Intel Sandy Bridge-EP versus AMD Interlagos

[Sean Evans]

Hello group,

In the 3rd quarter of 2011, Intel will be launching it's next generation CPU with Sandy Bridge-EP, which is expected to feature 8 cores, 16MB of L3 cache (although some rumours put this at 20MB), 4 DDR3 memory controllers, 2 QuickPath 1.1 links and 32 lanes of PCI-Express 3.0.

http://www.realworldtech.com/page.cfm?A ... 191937&p=1

In approximately the same timeframe AMD will be launching Interlagos Opteron, which is based on AMD's new Bulldozer core architecture, Interlagos has 8 Bulldozer modules, thus 16 cores per CPU!

http://www.hpcwire.com/features/Intel-A ... 60704.html

It looks like by the 3rd Quarter of 2011, Interlagos Opteron will finally be better than Intel CPUs for computer chess play!

Perhaps, someone more technical can comment.

Cordially,

Sean

[JCoit]

A couple of comments on this:

1. The most advanced engines have the ability to utilize 8 cores (8 threads), however there are marginal ELO gains going from 4 to 8 cores. Beyond 8 cores some programs actually perform at a lower level, due to overhead and thread-induced inefficiencies. From this perspective, I think it is unwise to make judgement on how good the processor will perform based solely on core count.

2. More cores might be better for running a number of chess engines at the same time, as is the case when running engine vs. engine tournaments or testing of engines.

3. I hope you're right, though... I'd like to see AMD back on top (mainly because they're cheaper processors)

[notyetagm]

A couple of comments on this:

1. The most advanced engines have the ability to utilize 8 cores (8 threads), however there are marginal ELO gains going from 4 to 8 cores.

So what would you prefer for running chess engines, the 6-core Phenom II X6 or the 4-core Core i5-760? MaximumPC magazine just picked the Intel CPU over the 6-core Phenom II X6 as their preferred budget processor.

I just assumed a 6-core CPU would be much better for chess engines than a 4-core CPU. Am I wrong on this? And more importantly, how much better? Please discuss.

[JCoit]

speed-up vs. core count depends heavily on the algorithm and implementation. These days, the engines that scale the best use some form of DTS (Dynamic Tree Splitting). I'll throw in a shameless 'Thanks!' to Dr. Hyatt here. See below:

http://www.cis.uab.edu/hyatt/search.html

+-------------+-----+-----+-----+-----+------+
|# processors| 1 | 2 | 4 | 8 | 16 |
+-------------+-----+-----+-----+-----+------+
|speedup | 1.0 | 2.0 | 3.7 | 6.6 | 11.1 |
+-------------+-----+-----+-----+-----+------+
Table 3 DTS performance results


So let me just dive further down the rabbit hole and make some crazy assumptions.

1. Assume the Phenom x6 and the Core i5 you're looking at have the same single-threaded performance
2. Assume their single-threaded performances scale linearly with Ghz

The i5-760 is a 2.8 Ghz processor, but with 4 cores gets a speedup of 3.7. (2.8 * 3.7) = 10.36 ghz-cores. Or in other words, it would be roughly similar to a single core at 10.36 ghz provided that there are no other bottlenecks in the system.

If the speedup for 6 cores is halfway between the 6.6 and 3.7 shown in Table 3, that'd make it a speedup of about 5.1. So (10.36 / 5.1) = 2.0 Ghz per core.

So if my assumptions were correct, so long as the model of Phenom II x6 you are looking at is operating above 2.0 ghz per core, then it would probably be better performing at chess. But the problem here is that the Core i5-760 has a better single-threaded performance than the Phenom II processor (could be as much as 30% better). At the end of the day, though, doubling the CPUs speed will increase the ELO of your system somewhere around 50 points... rarely a little more, and probably a little less. So my recommendation would be to buy the cheaper processor because they're going to perform roughly the same. Toms Hardware does benchmark CPUs with Fritz...however the numbers they report are shear NPS, not taking in to account the speedups as discussed above.

http://www.tomshardware.com/charts/desk ... ,2419.html

looking at these charts, I'll reiterate that my guess is they're going to perform roughly the same. With the i5-760 getting around 8,500 knps and the Phenom II x6 getting around 10,000 knps. 10,000 / 5.1 = 1,961 and 8,500 / 3.7 = 2,297 suggesting that the i5-760 would actually be a little faster than the 2.8 ghz Phenom II x6 once you account for the higher efficiency it has with only 4 cores vs. 6 cores.