CPU Speed Test: How to Measure and Improve Your Processor's Performance

[member 428835]

Hi PF!

I'd like to test how fast my cpu clock speed is. Any recommendations with benchmarks so I can see just where I fit?

 

[jim mcnamara]

Okay, what system exactly do you have? OS and version, cpu details. After that there are test suites to answer your questions. But we cannot recommend anything until we know.

 

[member 656954]

Have you looked at sites such as https://www.cpubenchmark.net which provide indicators of speed? But is this just an 'of interest' question or do you think your CPU is running slow?

 

[member 428835]

Okay, what system exactly do you have? OS and version, cpu details. After that there are test suites to answer your questions. But we cannot recommend anything until we know.

Ubuntu 18.04, AMD epyc 7281.

Have you looked at sites such as https://www.cpubenchmark.net which provide indicators of speed? But is this just an 'of interest' question or do you think your CPU is running slow?

I have not; didn't know where to start. I'm running some cfd (computational fluid dynamics) simulations and think my computer could go a lot faster. I'm in the process of comparing a standard simulation to several benchmarked cases, but I thought it would also be useful to test it outside of the cfd software I am using.

 

[member 656954]

AMD epyc 7281

Here is the benchmark result, https://www.cpubenchmark.net/cpu_lookup.php?cpu=AMD+EPYC+7281&id=3130&multi=2, ranked 195, so not that fast.

 

[member 428835]

Here is the benchmark result, https://www.cpubenchmark.net/cpu_lookup.php?cpu=AMD+EPYC+7281&id=3130&multi=2, ranked 195, so not that fast.

But how can I test mine to see if I've built it wrong.

Also (not relevant) may not be the fastest but I bought it for 500$ and just built it (have very little computer knowledge) with a grant I spent two years on and am so proud of it. For me, this is by far the fastest machine I've ever worked on. So, fo me, it IS that fast ;)

 

[jim mcnamara]

Your disk speed affects things like virtual memory allocation. One way to speed up memory intensive apps is to have the pagefile on a SCSI disk. Or increase the amount of RAM. Or both.

I do not know what you are doing, but fluid dynamics is not a trivial exercise, and algorithms are exceedingly important for performance. Often about 80% of bottlenecks result from your algorithms.

Do you know how to profile? See your gprof man page for more information.
Python profilers are also available for profiling python applications -- since you seem to mention python a lot.

 

[member 428835]

Your disk speed affects things like virtual memory allocation. One way to speed up memory intensive apps is to have the pagefile on a SCSI disk. Or increase the amount of RAM. Or both.

Sorry, I don't know what a SCSI disk is. I have 8x16GB of ram, so not sure I can do much better given the cpu limitations.

I do not know what you are doing, but fluid dynamics is not a trivial exercise, and algorithms are exceedingly important for performance. Often about 80% of bottlenecks result from your algorithms.

Do you know how to profile? See your gprof man page for more information.
Python profilers are also available for profiling python applications -- since you seem to mention python a lot.

Yea, I spoke with several cfd experts who run the same program I do and they recommended the hardware based on my cost. But the way the cfd program works, the algorithms are already in place. I'm just a user.

I'm really wondering if there is a generic test I can run to see how fast I am going and how fast I should/can be going. Perhaps none exist, but I don't know until I ask the experts (you all).

 

[DrClaude]

You should run some benchmarks, such as the LINPACK benchmark, on your computer. You can find online some results, so you can compare and see how your particular computer is doing.

Check for instance http://www.roylongbottom.org.uk/linux benchmarks.htm

 

[jim mcnamara]

Phoronix has a lot of specialty benchmarking software, all free. The site is devoted to this kind of reporting and testing.

https://www.phoronix-test-suite.com/

Do some reading on that site as well. If you are going to be using Linux, learn it as well as your application software. It pays back big dividends.

 

[sysprog]

CPU-Z is an excellent freeware performance reporter that is available at https://www.cpuid.com/

 

[Vanadium 50]

CPU-Z is an excellent freeware performance reporter

But he's running Linux.

 

[Vanadium 50]

AMD epyc 7281

This chip has 16 cores and can run 32 simultaneous processes. If you are not doing that, you are using only a fraction of the chip's capabilities. (And before you ask "how do I do that", it's a feature of the "cfd software")

 

[sysprog]

But he's running Linux.

Thanks -- it was remiss of me to not notice that -- there's an open-source ported-to-Ubuntu version called CPU-G available at http://ppa.launchpad.net/atareao/atareao/ubuntu/pool/main/c/cpu-g/

 

[jim mcnamara]

@joshmccraney - I made a boo-boo earlier - for virtual memory performance you probably want an SSD (solid state drive). They are often SCSI disks (how the disk/computer send metadata to each other about drive operations - a protocol flavor)

 

[sysprog]

@joshmccraney - I made a boo-boo earlier - for virtual memory performance you probably want an SSD (solid state drive). They are often SCSI disks (how the disk/computer send metadata to each other about drive operations - a protocol flavor)

It's true that almost all SSDs run via SCSI rather than by the older IDE interface; however. most SSDs are NAND-based devices, which are addressable at a block level, as distinguished from byte-addressable, slower, NOR-based devices -- ref: https://en.wikipedia.org/wiki/Flash_memory

 

[MikeeMiracle]

You need to look at the number of cores you have and if the program you run can utilise them. A problem with early multicore hips was even though they had say 4 cores running at 2Ghz, the program were only programmed to use 1 core so the other 3 just sat there. There is more to your question than just "How fast is my CPU running" and "Is that enough to run this program." For example your CPU can only process data that it has, if you have slow disk drives and slow memory then most of the time your CPU is sitting there doing nothing. Having 8x16GB of ram won't be useful if it can't be utilised if you have bottle necks elseware.

For me the CPU, Motherboard and memory are the most important items, The motherboard arguably the most important and everything plugs into it and if it's a budget board it doesn't matter how good the other components are if they can't talk to each other fast enough due to limitations in the board.

 

[jim mcnamara]

@MikeeMiracle good points. Which reminded me: the cpupower command and CPU "parking" plus cpu affinity settings which should/would/maybe/who-knows have been done at the application code level.

What Mikee said is that it not just the CPU. It is what the motherboard, memory, OS, and disks do (or not) as well.

Python runs well, IMO. ...so maybe we can assume that affinity is set and therefore those aspects of data localization are fine.

Do you know the python method sched_setaffinity()?
http://sorami-chi.hateblo.jp/entry/2016/04/29/000000

Find out more:
Linux taskset command: https://www.howtoforge.com/linux-taskset-command/

But cpupower settings for high performance are almost always up to the the user. Benchmarking code that I am aware of turns on "high cpu performance". The intent is to max out possible speed and remove unnecessary bottlenecks. Which does not mean that python "automagically" does that by default. AFAIK.
@PeterDonis probably knows.

How to use cpupower, which is relevant for AMD especially:
https://linux.die.net/man/1/cpupower

 

[member 428835]

Thank you all for your help! When I get better internet (thanks Coronavirus) I'll go deeper into this. I will say I've been running all 16 cores in parallel. Motherboard is Supermicro MBD-H11SSL-I, so not a budget board. I have 1 TB SSD drive I boot Ubuntu up with. Plenty of power and cooling too. Overall I think a solid build. Just wanted to make sure I had everything configured, so I really appreciate all the responses!

 

[Svein]

I would check out SiSoft Sandra (it may be available for Linux, I haven't checked).

 

[bigfooted]

Compiler options also have a large impact on computational times, especially things like -g -O0 versus a more optimized -O3. So if you are compiling a cfd code like openfoam, also check the compiler settings. You should also check your hyperthreading settings. the command 'lscpu' is useful for this. My workstation has hyperthreading enabled, so my 16-core machine shows up as having 32 cpus (2 sockets with 8-core cpu's, with 2 threads per core). So I tell my cfd software to run on 32 cores.

lscpu also gives use a 'bogus' performance indicator: bogomips. Mine is 4400, I hope yours is better (I have an old machine).

 

[davenn]

It's true that almost all SSDs run via SCSI rather than by the older IDE interface;

no it isn't true

SATA is the standard for the last at least 10 years, it took over from IDE

 

[sysprog]

no it isn't true

SATA is the standard for the last at least 10 years, it took over from IDE

I agree that my 'almost all' remark was incorrect. SCSI or SAS is used in some systems in which greater speed and reliability are especially important. Compared to SCSI or SAS, SATA is less expensive, not as fast, and usually higher capacity.