Nvidia have simply unveiled the primary official Volta GPU, the professional-level Tesla V100, and if you need one you’re going to must fork out $3bn. That’s how a lot the R&D prices have been and there aren’t too many out there proper now…
Can’t afford $3bn? Well, right here’s our choose of the best graphics cards a traditional checking account would possibly be capable to afford.
The Tesla V100 chip makes use of TSMC’s tiny 12nm manufacturing course of, however it’s nonetheless a completely monumental beast of a GPU. It’s an 815mm2 chip, round twice the scale of the 471mm2 GP102 of the GTX 1080 Ti and Pascal Titan X playing cards. Inside which are 21 billion transistors and 5,120 CUDA cores.
Nvidia CEO, Jen-Hsun Huang, introduced the GPU on stage calling it a masterpiece. I imply he would, wouldn’t he? But it certain is one hell of an engineering feat and has been created “at the limits of photolithography,” explains Huang, “which means you’ll be able to’t make a chip any larger than this as a result of the transistors would fall on the bottom.
“To make one chip work per 12-inch wafer, I might characterise it as ‘unlikely.’ So the truth that that is manufacturable is an unimaginable feat.”
As nicely because the GPU itself Nvidia have packed it out with 16GB of second-gen high-bandwidth reminiscence (HBM2) able to delivering 900GB/s in reminiscence bandwidth. That’s a powerful increase over the Pascal-based P100 GPU which nonetheless hit a staggering 732GB/s.
They’ve additionally practically doubled the bandwidth over the primary era of their NVLink interconnect, hitting 300GB/s in contrast with 160GB/s for last-gen NVLink. That’s round ten occasions the bandwidth out there to the most recent PCI Express connections. Like with P100 Nvidia may also be transport a PCIe-based model too.
Now clearly $3bn is a bit wealthy, so there are going to be cheaper variations than Jen-Hsun’s on-stage prototype out there later within the yr when quantity of those large GPUs has been elevated. There’s a server field sporting eight V100 GPUs costing simply $149,000, and one with 4 at round half that. Those will begin transport in Q3 of this yr with full OEM availability of the V100-powered machines on sale in This fall.
So, what does that imply for us? Well, it means the pro-level Volta V100 GPU is unlikely to primarily be the identical chip that drops into the GTX 2080 Ti, not at that dimension. And if Nvidia observe the identical launch cadence they adopted with Pascal we received’t be seeing Volta GPUs in our PCs till May subsequent yr.
But that’s if Nvidia observe the identical path, and there’s nothing to say they’re going to try this. We should still see little Volta’s this yr. Maybe.
Source
