NVIDIA introduced a variety of points at Supercomputing, such as an Arm web server referece layout. Despite the program being the significant center occasion for high-performance computing/supercomputers, it isn’t typically the place where NVIDIA introduces a GPU. Nonetheless we saw a brand-new version of NVIDIA’s high efficiency Tesla V100 at several cubicles at Supercomputing.
The brand-new GPU we saw was called the V100 S (or V100 s). Firstly, the name: I really did not understand it was new/unannounced up until it was explained to me. The means it was created on a few of the signboards resembles it is simply describing ‘several V100 systems’, yet a number of business verified to be that it is a brand-new item. For these suppliers, they were really informed prior to the program that NVIDIA was preparing to introduce it there, and also were amazed that the Chief Executive Officer Jensen Huang did not discuss it in his off-site 2 hr discussion to press and also companions.
Nonetheless, NVIDIA’s companions had actually published the signboards, developed the screens, developed the systems, and also had not been informed * not * to reveal it off. So they did. I was notified to keep an eye out for the gold shadow at one certain cubicle– they were setting apart by having the basic V100 with an environment-friendly shadow, and also their V100 S will certainly have a gold shadow. This is regardless of the gold shadow systems likewise simply state ‘V100’, which is indicated to symbolize the household of the card.
Finding out what is various concerning this card has really been a job– none of my normal calls appear to understand specific numbers, although a pair verified it was ‘faster memory’, describing the on bundle HBM2. I’m still checking out specific regularity modifications, and also probably the knock on results on TDP, yet as it stands ‘faster memory’ is the only details I have. There may likewise be a cost distinction for anybody thinking about these variations.
One idea is that NVIDIA may not really introduce the V100 S as a different version, yet simply a greater memory variation of the V100 and also consumers will certainly simply need to examine specifically what the memory regularity is when they acquire– much like various customer cards can have various memory rates. No- one was reviewing specific launch timing, yet it appeared NVIDIA’s companions were deep right into recognition, otherwise currently providing them to pick consumers.