Wednesday, February 17th 2016
NVIDIA GP100 Silicon to Feature 4 TFLOPs DPFP Performance
NVIDIA's upcoming flagship GPU based on its next-generation "Pascal" architecture, codenamed GP100, is shaping up to be a number-crunching monster. According to a leaked slide by an NVIDIA research fellow, the company is designing the chip to serve up double-precision floating-point (DPFP) performance as high as 4 TFLOP/s, a 3-fold increase from the 1.31 TFLOP/s offered by the Tesla K20, based on the "Kepler" GK110 silicon.
The same slide also reveals single-precision floating-point (SPFP) performance to be as high as 12 TFLOP/s, four times that of the GK110, and nearly double that of the GM200. The slide also appears to settle the speculation on whether GP100 will use stacked HBM2 memory, or GDDR5X. Given the 1 TB/s memory bandwidth mentioned on the slide, we're inclined to hand it to stacked HBM2.
Source:
3DCenter.org
The same slide also reveals single-precision floating-point (SPFP) performance to be as high as 12 TFLOP/s, four times that of the GK110, and nearly double that of the GM200. The slide also appears to settle the speculation on whether GP100 will use stacked HBM2 memory, or GDDR5X. Given the 1 TB/s memory bandwidth mentioned on the slide, we're inclined to hand it to stacked HBM2.
32 Comments on NVIDIA GP100 Silicon to Feature 4 TFLOPs DPFP Performance
That doesn't seem to make a lot of sense TBH. A 128-bit bus width GDDR5X card (using 70% of a conventional GDDR5 power envelope) running at 13000Gbps (the transfer rate currently being achieved) gives 208 GB/sec of bandwidth - that exceeds the 256-bit R9 380X and almost approaches the GTX 980. More than enough bandwidth for a 128-bit card. Double the bus width to 256, and the bandwidth jumps to 416GB/sec - comfortably higher than any current Nvidia card and every AMD card except the Fiji cards. Hardly a crisis situation.
Waiting for GP104 variant to come out as that's what I'll be getting.