[chad]

Hello everybody!

Looks like I will be able to proceed with a dual configuration AMD Epyc build (though this was a bit of hassle trying to track down everything).

With that being said I should be able to have this all built and hopefully running by the end of the month, so I should be able to do some performance testing if anyone wants.

The set-up will be a Supermicro H11DSI-O with two 7281 processors. The one downside is I was only able to afford 96GB of ram at this point (12 x 8GB:2400 Mhz) meaning I won't be able to take advantage of the 8 memory channels right now.

[flotus1]

I understand that you are not looking for any recommendations. Anyway, here are my thoughts on your setup
The Epyc 7281 only has 32MB of L3 cache. The Epyc 7301 costs ~150$ more and has the full 64MB of L3 cache. And then there is the issue with the inter-CCX communication with Naples. The speed of this "infinity fabric" is directly linked to the memory frequency.
If I had to choose on a tight budget, I would consider starting with one 7301 and DDR4-2666. It will be less performance initially, but this way you are not limited once you get the budget to upgrade to a second CPU. However you decide: make sure to get dual-rank memory modules. Just my two cents...

[chad]

Flotus,

Thanks for the suggestion, I haven't quite ordered yet so I'll still take any advice!

In regards to the 7281 vs 7301, would it be a better to get two 7301 right away and only 64 GB of 2666 MHz (32Gb per processor) or get the full eight channels for only one processor and try to upgrade later?

Furthermore, I would estimate a 50/50% chance of receiving extra funding so I want to make this as best as possible assuming I won't upgrade in the future.

[flotus1]

Originally Posted by chad

In regards to the 7281 vs 7301, would it be a better to get two 7301 right away and only 64 GB of 2666 MHz (32Gb per processor) or get the full eight channels for only one processor and try to upgrade later?

Two processors are definitely better than one with the same total memory banwidth. You get twice the amount of cores and cache. And even if scaling is bad beyond 16 cores performance will still be higher.