Message boards :
Number crunching :
Wondering about getting an nVidia Tesla K80 for MilkyWay.
Message board moderation
Author | Message |
---|---|
![]() Send message Joined: 18 Dec 20 Posts: 2 Credit: 7,944,564 RAC: 0 ![]() ![]() |
Hey everyone, this is my first post so I apologize if I am not in the right place. I was considering getting an nVidia Tesla K80 for my garage rig that computes for MilkyWay. Since the prices on these cards are quite cheap at about $135 USD, I would prefer to get this card over something more expensive like a GTX 1060. I am wondering if the performance of this Tesla will be comparable to that of the GeForce GTX 1060 3GB I have in my gaming rig. Will it be worth the cost, and will it work with the project? Thank you! |
Send message Joined: 7 May 10 Posts: 6 Credit: 67,026,308 RAC: 0 ![]() ![]() |
you will need to find proper cooling for it - since its passive for pull-push config. (or some fan mount or something special.) Now thats out the way, fp64 k80 is 20x faster than 1060, and 4x as fast as Titan V. I say go ahead and try - but note it may not work, as card is missing all kinds of codecs and video encoders. I assume someone from milky staff should state.. |
![]() Send message Joined: 18 Dec 20 Posts: 2 Credit: 7,944,564 RAC: 0 ![]() ![]() |
Yes, there are some 3D printed fan shoulds I can use to mount a fan onto the card for around $20. This seems like something worth trying though. |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 14 ![]() ![]() ![]() |
Yes, there are some 3D printed fan shoulds I can use to mount a fan onto the card for around $20. This seems like something worth trying though. As long as you can get drivers that have OpenCL in them and are new enough, although MilkyWay accepts some old drivers just fine, the card should do well. If you can't find one with OpenCL drivers try the Project SRBase and the TF workunits, I had an older gpu that just would not load the OpenCL drivers and it crunched just fine on those tasks. I have since switched the gpu to a different machine and the OpenCL drivers loaded right up...no clue but it's all good now. |
Send message Joined: 9 May 10 Posts: 4 Credit: 103,752,795 RAC: 0 ![]() ![]() |
I don't know if it helps, but I did something similar recently with an Nvidia K20M which I got 2nd hand on Ebay. It is installed in my "winter heating" PC running Ubuntu 20.x. It processes Separation v1.46 WUs in about 50s. Linux was able to identify the drivers it needed. This is computer ID 819072. you will need to find proper cooling for it - since its passive for pull-push config. (or some fan mount or something special.) This is absolutely true for the K20M. I initially bought it to try fitting it in to an HP DL380 G7 server, but it wouldn't post with the card in. I tried it in my desktop PC, but it overheated in about 30 seconds. The good thing is that if this happens, it just switches itself off. The CPU and other GPU in that PC are watercooled so I added the Tesla K20M in to the loop using a cheap waterblock and using a hacksaw to modify the thin aluminium heatsink that covers other parts of it. I'm not sure how good the contact is between this heatsink and the chips, since the previous thermal paste was more of a thick putty that I replaced with Arctic silver which doesn't have the same gap filling properties... Also, if it overheats, the card will not be able to be addressed and all the MW tasks will fail. The Tesla disassembled quite easily. The K20 used Torx bits. Another point to bear in mind is the physical size. I assume the K80 is similar to the K20, but they are long cards. It would be worth checking the physical dimensions against your case. If you have a route to solving the cooling issue and it will all fit then I say give it a go! PS. I'm only slightly jealous that you can find a K80 for such a good price. They are a lot more than that here. |
©2024 Astroinformatics Group