Bulldozer ES benchmark is out!

[RoyG]

This is a Chinese website: http://www.chiphell.com/thread-210890-1-1.html Use google to translate

2.8Ghz, Vcore: 0.936v
1M Pi: 23sec
Fritz Chess: 19.53

CPU frequency was increased to 3.2G and Pi is 26.7 seconds. Apparently there is a bug.

OP is testing right now, keep refreshing the page.

 

[Arachnotronic]

It looks fake. Look at this screenshot:

http://www.chiphell.com/thread-210890-1-1.html

It says, for L2 cache, "2048 KB x 8"...but in an 8 core BD, it'd be "2048 KB x 4"...

 

[996GT2]

It looks fake. Look at this screenshot:

http://www.chiphell.com/thread-210890-1-1.html

It says, for L2 cache, "2048 KB x 8"...but in an 8 core BD, it'd be "2048 KB x 4"...

The site looks legit to me. Maybe the errors are just due to the fact that it's an ES chip?

If it's real, those benchmark results don't look too great. 23 seconds for Super Pi 1M? That's about twice as long as it takes for a Sandy Bridge based chip clocked at the same speed.

In fact, the Core 2 Duo T8300 @ 2.4 GHz in my Thinkpad does it in about 20-21 seconds...

 

[Arachnotronic]

SuperPi is a single threaded benchmark, though. It's kind of useless for benching modern CPUs.

 

[Tsavo]

I want to see real #'s since we have zero way of knowing what state ES parts are in when they are being 'benched' like this.

 

[996GT2]

SuperPi is a single threaded benchmark, though. It's kind of useless for benching modern CPUs.

Super Pi is a good measure of single threaded performance. If you look at the progression from Pentium 4-->Athlon 64-->Core 2-->Sandy Bridge you'll see a progressive drop in Super Pi 1M times due to increased IPC.

Well, the Cinebench numbers looked pretty disappointing as well...

 

[Tsavo]

SuperPi is a single threaded benchmark, though. It's kind of useless for benching modern CPUs.

It's still useful.

 

[Arachnotronic]

And I'm just not sure if I believe these results. AMD doesn't suck that much

 

[RoyG]

It looks fake. Look at this screenshot:

http://www.chiphell.com/thread-210890-1-1.html

It says, for L2 cache, "2048 KB x 8"...but in an 8 core BD, it'd be "2048 KB x 4"...

I am fairly familiar with this forum and I am sure it is not faked.

 

[RoyG]

CPU frequency was increased to 3.2G and Pi is 26.7 seconds. Apparently there is a bug.

 

[996GT2]

And I'm just not sure if I believe these results. AMD doesn't suck that much

The ES chip was only clocked at 2.8 GHz. What did you expect? It's been known for a long time that Bulldozer is not a high-IPC design like Sandy Bridge, which is why AMD needed to release the chips at much higher clock speeds to even have a chance of being competitive with Sandy Bridge.

You can extrapolate from their results to get an idea of how a 3.5 GHz Bulldozer chip might perform. But even then it still doesn't look too great.

 

[Tsavo]

The ES chip was only clocked at 2.8 GHz. What did you expect? It's been known for a long time that Bulldozer is not a high-IPC design like Sandy Bridge, which is why AMD needed to release the chips at much higher clock speeds to even have a chance of being competitive with Sandy Bridge.

You can extrapolate from their results to get an idea of how a 3.5 GHz Bulldozer chip might perform. But even then it still doesn't look too great.

There's not being a high IPC unit, and then there's something that gets its ass kicked by the cpu in a toaster. :biggrin:

I want real numbers...such a long wait. :|

 

[Skurge]

CPU frequency was increased to 3.2G and Pi is 26.7 seconds. Apparently there is a bug.

Maybe they should underclock it

 

[tweakboy]

Oh these are all numbers, benchmaarks etc. put all aside and go play your favorite game this morning.

 

[Arachnotronic]

The ES chip was only clocked at 2.8 GHz. What did you expect? It's been known for a long time that Bulldozer is not a high-IPC design like Sandy Bridge, which is why AMD needed to release the chips at much higher clock speeds to even have a chance of being competitive with Sandy Bridge.

Sure, but it has 8 cores...you'd think that'd make up for the lower IPC...

 

[zephyrprime]

It looks fake. Look at this screenshot:

http://www.chiphell.com/thread-210890-1-1.html

It says, for L2 cache, "2048 KB x 8"...but in an 8 core BD, it'd be "2048 KB x 4"...

isn't cpu-z often wrong about that sort of thing for new processors?

 

[dma0991]

I'm wondering why is the CPU-Z of this Bulldozer ES is showing only 1 core. Could it be that CPU-Z is unable to read the info correctly or the other 7 cores were disabled by the person who was testing it.

Other CPU-Z screenshots would usually show the proper number of cores and threads.

 

[tweakboy]

Use formerly Everest use, Aida it has benches and compres your rig with others scores. really neat. you can find a evaulation version now. thx

 

[sm625]

You can run multiple instances of super pi. You just need to make a copy of the exectuable and put it in a new folder. When I run 8 instances my average time goes from 18 seconds to about 37. I am curious how fast can a 2600k do 8 instances? If you want to try it here is a batch file to help you run 8 instances:

REM create 8 folders named "1" thru "8" in the same folder as this batch file
REM place a copy of super pi in each folder
cd 1
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 2
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 3
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 4
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 5
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 6
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 7
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

cd 8
del *.txt
del *.dat2
start super_pi_mod.exe
cd ..

 

[podspi]

And I'm just not sure if I believe these results. AMD doesn't suck that much

The fact that the 1055T (@ 2.8ghz stock) can get ~ 21 seconds, and the fact that INCREASING the clockspeed slowed down the Pi calculation implies that something is wrong.

Hopefully AMD has just gimped every ES in some subtle way so that performance couldn't be leaked. If there really are such large problems with recent ES (that weren't intentional) that definitely is NOT a good thing. No wonder the launch was delayed...

 

[996GT2]

The fact that the 1055T (@ 2.8ghz stock) can get ~ 21 seconds, and the fact that INCREASING the clockspeed slowed down the Pi calculation implies that something is wrong.

Hopefully AMD has just gimped every ES in some subtle way so that performance couldn't be leaked. If there really are such large problems with recent ES (that weren't intentional) that definitely is NOT a good thing. No wonder the launch was delayed...

Why would they intentionally gimp an ES to reduce performance by a (seemingly) large margin? I haven't heard of any company doing that intentionally before.

 

[OCGuy]

Either this is some massive disinformation campaign by AMD (doubt it), or whatever that stepping that chip is on had some major issues. I'm thinking I know the reason for the delayed launch now.

 

[piesquared]

Super Pi is a good measure of single threaded performance. If you look at the progression from Pentium 4-->Athlon 64-->Core 2-->Sandy Bridge you'll see a progressive drop in Super Pi 1M times due to increased IPC.

Well, the Cinebench numbers looked pretty disappointing as well...

Super Pi's x87 code is completely outdated, and has been for many years. It shows nothing of modern workloads on modern architectures. It's funny that GPU benchmarks have evolved and have been updated to represent modern architectures (haven't seen any 3D Mark '01 benchmarks for quite a few years). Maybe the GPU market is just that much more progressive, or intel propaganda has seared it into enthusiasts brains that it is a good benchmarking tool since they currently do well in it.

Having said that, these are completely crippled samples, and I fully expect Bulldozer to knock it out of the park in Super Pi but it's irrelevant in modern processing unless you are concerned with legacy x87 code.

 

[Topweasel]

Either this is some massive disinformation campaign by AMD (doubt it), or whatever that stepping that chip is on had some major issues. I'm thinking I know the reason for the delayed launch now.

We know that at least 1 stepping had some problems, my guess is even if its not B01 that was at fault, that there are a lot of B0 and A0 chips that are slipping out to leakers as the new ES chips go out.

Also we don't know anything about the microcode for any of the boards. Could just be that whatever bios is loaded on these boards isn't ready for any of the chips.

This is something we have seen dozens of times leading up to releases. Snapshots taken of slow ES chips, on unfinished BIOS's, with unfinished drivers (though this is a finished board so that's not it).

Benchmarking right now is mostly useless and more likely to be damaging then helpful.

 

[formulav8]

and I fully expect Bulldozer to knock it out of the park in Super Pi.

Whats making you think that? Cinebench didn't look that great either?