I am working on a Project of a super computing infrastructure solution HPC.
I am working with FUJITSU servers using INTEL XEON GOLD 6148F processors.
My client is requesting that we inform the theoretical maximum performance (RPEAK) of each server for which we must complete the following formula:
rPEAK = CPUSpeed x # cores X Flops x # CPUS x # Servers
Additionally for the FLOPS we are requested that this value will correspond to the operations of double precision Fused Multiply Add (FMA) and vector instructions AVX-512.
Our set of servers must offer a computing capacity of at least 150TFLOPS, but since we can not clearly have the information of the processors' FLOPS, we can not accurately calculate the number of servers we require.
Thank you for joining the Processors Community. I sure can help with your inquiry.
The theoretical Max Value for the FLOPS of this processor model is:
Theoretical Max Value (FLOPS)
T= 2400000000 * 4* 20 = 192000000000 FLOPS.
Amy, thank you very much.
Thank you very much for your help. I do not know if I can bother you with some questions:
1. The variables in the formula "T", the first value I imagine is the frequency of the processor and the last the # of cores of the processor, but the value of 4 to what corresponds?
2.For the processor that I am using the Theoretical Max Value (FLOPS) would be 192 FLOPS or 192000000000?
3. Why do you use 2400000000 instead of 2.4 GHz?
Do you have documentation about this topic to read?
Sure, see below for the answers:
1. The "T" corresponds to Theoretical Max Value, the "2400000000" to the frequency in Hertz, "20" number of cores and "4" corresponds to a design constant number and it's fixed.
2. It's 192000000000 FLOPS.
3. The formula requires IS units.
You may want to check: https://www.intel.com/content/www/us/en/support/articles/000005755/processors.html Export Compliance Metrics for Intel® Microprocessors