Unfortunately AMD totally missed the boat on ML.
I think by this measure so did everyone else. Nvidia pushed cuda, and everyone took it, even though 8 or so years ago openCL was a reasonable competitor. But various orgs choose to make a proprietary technology a first class citizen rather than putting the effort into an open alternative. Big win for nvidia, big loss for everyone else not large enough to create their own ML frameworks and accelerators.When you build a processor that literally has "do 4x4 matrix multiplication" as an assembly statement, you'll get far superior ML performance compared to everyone else. (Except Google's TPUs, which have larger matrix-multiplication arrays)
The open-source OpenGL drivers are very good.
I know ML is where the current money is at, but coprocessors in general seem to be applicable to far more problems than just machine learning and graphics.
What if AMD actually acquires Xilinix?
Everybody thought it's about CPU-FPGA integration but, they may pull something very, very different.
https://www.xilinx.com/support/documentation/white_papers/wp...
True, a systolic processor can be built out of these components, but I doubt it'd be anything as fast as a proper matrix-multiplication unit that NVidia is putting out. And that's after the complications associated with FPGA programming.
----------
Xilinx's FPGAs are a combination of LUTs (4-bit or 6-bit look-up tables), Routing, and "DSP Slices" (prefabricated multipliers). The bulk of your math is still going to be performed in the DSP Slice.
Where FPGAs win is that their LUTs and Routing tables allow you to create custom high-speed glue-logic between these DSP slices. But when it comes to something like a regular Convolutional neural net, the routing is very straightforward.
------
I'm not really a deep learning expert. I know that FP16 seems to be pushed by NVidia. This INT8 stuff seems odd to me (there's no way INT8 gets the same dynamic range as FP16), but I'm also not really sure if the dynamic range is needed or useful in deep learning applications.
The only consolation for me is that I/O over PCI ends up being my ultimate bottleneck. If AMD did a full-on integration with their infinity fabric, it could actually be a game changer. For me, in my weirdo niche.
----------
AMD's GPU Infinity Fabric seems to be a different infinity fabric from their CPU. CPU Infinity Fabric is ~50GB/s, but the GPU Infinity Fabric is GPU-to-GPU links of ~90GB/s.
Realistically, PCIe 4.0 and 5.0 will be the path forward for standard PCs. There will always be incredible application-specific solutions (such as NVLink / NVSwitch on the DGX), but the general computer user will wait for the open standards to support such speeds before adopting them.
At the moment, GPU-to-GPU communications probably will scale better than CPU-to-GPU comms. CPUs just aren't built for the bandwidth (outside of POWER9 / POWER10).