Jim Ison, VP of Sales at One Stop Systems, walks us through their monster PCIe expansion chassis that can hold up to 16 full size GPU cards. In order to do this, this product is truly an innovative design - packing this amount of processing power into only 3U worth of rack space. Watch the video for a guided tour through the product by Jim.
Video Transcript of Interview:
Gabriel Consulting: I could not walk by this without stopping to see what it is. You’re Jim Ison?
Jim: Jim Ison, Vice President of Sales for One Stop Systems.
Gabriel Consulting: What is it?
Jim: What we have is 16 GPUs in a single 3U enclosure that can go into any server rack that can handle 36” deep chassis. Connects to any server, any manufacturer that has x16 slots. We use a Gen 3 small host interface board goes into any short height systems, just about any server can handle that.
Gabriel Consulting: And that could handle one row of GPUs?
Jim: It can handle one to four rows of GPUs. 16 GPUs can run from one of these cards, two, three or four cards.
Gabriel Consulting: You can run it pretty ragged with 16 GPUs.
Jim: You can run it as ragged as you want. The nice thing about it is the link stays up with all four systems so you can share GPU resources across GPUs without going back to the host, or you can send all the data back to the host as you need it.
Gabriel Consulting: That’s a really interesting point. I’m going to be writing about this in other articles. GPUs can intercommunicate between each other without having to go out back to the CPU or main memory, and it takes out a lot of latency.
Jim: It takes out latency and also makes a direct path, so your GPU direct applications will operate just as fast without having to go through a network connection.
Gabriel Consulting: It’ll operate faster?
Gabriel Consulting: I mean they’re losing some hops there, by c’mon don’t be so shy, don’t be so modest. It’s better.
Jim: It is better.
Gabriel Consulting: It’s faster. So how long have you had this out?
Jim: The Gen3 products start shipping in April; this is a prototype model here. It has 9000 watts of power, 6000 watts plus a spare so we have full redundancy on the power supplies. It’s got a full IPMI system monitor. The system monitor does handle, SMNP, RCMP type monitoring applications. We can monitor the telemetry of the GPUs as well as temperatures and voltages inside the box.
Gabriel Consulting: Now this is Gen 3, did you have a Gen 2 product?
Gabriel Consulting: How did that do?
Jim: Gen 2 worked very well. The other nice thing about this is we can handle full length cards and everything. So if you see some of the other large systems, they candle ¾ length GPUs. Well GPUs are starting to grow so you have a lot of passive GPUs out there, most of the connections to the power is on the back, so you start to see where you plug the power into the back of the GPU but since it’s ¾ length, you start getting into problems with fit. Now this will support full length cards, plus power in the back for both GPUs front and back row. So there are two ranks of 8 GPUs, to give you the sixteen.
Gabriel Consulting: What’s a typical customer look like for this? What are they doing?
Jim: Anything in the medical imaging market, image processing. In the financial market, people are using GPUs. This will also work with FPGA cards, and NAND flash cards so we do storage with the system. The canisters are removable, so you can remove them like bank of hard drives when you get into them. And then the Oil and Gas industry is another big product market.
Gabriel Consulting: Oh, makes sense. Well thank you. Oh! wait, wait. What’s the ballpark range? What’s this go for?
Jim: If you’re talking quantity one, you’re in the fourteen thousand dollar range.
Gabriel Consulting: What if I bought two?
Jim: Then you’ve got $28,000.
Gabriel Consulting: Alright well thank you very much Jim. Good stuff.
Jim: Thank you.