First Bulldozer benchmarks (dual Interlagos)

[Sharktooth]

Phoronix received some interesting data from an AMD server partner. The test platform is a 32 cores system (2x 16 cores CPU). Even if the clockspeed is pretty low (1.8 Ghz) the results appear to be really good.
Have a look: http://www.phoronix.com/scan.php?page=article&item=amd_bulldozer_linux&num=2

[aegisofrime]

Saw that yesterday, however trying to get an idea of performance as compared to Sandy Bridge is nearly impossible due to the fact that its 1. An engineering sample 2. 2x 16-core Interlagos and 3. 1.8 Ghz clock speed as opposed to the supposed 3.5Ghz~ launch speed of Desktop Bulldozer.

Some forum discussion of it, from playing around with the core count and clock speed seem to conclude that clock for clock it will lose to Sandy Bridge. Then we have this:


Work is planned to integrate x264 with the Sandy Bridge's encoding
ASIC for improved encoding performance. Current status is: waiting on
Intel (these guys move at the speed of an obese one-legged paraplegic
three-toed sloth swimming down a river of frozen helium while chained
to an osmium anchor stuck inside a black hole).

From the x264 development newsletter 12 (http://mailman.videolan.org/pipermail/x264-devel/2011-February/008293.html)

How much that boosts x264 encoding is unknown, of course. Still its one factor in favor of Sandy Bridge against Bulldozer.

Then we have this:

http://www.xbitlabs.com/cms/include/image.php?src=/images/news/2011-03/amd-pos.png&width=550&height=456&cache=1&quality=75&aspect=0&format=png

From Xbitlabs
(http://www.xbitlabs.com/news/cpu/display/20110313163718_AMD_Aims_to_Fight_Core_i7_Sandy_Bridge_with_Bulldozer.html)
If this is real (some forums have been calling it fake, although they sound suspiciously like AMD Fanboys), then AMD themselves are basically admitting that they can't match Sandy Bridge in terms of raw performance, they are going to win in thread count a la Thuban vs Nehalem.

As you have seen, I have been cracking my head over my next PC upgrade :p

[MasterNobody]

Then we have this:
Work is planned to integrate x264 with the Sandy Bridge's encoding
ASIC for improved encoding performance. Current status is: waiting on
Intel (these guys move at the speed of an obese one-legged paraplegic
three-toed sloth swimming down a river of frozen helium while chained
to an osmium anchor stuck inside a black hole).
From the x264 development newsletter 12 (http://mailman.videolan.org/pipermail/x264-devel/2011-February/008293.html)

How much that boosts x264 encoding is unknown, of course. Still its one factor in favor of Sandy Bridge against Bulldozer.
Currently this is zero boost with VERY doubtful chance of increasing more than zero because Intel seems don't want to expose details and low level API to QuickSync hardware. You can see this from the adjectives in this quote :)

[Sharktooth]

aegisofrime, dont forget the bulldozer architecture defines "cores" as a 4 pipes integer unit.
when we talk about bulldozer "cores" we should keep in mind they're not full fledged cores. the correct comparison would be 1 bulldozer module vs 1 sandybridge core.
if you keep that in mind you will see numbers are not so bad... ;)

[aegisofrime]

Yes, so I should be trying to calculate the score for a 8 core/4 Module Bulldozer right? The formula most people used:

25.97 (time taken for 32 core/16 module Interlagos) * 4 = 103.88

Adjusting for clock speed (since the CPU in question is 1.8 Ghz: Lets give Zambezi 3.6 Ghz) : 103.88 / 2 = 51.94

Thus, extrapolating the numbers, a 4 module / 8 core Zambezi will do the C-Ray benchmark in around 51.94 seconds.

By comparison, a i7-2600K does it in 37.79 seconds:

http://openbenchmarking.org/result/1103237-SPLA-SB5349780

Sandy Bridge is still looking pretty good... And I'm still wondering about the authenticity of that slide from Xbitlabs... :mad:

Mathcraft is fun! Reminds me of my Warcraft 3/Defense of the Ancients days :p

[Sharktooth]

ppl just assumed 32 cores are twice as fast as 16... that's not properly true.
threading overhead is not taken into account as well as bus bandwidth saturation.
however, numbers are not that bad expecially considering the fact there are still no compiler optimizations for bulldozer CPUs.

[aegisofrime]

I agree, the numbers aren't too bad, definitely not as bad as the Intel fanboys screaming "AMD is bad!"

Still, I wonder if it's too early to make a decision to get a i7-2600K? I really need the extra power (double x264 encoding speed of my Phenom II 955!). So far it seems like Bulldozer should perform on par with the 2600K,no?

Still, I also saw a picture of a supposed Cinebench benchmark showing a Bulldozer scoring double the points of a i7-980X(!!!)

[zinga]

Article relating to topic which may be of interest: http://www.realworldtech.com/page.cfm?ArticleID=RWT033011040021

Altogether, the results present a far from clear picture. Some of the tests show a significant performance advantage for Bulldozer – which is impressive, given that the focus was on increasing frequency more than IPC. Yet others show dire regressions reminiscent of the P4. Worse still, the benchmark data seems inconsistent at times. Perhaps most frustrating is that the sheer quantity of missing information makes it very difficult to reconcile the importance of the results. Compiler optimizations alone could easily change performance by 30% in some cases, and libraries can have a bigger impact still.

Given all the uncertainty surrounding the benchmark data, there are no hard and fast conclusions. The quality of the data only lends itself to vague and general impressions. Overall, the data suggests that for a number of applications, Bulldozer will have comparable IPC to its predecessor; sometimes better, sometimes slightly worse. Yet at the same time, the data also implies a very real risk that some workloads may hit particular bottlenecks in the architecture and suffer greatly. This seems to weakly support AMD's claims that Bulldozer will not substantially reduce IPC in favor of frequency. For more concrete answers about Bulldozer performance, we will have to wait for more information, preferably in the form of real benchmarks or real products or ideally both.

[aegisofrime]

That's the risk you take with a brand new architecture. With Pentium 4 (Netburst) to Core (Banias), Intel succeeded beyond their wildest dreams. Might we be seeing a failure here?

Honestly, the rhetoric from JF-AMD (AMD Director of Server Products) hasn't been very encouraging. He has constantly be trying to saying that single-threaded performance no longer matters and that multi-threaded is where it's at. That seems to imply to me that IPC won't be a great improvement over Phenom II.