Radeon Instinct Hopes To Compete With NVIDIA For Machine Learning, AI

[phoronix]

Phoronix: Radeon Instinct Hopes To Compete With NVIDIA For Machine Learning, AI

AMD/RTG is trying to better compete with NVIDIA in the AI and machine learning space with the introduction of Radeon Instinct...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-Instinct

[sykobee]

MI8 is clearly Fiji based.

The 3 PF rack is a good showing from AMD.

[boxie]

soo... Benchmarks of caffe and tensorflow are incoming? be interesting to see how well ROCm goes on the OSS side

[Michael]

soo... Benchmarks of caffe and tensorflow are incoming? be interesting to see how well ROCm goes on the OSS side

There's frequently Caffe benchmarks (on NVIDIA GPUs, haven't tried using the ROCm stuff lately), will see when I try the next ROCm release if I can get it running well. TensorFlow benchmarking support is coming as soon as NVIDIA is able to open-source some scripts for me.

[bug77]

MI8 is clearly Fiji based.

The 3 PF rack is a good showing from AMD.

Great strategy. Breaking into a new market with rehashed products. I wonder how bad things really are at AMD.

[juno]

Any news on ROCm+OpenCL?

[pal666]

Great strategy. Breaking into a new market with rehashed products. I wonder how bad things really are at AMD.

i wonder what makes you think nvidia makes separate chips for each market

[gururise]

This is huge! Myself and many of my colleagues have been forced to use Nvidia products because they have a monopoly on Deep Learning/Machine Learning. Most of the good DL libraries are written for CUDA. Really hope to see AMD make in-roads here.

[bug77]

i wonder what makes you think nvidia makes separate chips for each market

The first product based on Pascal was the GP100 based Tesla card.

But your question is irrelevant, AMD can rehash all they want, that's not the point. The point is, this is a really strange approach to entering a new market.

[pal666]

The first product based on Pascal was the GP100 based Tesla card.

they also have gp102 and gp104 teslas

But your question is irrelevant, AMD can rehash all they want, that's not the point. The point is, this is a really strange approach to entering a new market.

the point is, it is standard approach (nvidia entered with g80) and your question is irrelevant