DP FLOPS of intel xeon 5680

[fatalme]

Hi, I have used intel optimized LINPACK benchmark tool and I got a result of 70 GFLOPS on Intel xeon 5680. I think this should be for SP FLOP. Anybody knows the GFLOPS for DP FLOP? And how to calculate the theoretical peak performance for DP FLOP?

Thanks.

[TimP]

Peak DP performance is 3.33x10^9 ticks/sec (if Turbo is disabled) x 4 flops/tick x 12 cores (if HT is disabled) so your performance is a little short of expectation.

[merwin101]

Quoting TimP (Intel)

Peak DP performance is 3.33x10^9 ticks/sec (if Turbo is disabled) x 4 flops/tick x 12 cores (if HT is disabled) so your performance is a little short of expectation.

Doesn't the formula also hold if hyperthreading is enabled? My understanding is that even with hyperthreading enabled, each core can still execute at most 4 flops per tick, so the formula for peak DP performance in FLOPS is still the same. Hyperthreading only improves performance when there are less than 4 instructions ready for a given thread in a clock cycle, right? Or am I misunderstanding hyperthreading (and if this is the case, is there an additional x2 for hyperthreading, or is the factor something else?)

Thanks.

[TimP]

Sorry, you're right, the Xeon architecture doesn't supply additional floating point resource for hyperthreading.

[SKost]

Quoting fatalme

Hi, I have used intel optimized LINPACK benchmark tool...

[SergeyK] Where could I download it from? Could you give a link, please?

and I got a result of 70 GFLOPS on Intel xeon 5680. I think this should be for SP FLOP. Anybody knows the GFLOPS for DP FLOP? And how to calculate the theoretical peak performance for DP FLOP?

Thanks.

[SKost]

Quoting Sergey Kostrov

Quoting fatalme

Hi, I have used intel optimized LINPACK benchmark tool...

[SergeyK] Where could I download it from? Could you give a link, please?

...

That was easy to find:

http://software.intel.com/en-us/articles/intel-linpack-benchmark-download-license-agreement/