiplied against another number)". (a percentage is also known as "a fraction of 100".) So you divide that "x%" number by 100 to get what the fraction really means in decimal notation. (Decimal notation, not fractions, is the number format in which the rest of this math is done.)

But the "x%" has a secret. It is implicitly tied to the second number in this equation. "25% more than 80" doesn't just mean "0.25 more than 80", it means "25% of 80 more than 80". So once you have the decimal value of the percentage ("decimal value fraction" if you will), you apply it to the second number (using multiplication) to find that percentage of that number. So 25% of 80 is 20, because 0.25*80=20.

Then, the "x% (of y) more than y" means you add ("more") both the original "y" to the value of "x% (of y)" to get the answer to that equation.

"x less than y" also means "subtract x from y", which also means "y minus x", or "y-x", or 80-(25% of 80).

[informal]

The two (2) 128-bit FMACs per Module will only be paired to one 256-bit FMAC when we will have AVX.

If we only have one FP instruction per Module, then one 128-bit FMAC can have all the resources (FP Sheduller etc).

You are wrong.
Look here:
http://blogs.amd.com/work/2010/10/25/the-new-flex-fp/
If we have single threaded fp legacy code,one integer core can schedule instructions on both FMACs.

Core 1
2x128b AVX or 2x128b un-recompiled SSE

This means one core can have the whole shared FP resources on its disposal if Core 2 has no fp instructions scheduled in FP scheduler.

[pipoo]

Oh Jeez... something must be wrong with those scores. =/

[rintamarotta]

Seriously gimbed down Bulldozer ES, AMD doesnt want people know real numbers yet, Intel would panic.
NDA but, you can expect more from bulldozer

[XRL8]

When you look at overclocked results, Thuban is far.

Don't forget power consumption and efficiency too.

Well, duh, if you take cpu on a mature 32nm process its going to clock better than a 45nm one.
Your statement about "rape" is still invalid tho.TDP`s are kept inline only at stock settings.Fact of the matter is phenom X6 aint bad at highly threaded work.
Getting back to core of discussion, BD is going to be build on 32nm with power gating just like sandy.So NOT expecting it to clock better than X6 is not wise.
As for the results, they can pretty much be dismissed.
And no, not because im an AMD fanboy.but if it was that horrible, there would be no point in replacing phenom II with this.

[demonkevy666]

From SA forum,new image:

3dmark06 ~3034pts,lower than Phenom I 9850
C11.5 ~2.86pts, 2.5x lower than 1100T Thuban
Super Pi.... ah who cares anymore lol
The chip this Chinese guy has is evidently a seriously borked ES 6 core Zambezi @ 2.6ghz.

PS now the scores are going further down with each new picture,we are in K7 range now^^.This is clearly terrace testing that ES

well the only way you can get that low of a score is if you ran all of those at say start up all at the same time.
lol

it looks like he tried all of them at once somehow. lol

[Macadamia]

[Solus Corvus]

you make it sound as if most apps were well threaded and its rare to have apps that dont scale with more cores :P

That isn't my intention and I either wasn't clear or you misread me. Most apps aren't well threaded. But those that do scale with cores often benefit more from extra cores than extra clocks/IPC.

My point is simply that what matters to the average user doesn't matter so much to us. The high-throughput multicore case is just as extreme as the high clocks/IPC case. There are plenty of valid desktop uses that utilize many cores - encoding, rendering, heavy multitasking, etc. To determine if BD or SB is right for you will likely come down to using your brain to determine how benchmark scores actually apply to the apps you use and how you use them. There will probably be some cases where SB is the right choice and other cases where BD is the right one.

[FlanK3r]

Well, duh, if you take cpu on a mature 32nm process its going to clock better than a 45nm one.
Your statement about "rape" is still invalid tho.TDP`s are kept inline only at stock settings.Fact of the matter is phenom X6 aint bad at highly threaded work.
Getting back to core of discussion, BD is going to be build on 32nm with power gating just like sandy.So NOT expecting it to clock better than X6 is not wise.
As for the results, they can pretty much be dismissed.
And no, not because im an AMD fanboy.but if it was that horrible, there would be no point in replacing phenom II with this.

for me too

[w0mbat]

no, those scores are real. but theres a bug, the cpu always runns @idle, dont mind what cpuz says...

[informal]

no, those scores are real. but theres a bug, the cpu always runns @idle, dont mind what cpuz says...

Now that is an interesting piece of information
If we assume idle is 800Mhz,as it has always been,and apply 4x (3.2Ghz X8) to C11.5 score of 2.86 ,we arrive to 11.xx score,dangerously close to the score from leaked Donanimhaber slide . I dare not to apply 4x to the poor 3dmark06 score or anything else in that picture since it would go through the benchmarking roof .

[Dimitriman]

And what does running at idle do to it?

[Manicdan]

that kinda explains the .960v we see

^its stuck at lowest multiplier available, 200x4 = 800mhz