Thanks to Indiana University and also The Next Platform, we have a tip of what’s to find with NVIDIA’s future GPU strategies, with solid indications that NVIDIA will certainly have a brand-new Tesla accelerator (and also underlying GPU) on-line by this summertime.
In a post describing the setup of Indiana University’s Big Red 200 supercomputer– which additionally occurs to be the very first Cray Shasta supercomputer to be set up– The Next Platform reports that Indiana University has actually chosen to break up the implementation of the supercomputer in to 2 stages. In specific, the supercomputer was implied to be provided with Tesla V100 s; nonetheless the college has actually rather chosen to resist on distribution of their accelerators to ensure that they can rather have NVIDIA’s next-generation accelerators, which would certainly make them amongst the very first organizations to obtain the brand-new accelerators.
The discovery is remarkable as NVIDIA has yet to introduce any type of brand-new Tesla accelerators or matching GPUs. The business’s present Tesla V100 s, based upon the GV100 GPU, were initially introduced back at GTC 2017, so NVIDIA’s calculate accelerators schedule for a significant refresh. However it’s a little bit unexpected to see anybody aside from NVIDIA disclose any type of information concerning the repairs, provided exactly how buttoned-down the business typically has to do with such information.
At any type of price, according to Indiana University the team anticipates to have their brand-new accelerators set up by later on this summertime, with Big Red 200 running in CPU-only setting in the meantime. The Next Platform short article takes place to state that the more recent accelerators will certainly supply “70 percent to 75 percent much more efficiency” than NVIDIA’s present V100 accelerators, which thinking it’s precise, would certainly create a substantial generational upgrade in efficiency. Though as constantly, with numerous settings of calculate included– whatever from straight FP32 vector mathematics to tensor procedures to reduced accuracy procedures– the evil one remains in the information on where those efficiency gains would certainly most be understood.
In the meanwhile, NVIDIA’s following GTC occasion is arranged for mid-March So if NVIDIA is preparing to introduce a brand-new Tesla, after that I would definitely anticipate to see it there.