Folding Forum

Joe_H wrote:iGPU support is on the "maybe" list. They did get a lot of offers for software development help, and there was a very long list of wished for items to be worked on. Will have to wait and see which make it out to actual release. If this does make it out, watch for announcements here or on the Twitter or Facebook F@h sites.

Would that really be beneficial? I would think that making use of the IGPU would run temperatures even higher than CPU WUs do already, taking TDP and cooling capacity away from those CPU WUs. And after all, it's being mentioned regularly, that CPUs are needed for certain projects.

If FAHBench is anything go by, I guess yes:

Single Precision
OpenCL Score: 8.58 (my crappy 1050 laptop edition produces about 30)
CPU Score: 1.85

Double Precision:
OpenCL Score: 3.02
CPU Score: 1.86

But you have a point, what matters is real-world mixed performance, and perhaps there's greater need for CPU-specific WU performance.

Breach wrote:If FAHBench is anything go by, I guess yes:

Single Precision
OpenCL Score: 8.58 (my crappy 1050 laptop edition produces about 30)
CPU Score: 1.85

Double Precision:
OpenCL Score: 3.02
CPU Score: 1.86

But you have a point, what matters is real-world mixed performance, and perhaps there's greater need for CPU-specific WU performance.

From the comments on different threads and from what I see, it's GPU slots that are still running idle every now and then waiting for WUs. I haven't seen my CPU slots waiting for work in the past few weeks.
I guess for science it might be beneficial to have the GPU WUs done by the more powerful GPUs and and use CPUs for CPU WUs instead of running a GPU WU on a IGPU which most probably would take longer and delay the project's overall progress.
Also for CPU WUs it might be beneficial to have the entire TDP and cooling capacity available in order to run the WUs as quickly as possible on the given hardware rather than having to share TDP between IGPU and CPU cores.
The IGPU would possibly accumulate more points for the donor than the CPU cores, but how do we support science the best way?

Why the Assignment Server copy has not been updated to to match the current version which has these as ":::" instead of :3:0: is being checked into.

HugoNotte wrote:

Joe_H wrote:iGPU support is on the "maybe" list. They did get a lot of offers for software development help, and there was a very long list of wished for items to be worked on. Will have to wait and see which make it out to actual release. If this does make it out, watch for announcements here or on the Twitter or Facebook F@h sites.

Would that really be beneficial? I would think that making use of the IGPU would run temperatures even higher than CPU WUs do already, taking TDP and cooling capacity away from those CPU WUs. And after all, it's being mentioned regularly, that CPUs are needed for certain projects.

Slow CPU cores more than could double in PPD.
Even bigger CPUs, with better IGPs, like the top of the line 10th gen Intel CPUs, could potentially increase tremendously in PPD.
However, it would only be recommended in Desktop CPUs.
On many laptops (like the Pentium N5000, Celeron N processors in Laptops), the heat becomes an issue.

If you compare to Boinc, which does support Intel IGPs, those mobile CPUs are able to either fold on CPU, or fold on IGP (with 1 or 2 threads max).
Once you enable more than 2 threads, and IGP maxed out, they will thermal throttle.
One could provide active (fan) cooling on those laptops, to keep the temperatures below 90c.
But the real candidates are the desktop CPUs, which have better coolers (and where cooler upgrades are feasible).

Or, if you're lucky to get your hands on those single board computers, with an Intel mobile CPU (for industrial use) you can also easily upgrade the cooling solutions.

HugoNotte wrote:Would that really be beneficial? I would think that making use of the IGPU would run temperatures even higher than CPU WUs do already, taking TDP and cooling capacity away from those CPU WUs. And after all, it's being mentioned regularly, that CPUs are needed for certain projects.

Assuming that support will be provided, you don't need to use it. All systems are, to some degree, limited by the capabilities of the power supply and by the system's ability to cool itself. You are the one who ends up deciding what resources you choose to donate ... within the capabilities of your system.

The AVX instructions on your CPU generate more heat and increase throughput, but they may end up being throttled by settings in your BIOS. The same would be true for the instructions processed by an iGPU.

MeeLee wrote:

HugoNotte wrote:

Joe_H wrote:iGPU support is on the "maybe" list. They did get a lot of offers for software development help, and there was a very long list of wished for items to be worked on. Will have to wait and see which make it out to actual release. If this does make it out, watch for announcements here or on the Twitter or Facebook F@h sites.

Would that really be beneficial? I would think that making use of the IGPU would run temperatures even higher than CPU WUs do already, taking TDP and cooling capacity away from those CPU WUs. And after all, it's being mentioned regularly, that CPUs are needed for certain projects.

Slow CPU cores more than could double in PPD.
Even bigger CPUs, with better IGPs, like the top of the line 10th gen Intel CPUs, could potentially increase tremendously in PPD.
However, it would only be recommended in Desktop CPUs.
On many laptops (like the Pentium N5000, Celeron N processors in Laptops), the heat becomes an issue.

If you compare to Boinc, which does support Intel IGPs, those mobile CPUs are able to either fold on CPU, or fold on IGP (with 1 or 2 threads max).
Once you enable more than 2 threads, and IGP maxed out, they will thermal throttle.
One could provide active (fan) cooling on those laptops, to keep the temperatures below 90c.
But the real candidates are the desktop CPUs, which have better coolers (and where cooler upgrades are feasible).

Or, if you're lucky to get your hands on those single board computers, with an Intel mobile CPU (for industrial use) you can also easily upgrade the cooling solutions.

I hear you and agree, that an IGPU might well produce more PPD than the CPU it's attached to. My point is, would it be better for science? GPUs are already running idle every now and then, CPUs are needed for certain projects. Would it make sense from a science point of view to enable a weak IGPU to fold a WU in much longer time than a dedicated GPU would or rather make use of the 2, 4 or more CPU threads?
I'm not looking at PPD here, my argument is, that there is lots of work for CPUs.

FAHBench also supports the iGPU.

What 10th Gen Intel iGPU do you have? Does it show on GPU-Z?

HugoNotte wrote: I hear you and agree, that an IGPU might well produce more PPD than the CPU it's attached to. My point is, would it be better for science? GPUs are already running idle every now and then, CPUs are needed for certain projects. Would it make sense from a science point of view to enable a weak IGPU to fold a WU in much longer time than a dedicated GPU would or rather make use of the 2, 4 or more CPU threads?
I'm not looking at PPD here, my argument is, that there is lots of work for CPUs.

I would think that there are 2 types of WUs to do.
One, which GPUs can do.
And one, which need the complexity of CPUs to process the WU.
If you can offload the GPU WUs to GPUs that are very fast,
CPU WUs won't benefit from IGPs as they're not compatible, however, if some of the 32bit FPP could be offloaded to the IGP, it could potentially accelerate CPU WUs.
To find someone who full time tinkers with a PC to make it work isn't very easy; and I don't think FAH has the budget for it.
It's entirely possible they are hoping for someone in the open source community to chime in (just like they do with people folding, or helping out in the forum; mostly volunteers).