Joe_H wrote:I just checked the current version of GPUs.txt and one Intel entry is enabled. I am assuming the entry matches for the device ID iGPU you have in your i9-9900.
I just caused a download of the latest GPUs.txt file to one of my systems, the line looks like this:
VinceMH wrote:...Does pause on start just allow me to manually start after a PC restart?...
That's correct. Regardless of the previous state of the Slot, when the client starts, the Slots will be in paused status so you can resume folding on all or few of the slots, depending on what you would like to do.
ETA:
Now ↞ Very Soon ↔ Soon ↔ Soon-ish ↔ Not Soon ↠ End Of Time
Each slot operates independently. Assuming you now have three slots. A CPU slot, an Intel GPU slot (non-functional), and a Nvidia GPU slot you can set the PAUSE-ON-START globally or you can set it for individual slots. I'm assuming you'd want to leave (only) the Intel slot paused.
The experimental testing of iGPUs was intended to be a 'back-room' event but some aspects have been exposed and you're one of the "lucky" ones who inadvertently got picked to help. The slot will probably stay and won't get a new assignment -- at least until testing is completed.
Whoops! We're experimenting with enabling support for Intel iGPUs, and due to a mistake on my part, some of these Intel iGPU test projects made it out into full FAH. We've fixed this so it doesn't happen again.
Apologies for this, and thanks so much for all your help and support!
@JohnChodera :
If you're working with intel IGPs, the beignet drivers offer OpenCL 1.2 to the oldest of Core i (3rd to like 5th gen or so), and Atom processors as well.
They'd be 12 shader units max, running at a max of 500Mhz.
That happened to me, and I picked up several WU and completed them. I didn't mention because I was afraid to jinx it but if it was just a test that leaked into production that is not bad. I am all for adding as many GPU and other types of accelerators as possible. While I feel for every person that has been impacted by all of the scientific research conducted by FAH; COVID needs to be solved as quickly as possible. It is is all theory until it hits someone near you and losing a 1st cousin to it has focused my perspective. I am willing to baby-sit creaky drivers and old equipment just to get a few more WU solved.
Joe_H wrote:Support will not be going that far back for GPU folding.
I'm just assuming, if the newer Intel IGPs are going to be supported by using opencl, that (on Linux at least) those beignet drivers may make it easy for supporting those older CPUs as well.
I'm also wondering if it would be feasible to, instead of have Intel GPUs process WUs, to use Intel CPU WUs, and accelerate them with the iGPU, by letting the IGP help out the CPU in the small stuff it can do, to lighten the load on the CPU.
The capabilities of those older iGPUs are just not sufficient and there are issues with the hardware support from what I understand, so they will not be targeted for GPU folding.
A later version of Gromacs than used in the current CPU A7 folding core does have an option to offload some vector processing to GPUs, including the Intel IGPs, but that would take a new folding core. When such a new folding core might be released is not something I have knowledge of. In any case a new core would have to be tested, and validated before release. What level of GPU could be supported would also need to be determined.
iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
MeeLee wrote:@JohnChodera :
If you're working with intel IGPs, the beignet drivers offer OpenCL 1.2 to the oldest of Core i (3rd to like 5th gen or so), and Atom processors as well.
They'd be 12 shader units max, running at a max of 500Mhz.
Somehow a 12 shader GPU doesn't fit very well in the GPU world as seen by today's FAH.
Early iGPs were supported by early versions of Intel's OpenCL driver and it was full of bugs. I don't remember when they finally developed a version of OpenCL that actually worked but FAH can't afford to introduce errors into the science so there's a lot of hesitancy in even testing older versions. Intel, themselves, recommend not wasting FAH's time testing anything older than 7th Gen.
On the other hand, recent versions of GROMACS can support dual use of Intel processors ... doing part of as WU on the CPU threads and part of the same WU with the parallel capabilities of the iGP. That sounds intriguing, doesn't it. Although it could enhance the current CPU-only orientation of FAHCore_A7 I wonder how frequently the divided processing would have to sync up. Part of the WU processing would have to wait for the other part -- probably at every step.
I was surprised to see that iGPU slot added after debugging core interrupted issues happening since a few days (Basically deleted everything and reinstalled). I have a i7-8750H with that exact Intel UHD Graphics 630. Itself never managed to get any work so i set pause-on-start to True for the meanwhile until it's available.