AMD's CPU-to-GPU Infinity Fabric Detailed

[Marees]

News from AMD presentations in Rice Oil & Gas HPC conference yesterday

While AMD still appears to be a member of the HSA foundation, it no longer actively promotes HSA in communications with the press. In either case, it's clear the core tenets of the open architecture live on in AMD's new proprietary implementation, which likely leans heavily on its open ROCm software ecosystem that is now enjoying the fruits of DOE sponsorship

https://www.tomshardware.com/news/amd-infinity-fabric-cpu-to-gpu

https://twitter.com/addisonsnell/status/1234582848616697861

https://twitter.com/hatemltaief/status/1234572914432839687

 

[Thunderdolt]

This has its obvious use in large compute, but it also may allow something that matters much more to AMD than performance: cost reduction. If the GPU can seamlessly leverage system RAM, then AMD can reduce costs by reducing VRAM if not eliminating it entirely.

 

[Wat]

What about multi GPU access to the same ram pool? That's been one of the biggest hindrances to sli/crossfire,no?

 

[Nenu]

This has its obvious use in large compute, but it also may allow something that matters much more to AMD than performance: cost reduction. If the GPU can seamlessly leverage system RAM, then AMD can reduce costs by reducing VRAM if not eliminating it entirely.

System ram is 2 generations behind high end VRam, much lower bandwidth.
This wont be an enthusiast option.

 

[chameleoneel]

Sounds similar to what AMD did with the PS4. An extra bus to allow the CPU and GPU to access the RAM simultaneously. Less wait states and redundancy.