Boinc CPU Usage Question

[MillerPanda]

Hello, Quick question I hope someone can help with.

Let's say I built a system just to run Boinc and nothing else. Is it better/Faster to let it use 100% of the CPU or scale it back to where it is just under using maxed utilization?

and/or is it something that is different/based upon CPU(ie i5,i7,i9,e5)

I'm sure I am just overthinking it, but not sure if it bottlenecks just slamming it wide open.

Thank you for your time and any input.

[Eigenvektor]

The term bottleneck only really makes sense when you're talking about one component holding back another component. So for example if your GPU can't reach 100% utilization because your CPU isn't fast enough to keep it fed with data to work on.

 

In that case intentionally reducing your CPU's workload might increase the time it has to feed the GPU, increasing the GPU's utilization giving you better overall performance. Alternatively you could increase the workload of the GPU, so that it requires new data less often. That would reduce the GPU-specific workload on the CPU and give it more headroom for other stuff.

 

So if you're running BOINC on your CPU and GPU, it might give you better overall performance if your GPU has a high enough workload that doesn't require constant "refills" from the CPU, so the CPU can spend more time working on its own workload. But if you're just running on the CPU, intentionally reducing its workload will simply reduce the amount of work it will do over time.

 

Reducing workload might make sense if your CPU is thermal throttling, to reduce its temps and keep up clocks, but then a better cooler would have the same effect, while allowing you to run at full utilization.

[Mxyzptlk]

@Eigenvektor how do you reduce or increase the GPU workload in Boinc?

[Eigenvektor]

1 hour ago, Mxyzptlk said:

@Eigenvektor how do you reduce or increase the GPU workload in Boinc?

That was a somewhat theoretical point in the case of Boinc.

 

For games you can increase/decrease resolution or details to increase/decrease GPU workload relative to the CPU. In general a GPU will always run as close to 100% usage as it can⁽¹⁾. The only difference is how many CPU resources are required to achieve that. The best you can do in Boinc is experiment with different projects, see how many CPU resources are required for the GPU to run at 100%.

 

If the GPU is running below ~90-95% usage⁽²⁾, then you're likely CPU limited ("CPU bottleneck"). In that case reducing the number of cores or CPU percentage Boinc is allowed to use should give it more headroom to feed the GPU and likely results in more work being done overall. If the GPU is running at 100% regardless, then reducing CPU load won't help anything (other than reduce heat/noise/power usage).

 

So in regards to OP's question, it is faster to let it use 100% CPU, unless it causes GPU usage to drop well below 100%.

 

—————

(1) In games you can intentionally reduce GPU load by limiting frame rate below the maximum it can achieve – There's no equivalent limit in Boinc afaik

(2) …and you haven't limited frame rate

[Gorgon]

In general you never want to assign all your threads/cores to BOINC, Folding at Home or any other such Distributed Computing Application.

 

You always should leave at a bare minimum (for say a headless Linux System) one thread free for the OS. OSes with more overhead (say a Linux Desktop running X-Windows) will typically need  more resources and Windows tends to need more than a Linux System.

 

This will vary depending on the projects being run and the only way to tell for sure is to run a new project with say 1/2 your available threads and slowly increase the number of threads closely monitoring the yield and see if you hit a point of diminishing returns when approaching using all available threads or, in many cases, see your yields start to dramatically decrease at that point as your system is "thrashing" trying to juggle the Distributed Computing tasks and the overhead needed by the OS.