NVIDIA Tesla V100 is the first solution designed under the next-generation graphic architecture, “Volta”, which will replace the current “Pascal” platform in all kinds of segments, from supercomputers to mobile devices, to Desk and laptops.
It was presented by NVIDIA at the GTC conference being held in Silicon Valley. Tesla V100 is aimed at the field of high performance computing (HPC) and is specialized in deep learning applications, although its base can be used in any segment of supercomputing, artificial intelligence, autonomous cars or cloud.
NVIDIA Tesla V100 is based on the Volta GV100 graphic core, a development made by TSMC foundry in new 12-nanometer technology processes and, as you would expect, is a real monster. Tesla V100 is the largest and most powerful ever manufactured by the GPU industry, with an impressive die size of 815 mm2 and a whopping 21 billion transistors. It is 30% higher than the GP100 it will replace.
The main specifications give an idea of the power of this development. Among other:
- 5,120 cores CuDA FP32.
- 320 units of texture.
- 80 SMs
- 40 TPCs
- Core Speed 1455 MHz
- Memory: 16 GB HBM2 from Samsung
- Memory interface: 4096 bits
- Memory Ratio: 900 GB / s
- 640 cores Tensor (deep learning)
- TDP: 300 Watt
- Power: Up to 15 / 7,5 TFLOPS (FP32 / FP64) – Tensor 120 TFLOPS
In addition to maximum power components, NVIDIA Tesla V100 will feature optimized Volta software, new APIs, improved Volta multiprocessing service or the second generation of NVLink. It will be available by adding several of these cards into supercomputers like Volta DGX-1, servers like HGX-1 or professional workstations like DGX Station.
Of course, the most interesting thing for the consumer is to come with the new GeForce solutions for personal computers based in Volta which is NVIDIA Tesla V100, with presentation scheduled for the third quarter of 2017.