More Bulldozer Benchmarks (Supposedly from AMD)

[996GT2]

According to latest reports, AMD will launch it is highly anticipated FX CPU family at 12 October but before that press deck presentation and test results from AMD are finally unveiled.

Source: VR-Zone Forum.

Full presentation; http://www.donanimhaber.com/islemci/galerileri/AMD-Bulldozer-FX-resmi-test-sonuclari.htm[/QUOTE]

 

[dac7nco]

We'll see. If I can find these slides or corroborating data on AMD.com by this evening I'll continue to take you seriously.

Daimon

 

[gmaster456]

Techpowerup also posted these, so at least they are coming from a reliable source
http://www.techpowerup.com/152569/A...-Core-i7-2600K-in-the-Eye-AMD-Benchmarks.html

 

[inf64]

Actually the results in the slides are not bad at all. Take a look at this one:
http://img.donanimhaber.com/images/haber/amdfxpressdeck_12a_dh_fx57.jpg

You can see that SMT in case of 2600K vs 2500K brings a lot (8 threads vs 4) in terms of performance and not many of those listed applications scale perfectly with more cores/threads. So SMT kind of already extracted a lot of parallelism from the code and I doubt that bringing more cores in the case of intel (say SB-E) would increase the results by 50% . Well maybe in few cases it would,like wprime and Pov ray.In the rest I really doubt it would.

If you look at the 8150 vs 2500K numbers,the average is around 23.5% in favor of FX,give or take a few. That's pretty good advantage on desktop and it's more than what SMT (2600K) gets over non-SMT (2500K). We do need more applications that scale well with many cores to see this more clearly, but the point is that AMD's "cores" will scale better and will also use SMT in SIMD workloads to extract the ILP from applications that are not coded so well for many threads (just like i7 does).

I have a question though. Why is there no model number for FX listed in the slide above?? All other slides have 8150 listed but they are not compute intensive test but gaming ones. In this case FX6xxx and even 4xxx is just as good as FX8xxx. Anyone seen the "back up" slides that have system info listed?

 

[BlueBlazer]

These slides look suspect, probably fakes. Just look at the Cinebench R11.5 slide, as mentioned here earlier......

Are these slides made up? Looking at the Cinebench R11.5 numbers........

Exactly same as here >> Cinebench score of 5.95 @ stock and 7.8 @ 4.8GHz with a FX8150 mentioned on XS forums. :hmm:

Those numbers are from someone in Vietnam (who who supposedly have one)? :hmm:

 

[IndyColtsFan]

Even if they are straight from AMD, I'm not sure I'd trust them to configure an Intel system to run against in a benchmark, nor would I trust Intel to configure an AMD box if the situations were reversed.

At any rate, even if we accept these at face value, we're looking at performance more or less along the lines of a 2600K, give or take a few percent on certain tests. In my case, I'm sitting on an unopened 2600K and board and have been debating returning them for BD, but I'm not sure I see enough here to justify returning them.

 

[StrangerGuy]

Right, if these slides were true why AMD is still so quiet on actual benchmarks. They should have already done a preview like for the HD 4850, it's not like Intel wouldn't know how good BD is before release.

 

[IndyColtsFan]

Right, if these slides were true why AMD is still so quiet on actual benchmarks. They should have already done a preview like for the HD 4850, it's not like Intel wouldn't know how good BD is before release.

That's right, and I thought DonanimHaber had posted known fakes before. That makes me more skeptical.

 

[Arkadrel]

Soo... AMD slides claiming they can beat the 2500k, and almost match the 2600k.

AMD made huge improvements over the Phenom II if right... the benchmarks most impressive are not the gameing ones, but rather the software ones (those where the ones AMD was doing worst in compaired to Intel).

Not bad and if their prices are right... ~245$ for the top model...
thats actually gonna give Intel a run for its money (until SB-E is out).

I think these #s are real, because the slides look "real" (like the AMD ones).
*IF* they are Im extremly happy, this means AMD is nearly at 2600k levels in all area's.

 

[BlueBlazer]

That's right, and I thought DonanimHaber had posted known fakes before. That makes me more skeptical.

Another possible OBR copy-cat?

Two other oddities mentioned here......

They look pretty legit. The only thing that bothers me/looks fishy is that if you look at the Cinebench screen shot in the background the benchmark hasn't even been run. Further, they're not even using an AMD system if you look at the system info. It's a Xeon running Max OS X.

Also, and correct me if I'm wrong, but isn't 5.95 for BD a great score since it only has 4 FP units?

Edit:

Oh that's why... they took the image from the Cinebench site? http://www.maxon.net/uploads/pics/cinebench_screen_09.jpg

And here.....

Looking at their wPrime32M scores:

2600k is about 13% or so faster than the 2500k.

Real test scores:
http://www.bit-tech.net/hardware/cpus/2011/01/03/intel-sandy-bridge-review/7

2600k = 7.222 sec
2500k = 10.70 sec

That's 32.5% quicker. How is "AMD's" 2600k 20% slower than the 2500k in this benchmark?

 

[AtenRa]

Those slides are from AMD's Tech Day of 31 August 2011,

It may be that up until that day, AMD only had ES silicon and the slides where made with ES performance CPUs. That could explain why there is no CPU model reference for the performance slide.

Perhaps im wrong but, i believe that those numbers are low for BD in multithreading.

 

[trollolo]

the fact that these charts have no units on their x/y axis makes me rage with the force of a thousand suns

 

[Idontcare]

the fact that these charts have no units on their x/y axis makes me rage with the force of a thousand suns

dude, it's been long-established that both the x-axis and the y-axis goes to 11 for all things bulldozer.

 

[inf64]

Those slides are from AMD's Tech Day of 31 August 2011,

It may be that up until that day, AMD only had ES silicon and the slides where made with ES performance CPUs. That could explain why there is no CPU model reference for the performance slide.

Perhaps im wrong but, i believe that those numbers are low for BD in multithreading.

I was wondering about the most important slide in the deck too. Strange it only was listed as "FX" . No model number no nothing. Gaming slides have the model number but we all know games use up to 4 cores ,well at least 95% of games out there. Select few can make use of more but that is not that relevant. So even FX4xxx will be comparable in gaming performance to FX8150. The performance slide on the other hand encompasses applications that may make use of more cores and here we have no model number.

 

[LOL_Wut_Axel]

Why would AMD compare against a CPU that's been discontinued? The Core i7-980 performs the same and costs ~$415 less...

Also, it seems like they're cherry picking, using the 980X on some areas and the 2500K and 2600K on others.

 

[BlueBlazer]

Those slides are from AMD's Tech Day of 31 August 2011,

It may be that up until that day, AMD only had ES silicon and the slides where made with ES performance CPUs. That could explain why there is no CPU model reference for the performance slide.

Perhaps im wrong but, i believe that those numbers are low for BD in multithreading.

As I was speaking about OBR, saw this on his blog....

ps. thanks my friend F. for this slides ...

No thanks to the OBR watermarks. :hmm:

 

[Rubycon]

dude, it's been long-established that both the x-axis and the y-axis goes to 11 for all things bulldozer.

What about Z? Does anyone care about 3D any more? :biggrin:

 

[blackened23]

the fact that these charts have no units on their x/y axis makes me rage with the force of a thousand suns

How many ponies do you have in your life

 

[Venom20]

Even if BD doesn't beat the i7, a simple cost reduction would still give it a run for intels money. I'll be content with power somewhere between the i5 and the i7.

 

[LondonBurning]

I want to believe...

 

[bunnyfubbles]

At any rate, even if we accept these at face value, we're looking at performance more or less along the lines of a 2600K, give or take a few percent on certain tests. In my case, I'm sitting on an unopened 2600K and board and have been debating returning them for BD, but I'm not sure I see enough here to justify returning them.

They compare Bulldozer against the 980X for gaming (of which we know the 2600K is typically faster across the board in that regard) and only directly compare Bulldozer's multithreaded performance against the 2600K, of which the 2600K comes out on top more often than not. I don't like to think about the single or low threaded scenarios, where it seems likely AMD would get crushed.

Then there's the fact that the 8150 is a 125W TDP chip vs. the 2600K @ 95W TDP.

Honestly, if these slides are true, then all it really suggests is that AMD is conceding to anything at or above the 2600K level and positioning their flagship 8150 against the i5 2500K.

dude, it's been long-established that both the x-axis and the y-axis goes to 11 for all things bulldozer.

http://www.youtube.com/watch?v=EbVKWCpNFhY

What about Z? Does anyone care about 3D any more? :biggrin:

It's an acquired taste, otherwise we'd have more games with jetpacks (also preferably with blackjack and hookers)

 

[MentalIlness]

Totally pathetic. And I don't mean the slides. I mean "if the slides are real" the people still hoping they are fake just because they are in bed with Intel.

 

[krumme]

If these slides are not fake, those cherry picked bm does show that BD is not suited for the not important high-end desktop market. BD 4/8 core is like 320mm2 right? sans gpu, so the 2600 is far more efficient here. Its like comparing the NV570 to the far more efficient AMD6970. But at least AMD have something that is compettitive on the pure performance side.

 

[inf64]

Yes krumme, desktop market doesn't like CPUs with more cores. It like less cores and more IPC and clock. If your product fits this description (SB 2600K fits perfectly) ,then you are set in client space. FX just partially fits this description : it has more clocks but not the IPC of SB . Also it has 8 integer cores which means more power.

However, let's see how well this client geared product in the now new (server) form of SB-E will go against Interlagos on highly parallel workloads that can be recompiled to make use of FMA4 and XOP. I suspect that in this case,16C interlagos will have clear advantage in both integer and floating point throughput metrics.

 

[krumme]

Yes. But when will the FMA4 and XOP recompile happen, and to what degree in the server environment?

And we need the most important part, good power measurements. Really looking forward to Johans review.