Rafael et al,

I just tried getting my iGPU (HD 530) running PPR12M and I get the following error dialog:

Thread 0 starting
Detected GPU 0: Device Emulation (CPU)
Detected compute capability: 9999.9999
Detected emulator! We can't use that!

Any thoughts?

Rafael et al,

I just tried getting my iGPU (HD 530) running PPR12M and I get the following error dialog:

Thread 0 starting
Detected GPU 0: Device Emulation (CPU)
Detected compute capability: 9999.9999
Detected emulator! We can't use that!

Any thoughts?

We had a discussion yesterday about this. For different reasons, everyone agreed that we don't want to support iGPUs. Some of the reasons:

* It's of limited use. iGPUs are much less powerful than discrete GPUs. It wouldn't add much processing power to the overall project. Yes, there's a lot of them, but we don't think many people would actually use them (see below).

* Since most of our apps are limited by memory bandwidth, and the iGPU uses the same memory as the CPU, in many cases the iGPU would be impacting the performance of all the CPU cores.

* Screen lag can be brutal.

* Our past experience with a similar situation: Several people with fast computers were lobbying for the ability to run CPU GFN22 tasks. We did that. There's all sorts of warnings on the web site that this is ONLY for really fast computers. The result? People run it on slow computers that take forever to complete, despite the warnings.

From our perspective, this is something that would provide at best a modest benefit, would be difficult to maintain on our end, and the problems it would likely cause outweigh the potential benefits.

(Please note that this message is not an invitation to discuss why we're so utterly wrong. It's a statement of our position, and it's not likely to change at this time.)

It may seem like you're sitting there with all that unused potential in your computer, but we believe that few people would actually use this if it was available. If you're one of them, app_info is available.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

My hd4000 completes a weiferich/wss unit in about 15 minutes.

ppr12m looks like (which is why I don't try to run it on an igpu, almost a month for 1T is a little much):

gustav:sieve vzimmerman$ ./tpsieve-cl-x86_64-mac -p16325T -P16326T -k3 -K9999 -n9M -N12M -ffppr12M_16325T-16326T.txt -M2 -q
tpsieve version cl-0.2.3e (testing)
Compiled Sep 12 2011 with GCC 4.2.1 (Apple Inc. build 5666) (dot 3)
nstart=9000000, nstep=41
tpsieve initialized: 3 <= k <= 9999, 9000000 <= n < 12000000
Sieve started: 16325000000000000 <= p < 16326000000000000
Thread 0 starting
Detected 256 multiprocessors (1280 SPUs) on device 0.
Device 0 is a Intel HD Graphics 4000.
nstep changed to 32
CL setup complete.
cthread_count = 16384
p=16325000025952257, 432.5K p/sec, 0.04 CPU cores, 0.0% done. ETA 24 Oct 04:04

Running it on a windows box with an hd4400 works as well.

Just make sure you are using the cl version (for AMD).

Reporting an estimated completion date of 2 November....

Zimmer,

With such a lengthy run time, I agree that it is of very little benefit.

The HD530 can get you 150k credits per day with Collatz and which is the only project I know for which one can get a productive use out of their iGPU.

Very unfortunate that it is of no practical use for PG.

I hope nobody is planning on running PPR12M sieving on iGPU. That sieving is quorum 1, we'll never know if the code misses anything. Expect me to be unforgiving over a 1T minimum range missing my informal two-week deadline by that much. It takes 15 hours to sieve 1T on my first-generation Titan and 30 hours on my factory overclocked GTX 570. Two weeks to finish 1T of PPR12M is more than enough.