Xeon vs Xeon

[cineon_lut]

I have to get some new hardware for work. I am looking at used X5650 nodes, versus some new E5-2620s. I am having trouble finding real world comparisons between the two. I know the E5-2670s are much faster and more expensive, but I'm trying to get a bang for the buck here. E5-2620s in new hardware are 4x the price of used X5650s. Which means I can get 4x as many 5650s.

Anyone have experience with one versus the other? This will be for ray trace rendering CG. I have a bunch of 5650s now, and they're great. Fast, reliable, just a great all around experience.

Thanks!

[pinhodecarlos]

I have. For prime search E5-2670 are 3 times faster than X5650 because of the former ability of taking advantage of AVX technology.

[cineon_lut]

Thanks, Carlos. I would expect 2670s to be faster in their own right, but that AVX advantage is impressive. The raytrace renderer we use (Vray from Chaos Group) does not have an advantage with AVX for any but a few types of scenes. The ones I was looking at from a cost standpoint were 2620s.

[pinhodecarlos]

When talking about pure GHz the E5's are only 20-30 % faster than the Xeon at the same stock speed. Right now I am running a factoring program on both servers and only notice a slice better performance of the E5's against the Xeon. But when I run prime search there's a major boost. Example:
Prime test of numbers k*2^1700000-1
E5520 at 2.27 GHz: 2800 seconds (no AVX)
E5-2630 v2 at 2.60 GHz: 1160 seconds (with AVX)

E5 2.4x faster than the Xeon. Not too far away from my first value on my last post although stock speeds are different. Also don't forget that the E5 will consume less energy than the Xeon making it more efficient.

[pinhodecarlos]

Another example as the number tested gets larger.

Prime test of numbers k*2^4000000-1, for k=5
E5520 at 2.27 GHz: 14500 seconds (no AVX)
E5-2630 v2 at 2.60 GHz: 5500 seconds (with AVX)

[Slicker]

I've got a couple E5-2640 v2 @ 2.00 Ghz servers that I used on the NFS challenge. Each is a dual processor box in a 1U case. The case got really hot - like heat your coffee hot - in one place but they ran fine for the entire challenge. Maybe HP uses the as a partial heat sink. We have a couple pre-Core 2 Xeon servers that they are replacing and are 6x faster and use less power. The power savings was the main reason for the upgrade. The speed is just a bonus.

[pinhodecarlos]

About the heat generation I can't give an opinion because both servers are in controlled ambient. I found several primes on each of the servers that were confirmed by Top 5000 prime database.