A basic nvidia question

[acnash]

Dear all,

I had been running F@H on a Linux Mint 32 core Intel Zeon 3.10 machine with no GPU. My machine is shared between my own scientific work (minor protein folding and medical statistics; the more intense free energy calculation I do on a cluster at work) and the F@H client. The total estimated points per day would fluctuate between 100,000 and 200,000.

Yesterday, however, I was gifted three nvidia GTX 970 card! This is the first time since the days of HalfLife 2 and Doom 3 that I've played around with the graphics cards themselves rather than just using the ones at work handled by IT. So please excuse how simple my questions are.

I plugged one in, spent five hours fiddling around with openCL and CUDA and eventually it worked by accident. The points per day have shot up to 560,000. Interestingly, it took five minutes to see a sudden increase in points once I started folding on the GPU. I suspect, but I could be wrong, that this delay was down to the initial calculation of domain decomposition for some of the non-bonded interactions of whatever I was folding.

Firstly, as I have room on my motherboard, can I add a further GPU card? Secondly, as all three cards are identical and I have already installed the drivers and libraries required for the first card, will the second just work immediately? Finally, If yes to above, will the Folding@Home software display a second GPU card?

Thanks
Anthony

[JimboPalmer]

The F@H client software finds GPUs much better on install than on startup. If you have any issues, I would uninstall, then re-install the F@H software.
Assuming you really have 3 16 lane PCI-E slots in your computer, and at least 3 CPU threads to keep them fed with data, yes it should work.
Where previously you might have had 8 CPUs doing F@H, now 3 will feed the GPUs, and 5 can still crunch data. This will be a huge gain.

[kiore]

3 GTX 970's in one system will get pretty hot unless you have lots of space between them, I would add one at a time and monitor the temps closely.

[acnash]

Thank you both for those suggestions and advice. In terms of heat, I hadn't considered that!

Thanks again.

[ProDigit]

Due to the bitcoin crash, now's the time to upgrade graphics cards y'all!
Several GTX 1050 cards for under $75 (some as low as $55 on evil bay).
I recommend anyone to go online now and buy server hardware!
Seems like Qualcomm's Centrix processors are hitting the market, and Facebook and Google are updating their servers to these chips; leaving an excess of Xeon processors that used to be sold for $1200+ in 2014, and are sold for $580 new now, are sold for $50-75 on ebay.
I bought a $500 server that would have costed me in excess of $4k 2 years ago, all thanks to big companies dumping their old (and still good) hardware, as well as the bitcoin crash.

Grab em while you can!!

[Sven]

And don't forget, these graphic cards need a lot of electric energy => The power supply must be up to the task.

[bruce]

Thank you both for those suggestions and advice. In terms of heat, I hadn't considered that!

That's an important aspect (along with the power supply) that's often overlooked when systems are upgraded. When you buy a computer designed by a reputable computer manufacturer, they've spent a fair amount of time and money doing thermal design and testing. Even a laptop, which can't accept real upgrades are SUPPOSED to be designed to run at 100% CPU and 100% GPU (with 100% fan) and they can easily accumulate enough internal dust to start failing.