Message boards :
Number crunching :
RAC falling dramatically
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Wiggo Send message Joined: 24 Jan 00 Posts: 36836 Credit: 261,360,520 RAC: 489 |
I was just comparing your rig's times against my i5 2500K @3.4GHz times, which most would think is a far comparison, and all I can say is, WOW! Stop using that iGPU as my CPU completes workunits faster, and so should yours (but we've been telling people that for a few years now). "Shorties" take around 30mins to do while the rest take between 1hr to 1hr 30mins to complete on each of my 2500K's CPU cores (running 2 CPU tasks while feeding its 2x 1060's, 4 CPU tasks when it runs out of GPU work during outages). Cheers. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13855 Credit: 208,696,464 RAC: 304 |
Looks like the Lunatix GPU element eats up too much CPU on my rig. It's not the Lunatics application, it's the fact the iGPU is in the CPU package- they share power, heat, cache and bandwidth limitations. On a system with an addon GPU, even if it uses a CPU core to feed it, the output of the GPU is way more than the loss of the CPU core output. And with very low clocked, low core CPUs like the Celeron, the output from it's built in GPU generally offsets the reduction in CPU output. However- for a multi-core, reasonably high clocked CPU such as yours, the impact of the iGPU processing on CPU performance is so great, that the output of the iGPU may not be enough to offset the loss in CPU processing. Having a look at my graphs, there has been a slight decline in RAC over the last couple of weeks, but it's only been a slight decline where yours has been rather significant. It is possible that drop is due to some of the new work that's been released- not the larger sized WUs, but some of the new GBT tasks take a lot more crunching than the previous ones did. My GTX 1070s would do an average GBT WU in around 7min. Some of these new ones take over 9min. Grant Darwin NT |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I've noticed that too. I couldn't pin it on the 4 bit tasks but the latest BLC5 tasks are taking a couple of minutes longer to run compared to previous BLC tasks. And it doesn't matter whether they are crunched on a 970, 1060 or 1070.s Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I'm also seeing Arecibo "shorties" .... not being so short. Chalk it up to the 4-bitted-ness or something. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Suzuki Send message Joined: 17 Sep 01 Posts: 318 Credit: 4,474,402 RAC: 1 |
Hi all, Looks like disabling the GPU process has done the trick! RAC has been rising sharply since. Thanks for the info. Steve. |
BetelgeuseFive Send message Joined: 6 Jul 99 Posts: 158 Credit: 17,117,787 RAC: 19 |
You still seem to have a lot of GPU tasks on that system. If you do not plan to complete them please abort these tasks so they can be sent to someone else. If you do not abort them they will eventually time out and they will be sent to someone else, but it may be weeks (or even months) before this happens. |
Suzuki Send message Joined: 17 Sep 01 Posts: 318 Credit: 4,474,402 RAC: 1 |
You still seem to have a lot of GPU tasks on that system. Will do - thanks for the advice. Steve. |
betreger Send message Joined: 29 Jun 99 Posts: 11416 Credit: 29,581,041 RAC: 66 |
Looks like disabling the GPU process has done the trick! RAC has been rising sharply since. Once again the iGPU has proven to be a loser. I wonder why the project even has it as an option. |
Iona Send message Joined: 12 Jul 07 Posts: 790 Credit: 22,438,118 RAC: 0 |
Looks like disabling the GPU process has done the trick! RAC has been rising sharply since. I was going to say, because, some people only have the 'on-board' graphics - laptops, more so, but the results from iGPUs seem to be slower than a CPU core. As others have said, it obviously uses resources common to the CPU cores, thereby slowing them down. I have a suspicion that the on-chip cache is heavily involved in this, as an old trick to get old software to run at a useable speed in the days of the P2 (!), was to disable the on-chip cache - the reduction in speed was quite amazing. That may be an over-simplification of the whole picture, though. One silly thought.....has anyone ever tried the on-board iGPU, while doing the customary reservation of a CPU core? All things being as they seem to be, however, I'd have to agree about your query about it's inclusion as an option. Don't take life too seriously, as you'll never come out of it alive! |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
One silly thought.....has anyone ever tried the on-board iGPU, while doing the customary reservation of a CPU core?It has been awhile since I have tried the iGPU in my i5-4460, but it needed at least 2 cores to be shutdown in order for tasks to run somewhat normal - certainly not worth it. |
Wiggo Send message Joined: 24 Jan 00 Posts: 36836 Credit: 261,360,520 RAC: 489 |
A few people over the years have done testing here with Intel's iGPU offerings and the results have always turned out to be that you'll process more work by not using it. The problem always comes back to the shared cache between the CPU cores and the iGPU being so heavily thrashed that cycles are being lost due to more misses than hits happening. Another problem also occurs with stock cooling and poor airflow, and that's thermal throttling caused by the extra heat being produced by the iGPU. Cheers. |
Kissagogo27 Send message Joined: 6 Nov 99 Posts: 716 Credit: 8,032,827 RAC: 62 |
take a look there for some explanations http://lunatics.kwsn.info/index.php/topic,1735.0.html |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.