 |
 |
It's an interesting move but not unexpected - I noticed recently that NVIDIA new ToS forbid "Datacenter" deployment of their driver. Perhaps covering themselves for when we all start frying them. That's great to hear that Volta acceleration might be coming. It could really speed things up! Charles __ Charles Bowman Bioinformatics Analyst Ward Group The Scripps Research Institute [log in to unmask] ________________________________ From: Collaborative Computational Project in Electron cryo-Microscopy <[log in to unmask]> on behalf of Erik Lindahl <[log in to unmask]> Sent: Saturday, September 1, 2018 10:38 AM To: [log in to unmask] Subject: Re: [ccpem] RELION 3 Volta Benchmarks Indeed - I would recommend staying away from them for now. NVIDIA decided to overclock their cards too this time, which seems to have required a different cooling solution. There seems to be a couple of vendors that will produce "blower" cards (likely with slightly lower clock), and those will hopefully work fine with in servers or with 4 cards in a desktop, but it will likely be another month or two before they are out. We'll also be looking at moving a few more things to the GPU, so don't give up hope of Volta yet - we hope they'll be quite a bit faster in the future - and there might be some interesting opportunities with the raytracing units... Cheers, Erik On Sat, Sep 1, 2018 at 7:33 PM Yehuda Halfon <[log in to unmask]<mailto:[log in to unmask]>> wrote: I think the new founder edition design will be bad also for the workstation setup. I think that the 4 RTX in one case will be a problem. -------- Original Message -------- From: Charles Bowman <[log in to unmask]<mailto:[log in to unmask]>> Sent: Saturday, September 1, 2018 08:26 PM To: [log in to unmask]<mailto:[log in to unmask]> Subject: Re: [ccpem] RELION 3 Volta Benchmarks Craig - Yes they aren't hitting those bounds forward that we saw in the past. The biggest improvement in the P100 architecture was HBM2 memory, and now in Volta we have the tensor cores, which are great for neural network training but don't really help us out so far in cryoEM. I'm looking forward to the RTX line - they offer a refresh of the consumer cards with TFLOPS ratings near Tesla speeds in single precision. I am looking to get my hands on one to check out the cooler though - it seems like they may have eliminated the pass-through cooling and gone with the dual-blower design, which could cause problems for people who put them in servers. Charlie __ Charles Bowman Bioinformatics Analyst Ward Group The Scripps Research Institute [log in to unmask]<mailto:[log in to unmask]> ________________________________ From: Craig Yoshioka <[log in to unmask]<mailto:[log in to unmask]>> Sent: Friday, August 31, 2018 2:04 PM To: Charles Bowman Cc: [log in to unmask]<mailto:[log in to unmask]> Subject: Re: [ccpem] RELION 3 Volta Benchmarks Thanks Charles, A little saddened by the Volta numbers honestly! Looks like the 1080 or 1080Ti are still value king as long as your model/data fits in GPU mem. I'll test our 4 and 8 GPU 1080Ti configs you can throw in. Cheers, -Craig > On Aug 31, 2018, at 1:33 PM, Charles Bowman <[log in to unmask]<mailto:[log in to unmask]>> wrote: > > Hello Everyone, > > As promised earlier, you can now find my benchmark information comparing the currently available NVIDIA Tesla cards on the Ward lab website. > > https://ward.scripps.edu/benchmarks > > The data is split into two parts - benchmarks meant to compare GPU speeds, and benchmarks meant to push the hardware to its limits. > > I welcome any contributions or additions! I plan to leave this page up as a growing resource. > > Thanks, > > Charles > > __ > Charles Bowman > Bioinformatics Analyst > The Ward Group > The Scripps Research Institute > > ######################################################################## > > To unsubscribe from the CCPEM list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1 ________________________________ To unsubscribe from the CCPEM list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1 -- Erik Lindahl <[log in to unmask]<mailto:[log in to unmask]>> Professor of Biophysics, Dept. Biochemistry & Biophysics, Stockholm University Science for Life Laboratory, Box 1031, 17121 Solna, Sweden ________________________________ To unsubscribe from the CCPEM list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1 ######################################################################## To unsubscribe from the CCPEM list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1