Earlier this year NVIDIA revealed the first Volta-based cards for the datacenter market. The 12nm GV100 chip is close to the limits of what's physically possible, it's the largest GPU ever created and it has a huge die area of 815mm².
Larger chips are more expensive to manufacture than smaller chips. The larger a die is, the fewer chips you can produce from a 300mm silicon wafer. Additionally, larger chips are more prone to manufacturing defects so chip makers get hit by a double whammy that drives up costs if they design larger chips.
Given the Volta's huge die area I was wondering how high the manufacturing cost of this chip is. Thanks to yesterday's earnings call with financial analysts, we now know the Volta GV100 chip costs "close to $1,000" to manufacture:
Jen-Hsun Huang - NVIDIA Corp.
And so the price of Volta is driven by the fact that, of course, the manufacturing cost is quite extraordinary. These are expensive things to go and design. The manufacturing cost itself, you guys can estimate it, is probably in the several hundred dollars to close to $1,000. However, the software intensity of developing Volta, the architectural intensity of developing Volta, all of the software intensity associated with all the algorithms and optimizing all the algorithms of Volta is really where the value-add ultimately ends up. And so I guess the pricing – your question relates is pricing. We expect pricing to be quite favorable for Volta.
NVIDIA's GV100 is used by the Tesla V100 card. The company's DGX Station features four V100 cards and is priced at $69,000. DGX systems have shipped to over 300 unique customers, with more than 1,000 systems in the pipeline. NVIDIA says Facebook alone is using 128 DGX "AI supercomputer" systems. Consumers models of Volta are expected next year, those products will use smaller variants of Volta.