Dual Xeon Passmark Confusion

[Jae]

I am confused by some of the scores. I am about to buy a 2697v2. It gets the best score on the single processor passmark. It doesn't do nearly as well in the 2 cpu test. I understand that there's no logical reason this can't be true. But I am not sure I get it here.

First, I thought the 2697 has better systems to handle the greater number of cores. So I don't see why it would scale worse than other e5 Xeons.

Second, the tests have some other odd results:

The multiple cpu test has as number 1 an 8 core processor. But it isn't the 2687wv2. It is the one with a lower clock speed. (Sorry, I am on an ipad and can't look at the model numbers.) This result makes no sense to me. The only difference between these two processors is that the 2687w is faster and has a higher TDP.

In addition, the dual processor tests for the 2697v2 have a wide range of scores. The fastest score is on a machine that has a fast graphics card. The slower scores had worse cards. But why would this matter? Does the multiple processor Passmark somehow factor in the graphics card and the drives?

I would be very grateful for any insight. I am ordering my work station in the morning. I live on the east coast of the US.

Thanks!

[David (PassMark)]

The E5-2697 v2 has more cores, but lower clock speed.
The E5-2687W v2 has less cores but a higher clock speed.
We don't have examples of these hugely expensive CPUs to play with, but I would speculate that at some point adding more cores doesn't help. RAM and or the algorithms in use will become a bottleneck. At that point a small number of faster cores start to win out.

I think you are also saying that these two CPUs are the same and should score the same,
Intel Xeon E5-2667 v2 @ 3.30GHz
Intel Xeon E5-2687W v2 @ 3.40GHz
However they are not the same. The clock speed is different, the turbo speed is different, the amount of cache is different, one is a 22nm part while the other is a 32nm part, they support different RAM speeds, etc...
So there is no reason to think that they should give identical performance.

The GPU doesn't have any significant effect on the CPU score. This might change at some point in the future as the line between a CPU and GPU blurs, but for the moment the GPU doesn't effect the CPU results in PerformanceTest. Other factors must be the cause of any difference.

[Jae]

Thanks for the reply.

Going in reverse order, you have the faster cpu 2687wv2 doing worse on the dual cpu test. I'm not sure how that can make any sense. To the extent there are any differences, all of them favor the 2687wv2. (Also, I thought they were both 22nm as they are both version 2. But again if that is wrong, everything favors the 2687wv2. And that is reflected in the single cpu testing where the 2687wv2 wins.). The difference here cannot be due to scaling since each has the same number of corse.

On point one, I agree that part of the issue is parallel overhead. And that may be the explanation. On the other hand, dual 2697v2 beats dual 2687wv2 on a ton of benchmarks. And it wins by a lot on your single cpu test. That alone suggests that memory bottlenecks aren't a huge problem on a single chip.

Most oddly to me, there is one dual cpu test you list for the 2697v2 where it is far faster than anything else. Any idea why there is so much variance in the test scores for the same processors? That's when I started to wonder if graphics cards or something other than cpu could affect the score.

[edit: I verified that they are both 22nm. See http://ark.intel.com/products/75273/...ache-3_30-GHz]

Again, much appreciated. Please let me know if you have thoughts on these additional points.

Best

[David (PassMark)]

Actually you are right, I was looking at the wrong part number. There are too many CPUs on the market will too little difference between them. The Xeon E5-2667 v2 @ 3.30GHz
Intel Xeon E5-2687W v2 @ 3.40GHz are fairly close in their specs.

So it isn't clear why the Dual CPU Xeon E5-2667 v2 is scoring around 10% higher than expected. I suspect is partially because these are rare very high end CPUs (compared to the overall CPU market) and our average result isn't made up of many different machines.

Most oddly to me, there is one dual cpu test you list for the 2697v2 where it is far faster than anything else.

The average at the moment is a CPUMark of 23,526 for the E5-2697 v2.
The min is 18314, the max is 27078. There are several result near the max so there isn't an outlier. If anything it is the 18314 value that is unusually low. But having an odd low value from a large number of samples is common. It is easy to misconfigure a machine so that it runs slow.

[Venom]

I just bought a E5-2697v2 and I plan to buy a second for dual CPUs....

However, after reading this, I wonder which would be the more preferable all around CPU?

I do some gaming and a lot of Maya (VRay, etc.)

So I am wondering if I should go dual E5-2697v2 (for more cores), or switch to dual E5-2687W v2 (for less cost higher clock speed)?

[David (PassMark)]

Venom, this seems to be a duplicate post with this one.
http://www.passmark.com/forum/showth...eon-E5-2687Wv2