Intel integrated GPU - Page 7

Bdot · 2013-11-04, 14:54

Quote:

Originally Posted by Prime95

I don't have a params.h file.

This means you did not look at the mfakto sources, but just at the kernels which will still be called using the workaround.

Anyway, I noticed that even on my box, VectorSize=1 yields wrong results. I'm still analyzing why - it may have to do with the different result of comparisons.

kracker · 2014-03-10, 01:25

Don't want to poke too much but... is running mfakto on -thread title- still a slight possibility?

TheMawn · 2014-03-10, 01:59

Don't be too quick to neglect the thermals. If you're running mfakto on the iGPU, you're enough of a basket case that running Prime on the the CPU is a given.

To give you an idea, my laptop CPU (the only example of an iGPU I have) gets to ~72C running two Prime95 workers (leaving two of four logical threads idle) and I've seen 84C playing Starcraft 2. It officially supports two cores but is occasionally able to divert small tasks to two additional cores (which would ironically be the same two physical cores in my case) so I think it's a safe assumption to say that it, at the very best, also runs both cores to 100%.

This means the extra 12C comes from the iGPU (definitely stressed to the max).

Even if the mfakto does end up working, make sure you have the thermal overhead.

kracker · 2014-03-10, 02:12

:very confused: Thermals? Yes, that would be something to think about, but this thread is about just getting it to run.

My thinking is coming from your first phrase.

Prime95 · 2014-03-10, 03:28

Quote:

Originally Posted by kracker

Don't want to poke too much but... is running mfakto on -thread title- still a slight possibility?

It is theoretically possible. Debugging is required.

I would have investigated myself, but I couldn't get prime95 to run stable on Windows 8 and Windows 7 had a weird continuous interrupt problem siphoning off CPU time. I was forced into Linux and Intel does not support OpenCL on Linux.

kriesel · 2017-07-11, 00:31

Thanks for all your past efforts. Is https://github.com/Bdot42/mfakto current?

Quote:

Originally Posted by Bdot

This means you did not look at the mfakto sources, but just at the kernels which will still be called using the workaround.

Anyway, I noticed that even on my box, VectorSize=1 yields wrong results. I'm still analyzing why - it may have to do with the different result of comparisons.

Was there a resolution to the VectorSize=1 issue?
Time and hardware design progress marched on, and I'm now in possession of a system that claims (via mfakto 0.15pre6, the latest version I could find) to have two Intel gpus ("platforms"). I've coaxed mfakto into life on it sort of. It gets as far as appearing to be launching testcase 1/34071, then GPU-Z reports the gpu clock returns to the idle speed, gpu power drops to 0.1W or less, GPU Load 1% or so, and it gives me plenty of time to read the ini file more carefully. Where I find and finally read rather than skim, the following:

Code:

# Different GPUs may have their best performance with different kernels
# Here, you can give a hint to mfakto on how to optimize the kernels.
#
# Possible values:
# GPUType=AUTO             try to auto-detect, if that does not work: let me know
# GPUType=GCN              Tahiti et al. (HD77xx-HD79xx), also assumed for unknown devices.
# GPUType=VLIW4            Cayman (HD69xx)
# GPUType=VLIW5            most other AMD GPUs (HD4xxx, HD5xxx, HD62xx-HD68xx)
# GPUType=APU              all APUs (C-30 - C-60, E-240 - E-450, A2-3200 - A8-3870K) not sure if the &quot;small&quot; APUs would work better as VLIW5.
# GPUType=CPU              all CPUs (when GPU not found, or forced to CPU)
# GPUType=NVIDIA           reserved for Nvidia-OpenCL. Currently mapped to &quot;CPU&quot; and not yet functional on Nvidia Hardware.
# GPUType=INTEL            reserved for Intel-OpenCL (e.g. HD4000). Not yet functional.
#
# Default: GPUType=AUTO

GPUType=INTEL, and Not yet functional. Oh. (Please don't say, well run clLucas on it instead, because I also ran into roadblocks there.) On the plus side, it seems to have no impact on Prime95 throughput http://www.mersenneforum.org/images/...xtra/smile.gif

kriesel · 2018-10-20, 01:58

Quote:

Originally Posted by kriesel

GPUType=INTEL, and Not yet functional.

Was able to get mfakto v0.15pre6 running on an Intel Hd620. Then I installed some SDKs and it stopped working. Performance was 18ghzD/day without Prime95 also running, 20 with. See also https://www.mersenneforum.org/showthread.php?t=23394

ixfd64 · 2019-06-23, 20:45

Another data point: I recently got a new MacBook Pro with an Intel Iris Plus Graphics 640 integrated GPU. Although mfakto recognizes the device, all 30 self-tests fail. Even with GPU sieving disabled, there are still 13 failures.

kriesel · 2019-06-23, 21:02

Quote:

Originally Posted by kriesel

Was able to get mfakto v0.15pre6 running on an Intel Hd620. Then I installed some SDKs and it stopped working. Performance was 18ghzD/day without Prime95 also running, 20 with. See also https://www.mersenneforum.org/showthread.php?t=23394

Updates:

I uninstalled the SDKs, reinstalled OpenCl, and got mfakto function back. To run mfakto on it costs around half of the prime95 throughput, since they're sharing the same TDP budget, and the same memory bandwidth. But it's a net GhzD/day gain for that system. Running mfakto can make its screen updates sluggish though.

On another laptop, i7-8750H which includes uhd630, plus an internal gtx1050Ti, there seems to be little effect of prime95, mfakto, and CUDA apps on each other, and the screen which is served by the uhd630 only is still responsive. (Its GTX1050Ti is excluded from display use, reserved for compute use.)

ixfd64 · 2019-10-25, 21:32

I recently began experimenting with Intel HD Graphics again. On my Windows machines, mfakto does not detect the GPU at all. It does detect the GPU on macOS, but not all tests pass.

With GPUType=INTEL and GPU sieving enabled, all tests fail. If I turn off GPU sieving, then 13 out of 17 tests pass:

Quote:

ERROR: selftest failed for M50896831 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M50830523 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M50752613 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M51507913 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M51157933 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M51671491 (cl_barrett32_76)
no factor found
ERROR: selftest failed for M50805581 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M51157429 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M51760133 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M51090757 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M50989481 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M50856937 (cl_barrett32_87)
no factor found
ERROR: selftest failed for M45448679 (cl_barrett32_87)#
no factor found

However, using GPUType=GCN results in only four failures whether or not GPU sieving is enabled:

Quote:

ERROR: selftest failed for M51760133 (cl_barrett32_87_gs)
no factor found
ERROR: selftest failed for M51090757 (cl_barrett32_87_gs)
no factor found
ERROR: selftest failed for M50989481 (cl_barrett32_87_gs)
no factor found
ERROR: selftest failed for M50856937 (cl_barrett32_87_gs)
no factor found

For anyone who got mfakto to work on their integrated GPU, could you please share your setup?

kriesel · 2019-10-26, 05:08

Quote:

Originally Posted by ixfd64

I recently began experimenting with Intel HD Graphics again. On my Windows machines, mfakto does not detect the GPU at all. It does detect the GPU on macOS, but not all tests pass.

With GPUType=INTEL and GPU sieving enabled, all tests fail. If I turn off GPU sieving, then 13 out of 17 tests pass:

However, using GPUType=GCN results in only four failures whether or not GPU sieving is enabled:

For anyone who got mfakto to work on their integrated GPU, could you please share your setup?

1) Windows 10 on i7-8750H which includes a UHD630, mfakto v0.15pre6
2) Windows 10 on i7-7500U which includes an HD620, mfakto v0.15pre6; Vector size 2, grid size 4, gputype auto, gpusieving enabled.
34026/34026 on the long selftest; 30/30 on the short.

A working OpenCl driver is essential of course.
A separate OpenCl install checker can be useful.
Parameters can be examined in GPU-Z, advanced tab, OpenCl category, for example.

2014-03-10, 02:12	#70
kracker "Mr. Meeseeks" Jan 2012 California, USA 37×59 Posts	:very confused: Thermals? Yes, that would be something to think about, but this thread is about just getting it to run. My thinking is coming from your first phrase. Last fiddled with by kracker on 2014-03-10 at 02:13

2019-06-23, 20:45	#74
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 2³×313 Posts	Another data point: I recently got a new MacBook Pro with an Intel Iris Plus Graphics 640 integrated GPU. Although mfakto recognizes the device, all 30 self-tests fail. Even with GPU sieving disabled, there are still 13 failures. Last fiddled with by ixfd64 on 2019-06-23 at 20:56

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Can I run my CPU's integrated GPU along with my discrete GPU?	Red Raven	GPU Computing	9	2014-10-24 02:01
New integrated CPU-GPU programming paradigm	Dubslow	GPU Computing	1	2012-02-15 08:45
Ivy Bridge integrated GPU?	Dubslow	GPU Computing	7	2011-11-18 23:36
Can I use integrated graphics alongside a GPU?	mdettweiler	GPU Computing	9	2010-09-15 19:41
turn off your integrated Snd card in CMOS	nngs	Hardware	0	2005-05-20 01:31

2014-03-10, 01:25	#68
kracker "Mr. Meeseeks" Jan 2012 California, USA 37·59 Posts	Don't want to poke too much but... is running mfakto on -thread title- still a slight possibility?

2014-03-10, 01:59	#69
TheMawn May 2013 East. Always East. 1727₁₀ Posts	Don't be too quick to neglect the thermals. If you're running mfakto on the iGPU, you're enough of a basket case that running Prime on the the CPU is a given. To give you an idea, my laptop CPU (the only example of an iGPU I have) gets to ~72C running two Prime95 workers (leaving two of four logical threads idle) and I've seen 84C playing Starcraft 2. It officially supports two cores but is occasionally able to divert small tasks to two additional cores (which would ironically be the same two physical cores in my case) so I think it's a safe assumption to say that it, at the very best, also runs both cores to 100%. This means the extra 12C comes from the iGPU (definitely stressed to the max). Even if the mfakto does end up working, make sure you have the thermal overhead.