Recent content by HKS

For the 20-series, 2080 was released in September 2018, and the "2080 Super" refresh was released in July 2019. A 30-series refresh summer 2021 with some SKUs produced at TSMC sounds realistic to me.

My guess is that AMD isn't using the Tensor Cores.

https://videocardz.com/76953/nvidia-geforce-gtx-2080-1180-pcb-has-been-leaked Apparently, photos of the new PCB has been leaked. Looks like NVLink will be introduced at consumer products...

I guess they are. On the new Xavier SoC, they have a hardware-block dedicated to "machine learning" called NVIDIA Deep Learning Accelerator (NVDLA). They have also open sourced this implementation. http://nvdla.org/

And we also confirmed that GV100 support fast FP16 math, even though it is currently only exposed in CUDA.

Yes, I work in a University/Lab that among other thing work on machine learning, and we have already ordered several cards. For us, this is not ridiculous, I had personally expected it to be a "Quadro GV100" before GV100 was released as a Titan card. And for the performance this card will give...

I you want more details on using the tensor cores for general compute, check out Mark Harris CUDA 9 and beyond presentation from GTC: http://on-demand.gputechconf.com/gtc/2017/presentation/s7132-mark-harris-new-cuda-features-and-beyond.pdf

Plus ECC on the memory and features such as GPUDirect RDMA and GPUDirect for Video.

Tegra X2 (with a Pascal-based GPU) also have "fast" FP16.

That understanding is not correct... GTX 1080 cards starts at 5600 SEK. GTX 1080 Ti starts at around 7600 SEK.

iMac Pro will release in December this year. Yes they did. Same capabilities as GP100, so 30 TFLOPS for the Tesla V100

P100 is last years architecture. Would be more interesting to se how it compares to V100.

Yes, and according to AMD the TDP on Instinct MI25 is < 300W, so same envelope as Tesla P100 and Tesla V100 (which will ship in Q3) http://instinct.radeon.com/en-us/about/

Well, that's true. In many cases you can not write a singel OpenCL kernel and belive that you get maximum performance out of all architectures. You need to optimize for different vendors and different architectures.

Yes... For "Big Kepler" they have two GPUs. Titan, 780 Ti, Quadro K6000, Tesla K20/40 uses different revisions of GK110 Tesla K80 uses GK210 which doubles register file and shared memory.