I am bechmarking a server with intel xeon DP 3.6GHZ on a Brandon2 mother board. I wanted to check the performance of the machine with a single processor and i removed a processor leaving the setup with a single Xeon (Nocona)3.6 Ghz DP processor on board.
While running the intel optimised linpack benchmark on this setup to my surprise the result had a processor information with 2.72 Ghz. When i checked the /proc/cpuinfo in RHAS 4 update platform i was using the same frequency was available.
But the BIOS information shows a single processor with 3600Ghz. HT is enabled.
With linux both smp kernel and up kernel are showing the same cpu speed.
What could be the reason for this?
When you say: "When i checked the /proc/cpuinfo in RHAS 4 update platform i was using the same frequency was available." do you mean that /proc/cpuinfo returned 2.7 Ghz or 3.6 Ghz. I'm one of the developers for this particular piece of code, but I just wanted to make certain I understood the issue. Is this LINPACK 2.1.2?
Greg Henry (Intel MKL)
You can reply as quickly or as slowly as you need... But I'm afraid I have bad news for you. There may not be an MKL-related solution here. You see, we have several checks to ensure we get the frequency right in this version of LINPACK, and if it turns out that the OS (i.e., /proc/cpuinfo) disagrees with one of our calculations, we just assume the OS is right. So, one might argue that you could use a version of LINPACK that ignores the /proc/cpuinfo frequency; however, it seems to me that this is just skipping over a problem and not really fixing it. The real problem is why does the OS think that your frequency is 2.7GHz? I'm afraid I don't know that answer. The fact that LINPACK 2.1.2 agrees isn't really meaningful, since as I say we just take the /proc/cpuinfo value when encounter a problem.
As to the best way to follow-up on this, perhaps someone else out there knows something about this OS distribution? It seems unlikely that you'd be the only one to encounter this problem. But I'm guessing one needs to first determine if the OS is right- that things are running at 2.7 Ghz and there is a BIOS setting or something else that is necessary, or that the OS is wrong- in which case the question becomes "why"?
Just some thoughts- I hope we can help you resolve this.
Intel MKL Team
We got an issue when running Linpack on a Harcuvar platform with Denverton (16 Cores, 16 threads, @2GHz). I want to simplify my question:
Why Linpack is running at 800MHz fixed (based on EMON report) while Linux OS is saying it is working @2GHz fixed?
/proc/cpuinfo is saying 2GHz in all 16 cores
Linpack is saying CPU frequency = 800MHz
We have tried everything to fix frequency @2GHz: cpupower, disabling intel_pstate driver, BIOS knobs... In all the cases the Linux OS is saying all cores are running at 2GHz, but when launching the "runme_xeon64" it prints at the beginning "CPU frequency = 800MHz". By running any OS checks like lscpu or /proc/cpuinfo or cpupower frequency-info, all of them say CPU frequency = 2GHz, only Linpack and EMON are saying CPU = 800MHz.
This is really weird for us since EMON reported CPU average frequency = 800MHz, seems that Linpack ran at 800MHz fixed.