If you’d think the GPU on the new Titan X is powerful, wait till you see what Nvidia unveiled at this year’s Hot Chips conference. The green company showcased a die shot of its most powerful Pascal-based GPU called the GP100, which currently can only be found on Nvidia’s supercomputing monster: the Tesla P100 hyperscale accelerator.
Nvidia’s 16nm Pascal GP100 GPU is not only its most powerful GPU to date, but it’s also the biggest. Measuring in at a huge 610mm2, Nvidia’s HPC-oriented GPU is larger compared to the older Maxwell GM200, which had a die size of 601mm2. That’s not all, the transistor count on the GP100 GPU have also been increased to 15.3 billion – almost twice that of the Maxwell GM200 GPU (8 billion).
“The die shot of Nvidia’s Pascal-based GP100 GPU”
Of course, being a HPC (high performance computing) GPU means that the Pascal GP100 will be equipped with ridiculous amounts of compute performance. According to Nvidia, its most powerful Pascal GPU comes with an FP32 (single-precision floating point) performance of around 12 TFLOPs, while FP64 (double-precision floating point) performance stands in at 5.5 TFLOPs.
The Pascal GP100 GPU comes in two VRAM capacities: 16GB and 32GB of HBM2 memory and it is capable of tear-jerking bandwidth speeds of up to 1TBps. That said, due to its overwhelming performance, the maximum TDP of the GP100 GPU was increased to 300W, which is 50W more compared to the older Maxwell GM200 and Kepler GK210 GPUs.
“Behold the most powerful GPU on the planet”
Unfortunately, the Pascal GP100 GPU would probably never be used for consumer-grade graphics cards, simply because there is no need for such insane performance in the mainstream market, yet. However, if one simply cannot resist the temptation of owning the best of the best, the GP100 can be bought with Nvidia’s DGX-1 supercomputer, which comes with a modest price tag of US$129,000 (about RM519,437).