Cpu\ gpu max clock balance for boinc?
8 Feb 2017 11:24:22 UTC
Topic 205273
(moderation:
Need help
Does any one know what the balance of cpu\gpu should be to get maximum computation speed?
AMD FX[tm}-63006core 3.50 gig clockable to 3.8 gig
R5 200 1gig memory card
Model GA-78LMT-S2
Clockable to
cpu 400 MHz to 750 MHz
memory to 500 to 850 MHz
When I try to speed up boinc projects I get comutation errors.
I have been told I need the right balance of gpu and memory on card.
I wonder if I need to overclock the cpu also.
Any help I can get would be greatly appreciated.
Firebyrdman
Speeding up, or overclocking,
)
Speeding up, or overclocking, is often not a good thing for crunching, yes it can be done and it can work but the circumstances have to be just right, right on board ram, right cpu, right gpu, good ariflow in and out of the pc etc, etc. The easiest way is to crunch at stock speeds for a bit to get a good baseline set of numbers, then SLOWLY increase the speed on ONE device at a time until you get errors then back off to where you were getting nothing butt good results. THEN slowly overclock the next device until you get errors, then again back it off until you were getting nothing but good results, then go back to the first device and overclock it again until you get errors, etc, etc. It's usually a combination of things that can make things faster, BUT each component of your pc will be stressed as you overclock them and probably shorten their lifespan a bit. There is also usually one component that prevents you from overclocking as far as you might otherwise want too and that is then on your list of possible parts to upgrade. At some projects some people have actually found underclocking gives better results if they try to run more than one workunit at a time, I'm not sure that will work for you as your gpu only has 1gb of memory on it, again that is something you will have to test.
firebyrdman wrote:Does any
)
I'm unclear what you really mean by "balance".
For CPU tasks, your 6 core CPU will be a little slow because there are only 3 floating point units (FPUs), each one being 'shared' between 2 cores. For GPU tasks, the performance will be quite poor because the GPU you have is quite unsuited to the crunching task. Overclocking either of these would have little potential for improvement and a big potential for unstable operation.
With 1GB video RAM, you can only run 1 GPU task, period. If you had the ability to run 2 concurrent GPU tasks, you would likely see a nice improvement in output. You would need at least a 2GB card to do that.
With your current hardware, you may be able to improve the GPU crunch time (currently around 10hrs per task) by limiting (through your preferences) the number of CPU tasks crunching simultaneously. I presume you always have a GPU task running. How many CPU tasks run concurrently? Under default conditions there should be 5. If (with preference changes) you were to run only 3, you may find that CPU tasks run faster because of less FPU contention and the GPU task might also run a bit faster (better servicing from CPU cores) so that overall there might be more efficient use of the hardware. You will only know by doing the experiment. However the gains (if any) will be modest at best.
If you want to make a significant improvement in output, your best option would be to consider changing the GPU to something more suited to crunching. There are options available for around $US100 that would take you from your current RAC of around 8K to the vicinity of 200K - a >20-fold increase. You would just need to check the suitability of your current power supply. Your current GPU draws little power. If your PSU was a modern, efficient, good brand name unit with at least 300 watts available @12V (ie. at least 25 amps) you should be able to swap in a new card with no other changes needed. There are plenty of cards that don't need an external 6-pin power connector if your PSU doesn't have one. If you supply the brand/model of your PSU we could soon advise you about it.
Cheers,
Gary.