[bindesboll]

Originally Posted by RodriguezFatz

What about one motherboard with two CPUs? Do they have twice the memory bandwidth, compared with a single CPU board?

The number of memory channels is associated with the individual CPU. So yes, two CPUs on one motherboard will totally have twice the memory bandwidth as a single CPU (of same type).

Originally Posted by RodriguezFatz

Do you get any improvement for 8-core license, when using two CPUs in one workstation?

I guess your question is: "Does a system with dual 4-core CPUs on one motherboard perform better than a single 8-core CPU?" (correct me if I am wrong).

As it appears in one of my previous posts above regarding scaling efficiency:

Originally Posted by bindesboll

From 4 to 8 cores: 0.62 (clearly memorybandwidth limited, more cores will not increase speed).
From 8 to 16 cores (Motherboard CPU interconnect): 0.92

The scaling from 4 to 8 cores is within a single CPU, showing a pretty poor scaling (0.62), which indicates that the calculation power of the 8-cores is restricted by the external bandwidth.
However, the scaling from 8 to 16 cores (going from 8 cores in a single CPU to 8 cores in two CPUs on the same motherboard) is very good (0.92), which indicates that the motherboard CPU-interconnect is adequate.
Conclusion: A system of dual 4-core CPUs will have a performance that is better than a single 8-core CPU by roughly a factor 0.92/0.62 = 1.5.

Please note that these numbers are based on cluster with Xeon E5-2670 (Sandy Bridge EP), whereas the new Xeon E5 26xx v2 (Ivy Bridge) might scale better as it has increased memory speed and cache. However, still I would expect the most efficient 8-core system to be a dual CPU motherboard with 2 x Xeon E5-2637 v2.

Best regards
Kim Bindesbøll

[f-w]

Does anyone have any updates on this topic. How about actual benchmarks of a single node with dual e5-2600(series)v2 CPU's vs. a cluster of i7 nodes?

[evcelica]

Originally Posted by f-w

Does anyone have any updates on this topic. How about actual benchmarks of a single node with dual e5-2600(series)v2 CPU's vs. a cluster of i7 nodes?

I tested dual XEON-E5 vs 2 i7s. Post is here:
http://www.cfd-online.com/Forums/har...-3930k-x2.html

I also tested the exact same benchmark on only one i7 and saw that it solved in almost exactly twice the time. Performance scaling from 1 to two machines was 100.2%. Better than perfect with infiniband.

In the past I had used gigE and saw 99% efficient performance scaling going from 1 to two machines, but this was on a large problem, smaller models do not scale as well of course.

[acasas]

Maybe its too late, but please , check this out

http://www.cfd-online.com/Forums/har...5-2650-v3.html