NVIDIA’s Tesla V100 is an absolute powerhouse of a GPU, made most useful when used for deep learning and artificial intelligence. In case it somehow isn’t obvious by now, AI and deep learning are areas NVIDIA has put a ton of focus on in recent years (the company even held a deep learning award show at GTC 2017), so it’s no surprise that its first Volta-based chip is targeted at that audience.
As we learned at GTC 2017 a few months ago, the Tesla V100 offers 7.5 TFLOPS of double-precision performance, 15 TFLOPS of single-precision, and a dedicated Tensor Core offering 120 TFLOPS of compute power for deep learning workloads.
NVIDIA hasn’t announced the V100 SRP, but we can assume it costs at least $8,000 USD based on the fact that the top-end Quadro GP100 sits around the same price-point. So you can imagine the extreme joy some researchers must have felt on Saturday as NVIDIA CEO and co-founder Jensen Huang presented 15 of these beauts to extremely lucky research groups. Just look at their joy!
NVIDIA didn’t randomly hand these V100s out, but rather made sure that at least 15 different research institutions received one. Each of these were signed by Jensen in gold pen, with the simple message, “Do Great AI!” That shouldn’t prove a problem with 120 TFLOPS of compute power under-the-hood. A total rarity, Jensen wasn’t sporting a leather jacket at this event, presumably because the it (Computer Vision and Pattern Recognition conference) took place in Honolulu, Hawaii. Instead, he wore a short sleeve dress shirt, white jeans, and vans, calling it his “aloha uniform”.
Once these researchers get the V100s installed into their machines, they’ll be able to tie their apps into CUDA, cuDNN, and TensorRT to accelerate their research. There’s still no word on when V100 will become readily available, but with 15 being freed at this event, it gives the impression that the long-awaited availability is coming soon. If only we could say the same about Volta for gaming!