AMD Vega10 Update from Hot Chips 2017

https://www.servethehome.com/amd-vega10-update-from-hot-chips-2017/

On a day when both NVIDIA disclosed more about its (now shipping Tesla V100) and Intel discussed Xeon Phi Knights Mill for deep learning, AMD was relatively quiet on this front. While AMD did discuss ROCm and increased compute capacity, both NVIDIA and Intel are showing specific vector compute acceleration technologies. At the same time, during the talk, AMD did specifically note that it has cryptocurrency mining. Perhaps the most exciting discussion was around the SR-IOV for virtualization. For those that are using GPUs for applications such as VDI, AMD has an interesting approach with a significantly friendlier licensing model than NVIDIA GRID.

Click to expand...

Is Vega10 the professional side? Looks that way from above.

Dayaks said:

Is Vega10 the professional side? Looks that way from above.

Click to expand...

I didn't have time to read the article or get into the pics, just enough to post. But don't think AMD segments architectures so likely a driver on/off feature.

Dayaks said:

Is Vega10 the professional side? Looks that way from above.

Click to expand...

I'm pretty sure Vega 10 is the base GPU for the 64 (fully enabled) and 56 (cut down)

that 4GB slide is interesting..I guess we might see a Vega 50 or the APU will have 4GB of HBM (One can dream). Still pissed about the launch. Was going to buy a R9 Fury with a GPU block until I saw Kyle's latest article. The 56 just crushes the Fury and it will be even worse on water.

ccityinstaller said:

that 4GB slide is interesting..I guess we might see a Vega 50 or the APU will have 4GB of HBM (One can dream). Still pissed about the launch. Was going to buy a R9 Fury with a GPU block until I saw Kyle's latest article. The 56 just crushes the Fury and it will be even worse on water.

Click to expand...

And some of the tech on it isn't currently utilized either so may get even better with a lot of luck

Dayaks said:

Is Vega10 the professional side? Looks that way from above.

Click to expand...

Yes and no, it was to be a chip that does it it all, but doesn't do either that well,

Vega doesn't have the bandwidth to sustain the professional market Virtualization (they are serious handy capped with that 483 Gbs). There was a reason why nV went with 4 stacks of HBM2 with P100, cause things like HPC, Virtualization, etc, when having multiple uses or multiple things going on at once, that bandwidth gets eaten up.

The gaming market, its got its own issues.

Where AMD has 6 frameworks ready to go with Vega, nV has 14, and more optimized middleware libraries.
All of the frameworks work with one library from nV too. So complexity wise its not like a person has to learn different libraries to attach to different frameworks, which if a college or company is exploring something that would best fit their needs, they don't need different groups of people with different skill sets to test.

Some of the actual applications of Vega (used for in DL), like training, if this was a year ago prior to Google launching their tensor chip or now Volta, its miles behinds.

Its really a proof of concept chip more that anything else.