All, Do we have any interest in adding the K40 GPUs from IIT/SCC'14 back into JLSE? I'm sure there will be some amount of paperwork that will be needed to loan hardware for IIT. Thanks, Ben

Begin forwarded message:

Date: January 20, 2015 at 2:26:57 PM CST From: Ioan Raicu <[email protected]> Reply-To: <[email protected]> To: "Allen, Benjamin S." <[email protected]> Cc: "Scullin, William R." <[email protected]> Subject: Re: Follow up on SC14 SCC from NVIDIA

Hi Ben, Since we will be trying for SCC again this year, our undergrads could be using them to get more familiar with GPUs, writing CUDA code, etc, in an environment most similar to what they will likely have again. We have machines at IIT to put them in, but they are 6 or 8 core nodes, 16GB of RAM, and 1Gb/sec Ethernet, nothing compared to the 36-core, 128GB RAM, and 56Gb/sec IB the SCC nodes have. My understanding is that the SCC students (there are 6 of them at this point) will get JLSE accounts and will do much of their testing and tinkering on JLSE until new hardware would arrive in August/September. So, at least our UG students could use these K40 accelerators. But again, if its too much trouble, and paperwork, etc, we will bring them online at IIT. Thanks for looking into this. Ioan

On 1/20/2015 2:09 PM, Allen, Benjamin S. wrote:

...

Hi Ioan,

Let me ask around with JLSE folks. Generally our users aren't that interested in GPUs, but we'll see.

Thanks so much for the offer, and I'll let you know hopefully be the end of this week.

Ben

...

On Jan 20, 2015, at 12:41 PM, Ioan Raicu <[email protected]> wrote:

Hi William, Ben, We have these 8 NVIDIA cards that NVIDIA has graciously allowed us to keep. If you think we could load them up in the old 8-node SCC cluster, I'd be happy to loan them to you/ANL. If its too much trouble, then I'll load them up in a 20-node cluster we have here at IIT. Let me know what you decide. Cheers, Ioan

-------- Original Message -------- Subject: RE: Follow up on SC14 SCC from NVIDIA Date: Tue, 20 Jan 2015 18:04:26 +0000 From: Mark Ebersole <[email protected]> To: [email protected] <[email protected]>

“Provide us with more than 8 GPUs, perhaps 12, or even 16. Some of the applications were not driving the GPUs that hard, and we could have fit more than 8 GPUs in the power budget” – hardest part about this is budget. With the majority of teams using GPUs for every cluster competition throughout the year, it’s hard to give more than 8 per team! Teams are welcome to ask other vendors (like their server vendor) for more GPUs in addition to what NVIDIA gives. “Have other GPUs available, such as the Titans, etc. Some of the other GPUs compared to the Tesla's, might make the competition more interesting.” – Teams are welcome to ask us for different GPUs J I try not to dictate what should be used, just provide the necessary information for teams to make their own decisions. “Who do I need to talk to in order to return the K40s that were sent to us for the competition?” – If you’re going to use them, Keep them!! From: Ioan Raicu [mailto:[email protected]] Sent: Wednesday, January 14, 2015 8:08 PM To: Mark Ebersole Subject: Re: Follow up on SC14 SCC from NVIDIA Hi Mark, Sorry your email got buried under many other emails. I am just now starting to recover from SC14 and the holidays.

See my answers below inline: On 11/25/2014 11:11 AM, Mark Ebersole wrote: [BCC advisors for teams using GPUs] All, Hopefully everyone made it back home safely from New Orleans! While I was stuck in our booth most of the time, I enjoyed the interactions I did get with the teams. I wanted to get in touch while the competition is still fresh in everyone’s minds. Would you mind taking some time to help answer the following questions so I can try to help improve NVIDIA’s involvement in future competitions? It would be great if you could ask some or all of these questions to your team members as well. What can NVIDIA do better for the next competition? Provide us with more than 8 GPUs, perhaps 12, or even 16. Some of the applications were not driving the GPUs that hard, and we could have fit more than 8 GPUs in the power budget.

What could the committee change to make the competition more fair/interesting/relevant/etc.? Have other GPUs available, such as the Titans, etc. Some of the other GPUs compared to the Tesla's, might make the competition more interesting.

What did your team enjoy most about the competition? We learned a whole lot, very practical things. We will try again next year, and hopefully we will do that much more next time!

What worked well? Everything went smoothly, once we got our hardware!

Do you plan on using GPUs (probably K80s) in your next competition? Yes, definitly!

Anything else? Who do I need to talk to in order to return the K40s that were sent to us for the competition? Since K80 GPUs are now available, perhaps the K40 GPUs are not as valuable to NVIDIA. Is there any way we could keep the GPUs on loan for longer? At IIT, we are both a CUDA Research and Teaching center, and we have built a cluster of 20 nodes with GPUs (mostly donated from NVIDIA) so that we can teach CUDA in the classroom. At the moment, we have 1 K40, 1 K20, and a bunch of low end GPUs with 384-cores each, that have no double precision performance. If we could keep the 8 K40 GPUs we had on loan, for an extended loan, we could get half the cluster to house K20/K40 GPUs, and get more students to use the K40s instead of the low end gaming GPUs. If you are not the right person to be asking about this, perhaps you can put us in touch with who is? If you don't know, I'll track back through my emails and figure out who sent me the GPUs in the first place.

Cheers, Ioan

Thanks! P.S. If you have any team members looking for internships/jobs at NVIDIA, please have them send me their resume. Mark Ebersole CUDA Educator | NVIDIA Corporation [email protected] Click to learn: <Mail Attachment.png> <Mail Attachment.jpeg> This email message is for the sole use of the intended recipient(s) and may contain confidential information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.

-- ================================================================= Ioan Raicu, Ph.D. Assistant Professor, Illinois Institute of Technology (IIT) Guest Research Faculty, Argonne National Laboratory (ANL) ================================================================= Data-Intensive Distributed Systems Laboratory, CS/IIT Distributed Systems Laboratory, MCS/ANL ================================================================= Editor: IEEE TCC, Springer Cluster, Springer JoCCASA Chair: IEEE/ACM MTAGS, ACM ScienceCloud ================================================================= Cel: 1-847-722-0876 Office: 1-312-567-5704 Email: [email protected] Web: http://www.cs.iit.edu/~iraicu/ Web: http://datasys.cs.iit.edu/ LinkedIn: http://www.linkedin.com/in/ioanraicu Google: http://scholar.google.com/citations?user=jE73HYAAAAAJ ================================================================= =================================================================

-- ================================================================= Ioan Raicu, Ph.D. Assistant Professor, Illinois Institute of Technology (IIT) Guest Research Faculty, Argonne National Laboratory (ANL) ================================================================= Data-Intensive Distributed Systems Laboratory, CS/IIT Distributed Systems Laboratory, MCS/ANL ================================================================= Editor: IEEE TCC, Springer Cluster, Springer JoCCASA Chair: IEEE/ACM MTAGS, ACM ScienceCloud ================================================================= Cel: 1-847-722-0876 Office: 1-312-567-5704 Email: [email protected] Web: http://www.cs.iit.edu/~iraicu/ Web: http://datasys.cs.iit.edu/ LinkedIn: http://www.linkedin.com/in/ioanraicu Google: http://scholar.google.com/citations?user=jE73HYAAAAAJ ================================================================= =================================================================