2x NVIDIA 1060
27 Mar 2018 7:25:05 UTC
Topic 214219
(moderation:
Hi all. I have system with two Nvidia 1060 GPU and AMD Ryzen CPU.
In BOINC settings allowed use GPU without any restrictions, but all jobs i receive computes on CPU only.
Why? is it ok or i do something wrong?
Hi Vitaliy, currently your
)
Hi Vitaliy, currently your systems thinks that your 1060 doesn't support openCl. So please go to Nvidia driver download (not Microsoft driver download), select your Geforce 1060 64 driver, download, install advanced, make sure to select New install during installation procedure. Come back here if problems persist. :-)
thx, it really works) and
)
thx, it really works) and work fast! 1 gamma ray job - less 20 min!
I would add just a couple
)
I would add just a couple things. When a new driver is released, download it, go to your boinc manager in advanced view, select activity. Stop the CPU thus stopping all processing. Go to your computers device manager, click on one of your GPU's inside display adapters, select driver, select uninstall device and select delete driver to free up space on your memory disk, let the computer do this step then restart your computer and it should reinstall the generic driver. Once you only have the generic driver left, install the newest driver for your Nvidia, restart the computer and begin CPU again and you have eliminated all unnecessary drivers, freed up space and have only the best driver in your memory (C:). I only know for sure this works for NVIDIA and EVGA based GPU's.
Work runs fine on Bosons reacted into Fermions,
Sunny regards,
earthbilly
vitaliy.mikhailenko
)
If you run 2 at a time on a 1060 your run time will increase to 30 min but you get 4 done in an hour instead of the 3 you are currently getting.
Be sure you add extra CPU %
)
Be sure you add extra CPU % to your computing preferences so you have enough CPU for the additional GPU. I find that if I run just half of the processors and add the extra processor % to equal the processor for each GPU task, everything speeds up so much the graph line for HOST TOTAL actually gets steeper than running more processors.
Key here is we see how much processor is needed for each GPU and add that much in the preferences.
Example for an 8 processor, 4 core, like an intel core i7-4790. If the GPU task needs .45 Processor for it's task and you have two GPU's, And want to run four processors so as not to overload your bus,
.45 X 2 = .9 processors. + 4 processors each running tasks = 4.9 processors needed. 4.95 / 8 = 61.875% is put into your processor window in computing preferences. See there I added just a little more than needed but not enough to excite a new task to a single processor.
And for heavens sake, do not delete the MICROSOFT generic driver when deleting the old NVIDIA drivers.
Sunny regards,
earthbilly
Work runs fine on Bosons reacted into Fermions,
Sunny regards,
earthbilly
earth_4 wrote:I would add
)
One thing to be wary of with new drivers, not all of them work for crunching, they are being sold and updated mostly for gaming and in the past several new software versions have made crunching slower or full of errors.
I have Nvidia 391.01
)
I have Nvidia 391.01 installed and it seems to work fine. What do I look for to know if a new driver version has made crunching slower or full of errors???
What do you consider the latest best Nvidia driver to use???
TIA
Rig-9 wrote:... What do I
)
Just observe the average crunch time for a group of tasks, 'before' and 'after'. The trick is to use a big enough sample of tasks and also to be aware of the 'natural' crunch time variations that occur for a given series of tasks. Archae86 has posted this thread which will explain what I mean by 'natural'. The crunch time increase with frequency seems to be much the same for the different series of tasks that depend on different data files. Since the crunch times seem to plateau at frequencies above 1000Hz, you could use tasks above that frequency for the comparison of the two averages. A group of 20 - 30 tasks or so, should be enough to establish a reasonably accurate average for crunch time in each case.
If you do some experiments and make any discoveries, please post your results as a 'community service' just as Archae86 did.
With regard to errors, you will know pretty quickly just by observing whether or not the completed tasks validate. As a rule of thumb, you should expect validation failures not to exceed around ~1%, if all is well. There is always going to be the odd failure even if everything is running perfectly, since the validator does enforce fairly strict tolerances.
How long is a piece of string???
My impression is that drivers evolve in order to tweak the performance of the latest hardware that a particular company wants to sell the most volume of. It has happened before that in extracting the last bit of performance out of the latest hardware, there can be regressions for older hardware. Most likely, the changes will be fairly small anyway, so unless there is a known problem or other compelling reason to change, you are probably best to stay with what has already been working well, particularly if your hardware is not the very latest high end model.
It can be worthwhile testing drivers over a period if there has been a change in hardware architecture and you have decided to become an early adopter :-). If there are any issues with new stuff, there's going to be a lot of comment that will be easy to find with a quick search.
Cheers,
Gary.
I had an experience with an
)
I had an experience with an MSI branded GTX980. I had to go into Invidia control panel and select Manage 3D settings and select (use for compute). Then restart the computer. This is the only GPU I have had experience with that required special selection like this and it bridged several driver updates. In fact I always had to select it if I used it in any platform. Just this one. There could be many more. I am no expert across many GPU's.
Work runs fine on Bosons reacted into Fermions,
Sunny regards,
earthbilly
I switched my MSI X99 RAIDER
)
I switched my MSI X99 RAIDER MOBO with intel i7-6800k running at 3.7Ghz with 32GB Crucial ballistic sport RAM stock speed over to Einstein with 2- founders edition 1060Ti GPU's and run 4 processor tasks and 1 processor per GPU and my daily host total increases by 685,000+ per day.
With the SLI capable MOBO I have no problems running both GPU's without one dropping out to a fifth processor task. My computer options is set to 52%. I have a 20" floor fan set to low pointed into the case with the side cover off. The CPU has Corsair water cooling with dual fan radiator. CPU runs at 43C and the GPU's run at 52C
COOL!
Work runs fine on Bosons reacted into Fermions,
Sunny regards,
earthbilly