The announcement and launch of Nvidia Volta GPU is still a while off, but leaks and rumors regarding the next generation graphics architecture keep popping up now and then. According to the latest leak, Nvidia Volta GV100 GPU has appeared in the AIDA64 database, which sort of confirms that initial Volta samples have already been created.
Next-Gen Nvidia Volta GV100 Teased – Multi-TFLOPs GPU With Fastest HBM2
Developers found a PCI device ID in the latest AIDA update while they were searching for references to Nvidia GP102-based GTX 1080 Ti which is expected to launch next month. The PCI device ID belongs to 1D81 = Graphics Device [GV100], the company’s new flagship GPU code-named Volta.
GV100 is the full-fat Volta GPU that will most likely replace the Pascal GP100, or GP102. The specifications of the GPU are not confirmed yet but a previous analysis down by NextPlatform reveals that it will power the next generation supercomputers known as Summit and Sierra.
In that case, it will be an insanely fast chip capable of delivering multi-TFLOPs with the fastest HBM2 configuration. In general, Volta is rumored to have twice of everything that Pascal features, ranging from memory capacity, compute to power efficiency and bandwidth.
As per Nvidia’s GPU roadmaps, GV100 GPUs will deliver SGEMM (Single precision floating General Matrix Multiply) of 72 GFLOPS/Watt compared to 42 GFLOPs/Watt on Pascal GP100. Based on these figures, a Volta GV100 GPU with a TDP of 300W can theoretically output 9.5 TFLOPs of double precision performance, which is almost 2x the current-gen GP100 GPU.
Furthermore, Volta GV100 could feature 32GB HBM2 capacity of Pascal GPUs coupled with a bandwidth of around 1 TB/s.
It was originally thought that Nvidia’s 2017 GeForce 20 series would be based on the Volta architecture. But, recent reports suggest that the next GPU series will in fact be refreshed Pascal while the GeForce 30 series will feature Volta.
Built using the smaller 12nm process from TSMC, Volta is set to replace the Pascal refresh in 2018, though the official unveiling is expected at this year’s GPU Technology Conference in May.