Seems like Nvidia has taken the GPU challenge to a whole new level as they announced the biggest ever graphics chip named ‘Volta GV100 GPU‘. Based on the monolithic GPU design, Nvidia has made a damn impressive piece of technology which is aimed at the compute intensive market.
Technology Getting Compact
Nvidia has a ton of brilliant ideas in the pipeline which includes making GPUs more compact hence fitting several GPUs and several stacked DRAM dies on a single package. This news is a blessing for those looking for the newer speed challenges and making things possible. Currently, Nvidia is playing with 2 of the fastest GPU accelerators:
- Tesla P100 based on Pascal architecture
- Tesla V100 based on Volta architecture
Biggest Chips of All Times
Looking at the size of Tesla P100 and V100, one thing is for sure that their size is touching the limits and it can’t any bigger than this. The die size of Tesla P100 is 610mm2 to be exact while the die size of Volta V100 GPU is 815mm2 which is 33.1% larger than the former one.
NVIDIA’s CEO Jen-Hsun Huang is pretty confident about the claim of biggest ever chips of al time as he expressed his thoughts like:
The part that is really shocking is this is rectile limits. Rectile limits basically means that is at the limit of photo lithography meaning you can’t make a chip any bigger.
Volta Die Size Increasing
Looking at the past journey of Volta, it seems like the die size is on an ever increasing role and especially this year, it increased vastly over Pascal. The dedicated FP32 and FP64 cores of Volta and Pascal make them superior to those of AMD’s.
NVIDIA’s MCM GPU Package
The MCM(Multi-Chip-Module) seems to be the solution proposed by Nvidia to the increasing die size with the addition of cores. A concept design is featured in a research publication which has been posted by NVIDIA. The Multi-Chip-Module is simply a combination of several chips (GPU/CPU/Memory/Controllers) on the same chip interposer and is interconnected via fast I/O lanes.