AMD FX-8120P benchmark from Coolaler

[BlueBlazer]

Another leaked benchmark, this time from Coolaler >> [AMD Bulldozer FX-8120P] how Taiwan, like the moon with the foreign round! Put a XtremeSystems bulldozers measured multi-test pattern (orignal [AMD Bulldozer FX-8120P]原來台灣的月亮跟國外的一樣圓!放一張滄者極限所測的推土機多工測試圖). The translator's messed up, by the way.

Seems he's comparing it with VR-Zone's benchmarks, and the scores look similar (Coolaler's result slightly higher by only 0.03 points)... :hmm:

Edit: It also seems there's another newer CPUZ version? :hmm:

 

[inf64]

That's the performance it seems. Not good at all.
Thuban will roll over 8120 badly...

 

[Arzachel]

That's the performance it seems. Not good at all.
Thuban will roll over 8120 badly...

That should be the first red flag.

 

[Idontcare]

If these leaks of BD have any bearing in reality then it is entirely clear now why Dirk was fired back in Jan as that would have been around the time AMD would have had silicon in hand well enough to project the veracity of the microarchitecture.

 

[Avalon]

It doesn't make any sense. Why would AMD release something that poor when they could have moved Thuban over to 32nm and added extra cores + tweaks? Power constraints? Die size limits?

 

[formulav8]

Do you have to keep sticking your tongue out as us Blazer.

 

[cantholdanymore]

Out of ignorance. What is a CEO to do with the u-arch? Shouldn't it be the CTO?

 

[LOL_Wut_Axel]

LOL, it can't be THAT bad. That's worse than a Phenom II X6 1075T, and that's a CPU available for only $160.

 

[formulav8]

Well hopefully they will be able to tweak the next revision to really help bd's shortcoming.

 

[nonameo]

incredibly frustrating. Hard to refute when 2 different groups give you the same thing.

Seriously, it has 8 cores on a new architecture, more cache, etc. and can't beat a 6 core thuban with only a few more 100 mhz? Just scrap that crap and go back with stars. This is a downgrade.

 

[Arachnotronic]

Coolaler is a reputable leaker, so I'm going to choose to believe these.

 

[inf64]

Yes it makes no sense people. But that's how things stand right now. This perf. level is in press briefings. Something is broken in the uarchitecture since 9 moths ago we had a projection (by AMD themselves) on 3 workloads : 1 render test(C11.5!),1 pcmark and 1 3dmark06 subtest. Cinebench was showed as having the greatest speedup versus thuban,1.78x. Now it can't touch Thuban.

 

[formulav8]

LOL, it can't be THAT bad. That's worse than a Phenom II X6 1075T, and that's a CPU available for only $160.

As much as I would like BD to be a powerhouse, it unfortunately can be that bad since AMD has no other recourse but to release it anyways.

Hopefully its just that AMD's samples were debug versions but this late in the game I believe that the latest samples are the real deal...

I just can't see how their simulations could be that far off. Its similar to Ph1 simulations I guess. It showed the cpu in a much better light than in the real world.

 

[BlueBlazer]

Do you have to keep sticking your tongue out as us Blazer.

I like the tongue smiley!

If these leaks of BD have any bearing in reality then it is entirely clear now why Dirk was fired back in Jan as that would have been around the time AMD would have had silicon in hand well enough to project the veracity of the microarchitecture.

Probably they did not expect Sandy Bridge to improve so much. Looking the scores, it matches 3-year old Core i7 but Intel wasn't sitting still with their aggressive tick-tocks. :hmm:

 

[LOL_Wut_Axel]

As much as I would like BD to be a powerhouse, it unfortunately can be that bad since AMD has no other recourse but to release it anyways.

Hopefully its just that AMD's samples were debug versions but this late in the game I believe that the latest samples are the real deal...

I just can't see how their simulations could be that far off. Its similar to Ph1 simulations I guess. It showed the cpu in a much better light than in the real world.

LOL. WHY is it so slow?

Sandy Bridge will mop the floor with this. Even Llano would be faster. How can something be so hyped yet so disappointing? I'm betting on the module design being way worse than 180% scaling, more like 150%.

But I guess the fact that they're selling you a highly overclockable Eight-Core that Turbos to 4GHz for $230 should tell you something. Everything that keeps coming out is bad news, apart from the overclocking.

 

[LOL_Wut_Axel]

I like the tongue smiley!

Probably they did not expect Sandy Bridge to improve so much. Looking the scores, it matches 3-year old Core i7 but Intel wasn't sitting still with their aggressive tick-tocks. :hmm:

The Phenom II X6 1100T at least matches the Core i5 2500 in very multi-threaded, but this is just terrible.

 

[formulav8]

I like the tongue smiley!

Probably they did not expect Sandy Bridge to improve so much. Looking the scores, it matches 3-year old Core i7 but Intel wasn't sitting still with their aggressive tick-tocks. :hmm:

Honestly, SB is not much of an improvement over Nehalem in most cases as it relates to performance?

 

[BlueBlazer]

Honestly, SB is not much of an improvement over Nehalem in most cases as it relates to performance?

Depending on the application used IMHO. On this particular benchmark which Phenom II X6 was faster than Nehalem due to having 6 real cores, but Sandy Bridge outperforms the Phenom II X6 which is quite an achievement considering having only 4 real cores (like Nehalem). Somehow Sandy Bridge's multi-threading (hyper-threading) performance is much better. We can also see that in other multi-threaded applications.

 

[podspi]

I honestly can't imagine BD being released with that kind of performance. They already have Stars (Husky) on 32nm, it makes better sense to just shrink that and run with it, since the die size would actually be lower (we're comparing Thuban on 45nm to Zambezi on 32nm now).

As IDC said, if this actually is the performance, it is obvious why Dirk was let go. Yes, the CEO does not directly design a chip, but the fact that he allowed such a project go gold would prove incompetence*.


*I still don't believe it.

Edit: Also, while I DO believe this leak is legit, it is still an ES. Might be gimped in some way...

 

[GaiaHunter]

If these are real results, there must be a serious problem in the architecture - on paper the cores seem to be much beefier, even sharing a FPU and front end, should turn faster.

Guess the most likely culprit is cache problems.

If AMD release these CPUs with this bottleneck somewhere, the only explanation is that this bottleneck doesn't affect all workloads/benchmarks.

 

[formulav8]

LOL. WHY is it so slow?

Sandy Bridge will mop the floor with this. Even Llano would be faster. How can something be so hyped yet so disappointing? I'm betting on the module design being way worse than 180% scaling, more like 150%.

But I guess the fact that they're selling you a highly overclockable Eight-Core that Turbos to 4GHz for $230 should tell you something. Everything that keeps coming out is bad news, apart from the overclocking.

Well BD was designed when AMd had fabs and definitely die size to worry about. Performance was probably secondary at the time compared to saving die space. Well not secondary but they would have had to make some compromises they not have made otherwise.

But i'm not exactly sure why it performs as bad as it might be. I guess the shared design isn't near as efficient as AMD hoped for. Like someone else said, the module approach is more like 50% of a real core than 80% like AMD mentioned.

 

[Phynaz]

Not good. Really, really hoping this is not true.

 

[Vesku]

I would like to think they wouldn't release gen 1 BD if these results are true and not due to a pre-release issue that is fixed with the final silicon. I wouldn't say these are so bad they are fake. Instead, I would say they remain unverifiable. If somehow this is how the retail product performs then AMD's pricing is too high and I'll be keeping an eye out for a deal on an Phenom II x6 before they go out of production.

Is this going to be a bizarro launch where BD low thread count is faster than SB but multithread is weak (to explain pricing above x6)? Feels a bit FUDdy to me.

 

[LOL_Wut_Axel]

I honestly can't imagine BD being released with that kind of performance. They already have Stars (Husky) on 32nm, it makes better sense to just shrink that and run with it, since the die size would actually be lower (we're comparing Thuban on 45nm to Zambezi on 32nm now).

As IDC said, if this actually is the performance, it is obvious why Dirk was let go. Yes, the CEO does not directly design a chip, but the fact that he allowed such a project go gold would prove incompetence*.


*I still don't believe it.

Edit: Also, while I DO believe this leak is legit, it is still an ES. Might be gimped in some way...

The worst thing is that Bulldozer is supposed to be an architecture to be used as the foundation for some years. If the foundation is so weak, like the Pentium 4, the only saving grace is extremely high clock speeds. Even in that, the Pentium 4 was much better. While AMD only achieved 2GHz clock speeds with K8, you could reach 4GHz on the Pentium 4. With BD the clock speeds only seem to be around 500MHz higher.

Looks like a lot of bad compromises and design decisions were made for desktops.

 

[psolord]

Can you guys please not get so depressed please? You are depressing me as well, lol.

This is AMD playing as. Haven't you learned anything from past years?

The benchmarks may very well be real. This is not final hardware. JF warned us about all this. At the very least, wait for the launch day to get disappointed!