AMD EPYC Infinity Fabric v. Intel Broadwell-EP QPI Architecture Explained

[Patrick Kennedy]

We provide a different view of AMD EPYC Infinity Fabric v. Intel Xeon E5-2600 V4 (Broadwell-EP) QPI architecture and NUMA design to share with colleagues

The post AMD EPYC Infinity Fabric v. Intel Broadwell-EP QPI Architecture Explained appeared first on ServeTheHome.

Continue reading...

 

[DWSimmons]

@Patrick did you get any sense of the run length maximums for infinity fabric (IF). We know for every published spec the maximum not IF since it's proprietary. The distance from node to node for a two socket EPYC is more than a couple inches but nowhere near the distance of a bottom riser 1x pci-e slot to the CPU. We were spitballing in another thread about the possibility of a GPU numa cluster a la the NVidia DGX-1. AMD has switching and routing going on with IF while Intel is bailing on ring to go mesh (as well). They have the pathing, numa, and timing coordination down, I wonder what is stopping them. Do you have a sense of AMD pushing the IF out to the slot?

 

[gradinaruvasile]

While we are still awaiting some of our performance data to settle (e.g. with new firmware that came in over the past few days)

This means the AMD final firmware arrived?

 

[Patrick]

No sense on maximum trace lengths for IF @DWSimmons

@gradinaruvasile I think this may be final firmware. Still doing some testing.

 

[Patrick]

This means the AMD final firmware arrived?

Looks like the box we have is getting an upgrade later this week. The good news is that will push us to DDR4-2666 which is important for IF as well.

Hopefully the new configuration we can publish benchmarks on. I have two sets of charts for the Skylake launch. One with 2x EPYC 7601 + DDR4 2400 and one without.

AMD is sensitive to us publishing numbers on de-tuned platforms, which I understand and agree with given it is a new platform and we are still a few weeks from folks being able to buy them publicly.

It is one of the hard bits about the work and a lesson learned from Ubuntu 14.04 with ThunderX that we are carrying forward. I do not think it helps anyone publishing pre-release system benchmarks if there is an expectation that those numbers can change.

Also, STH is not using an AMD reference platform tuned by AMD engineers so we are going more for real-world performance.

 

[gradinaruvasile]

Thanks for the update. Looking forward to the tests.
But yes, the best approach is to wait for the production-ready stuff and do thorough testing. Another "it is not good for gaming" (Ryzen) - like initial impression would be bad for an enterprise platform.

Do you have 1 socket test platform? I am really interested in the P series performance. Also, i am really curious as to why the 2S-ready CPUs have 2x the price of the P series...