CPU speed and hyperthreading

[marypoppins]

Dear All,

I would like to ask is there any deeper differences between for example an Xeon E3-1220V2 3.1GHz and Xeon Quad-Core X5450 3GHz in virtualizing?

Because these cpus have high differencies in the price, and as I try to plan the required horsepower those seem the same the point of view of MHz.

As I know the normalized speed is CPUcore x CPUfrequencies so the above two cpus have near the same normalized speed (4core x ~3Ghz)?

Also I would like to ask whether the hyperthreading relevant to the vCPU calculating? Because the HT can improve performance 30% only certain situation.

So I totally ignored HT capability in my calculating, however it seems 'doubles' the cpu.

I'm calculating based on the normalized form: I have a xeon 4core 3GHz CPU in the old hardware so I have a grand total 4corex3000Mhz=12000MHz. I also measured that the OS using a total CPUpeak 60% CPU, while avg total 2%!!! .  So it seems the old system only use 12000x0.6=7200MHz at the peak load. Because of this I should have at least this 7200MHz in the new virtual system (and it can be a cpu with 2 cores 7200/2=3600MHz or a 6 cores 7200/6=1200MHz) and I sucessed in keep the CPU power. And HT is totally forgotten in this logic conclusion. Could you please confirm is this correct, or I made a big mistake somewhere?

Thank you very much

[rcporto]

I will recommend you go to Intel Xeon E3-1220 v2... there is no much difference in clock speed, but the Intel Xeon E3 will gives you not only Hyper-threading but VT-x EPT that is very important to virtualization. Here is a link about EPT: http://www.vmware.com/pdf/Perf_ESX_Intel-EPT-eval.pdf

And here is a comparison of two processors: http://ark.intel.com/compare/34446,65734

[marypoppins]

Thank you for your answer, however the two cpus were only examples. The question is without the architectural differencies (bus type, fsb, cache) and other capabilities, there is no difference between cpu-s in virtualization point of view if the normalized speed is the same? So there are cores and frequencies and multiply that two parameteres to compute the normalized speed is enough to compare the cpus (roughly), is it right? Or is there something other parameter (for example HT) which has high importance, and therefore I can change the calculation formula when choose CPU?

So in one word the above mentioned exapmle with the 7200MHz is correct roughly speaking?

Then in the next step I have to care the little pieces of puzzle, including that you answered (thank you) the HT and other EPT capabilities, cache size etc...

[PeteKoehler]

When stepping up to a newer generation of chip, one may have some performance improvements that will likely not show up as a result of reviewing the specs or comparing clock cycle speed.  For instance, check out my post Vroom! Scaling up Virtual Machines in vSphere to meet performance requirements.  You will see a substantial difference between the old Intel Harpertown based chipset, and the relatively newer Nehalem and SandyBridge chipsets.  If given the option, choose the newest chipset possible, and models that were intended for servers.