It was necessary?
Because Nvidia is now in the largescale systems (kit) business?
Because SuperMicro has always been in bed with Intel and Nvidia?
No matter how hard AMD trys to kit Epyc with Nvidia accelerators?

“AMD buy SuperMicro”, Intel’s 30-year system development partner like Dell Intel sales that is now Dell Nvidia sales.

It was necessary?

Whyt does Mercury under report AMD server share?
Never reporting on current generation.
Because Xeon might make a comeback dashing expectations?

Now let’s look at primary channel share leaving q2.

Bergamo + Genoa + Sienna + 4000 = 23% same as Mercury relies on ebay data
Emerald + Sapphire + Ice = 77%

AMD adding Milan Epyc channel increases to 32.7% and Intel drops to 67.2%.
AMD adding Rome Epyc channel increases to 46% and Intel drops to 54%.

AMD Bergamo + Genoa + Sienna = 54.7%
Intel Emerald Rapids + Sapphire = 45.2%

See how Mercury under reports AMD server share by not detailing all the options.

I do speculate this is about not [?] foretelling an Intel comeback and done purposely?

Mike Bruzzone, Camp Marketing

Fair points. I hope to see AMD spec’d and branded systems though.

They can control the manufacturing process without it being on their books, just like Google, Microsoft, AWS, and Meta do. Nvidia doesn’t literally make DGX machines, but it has them made to its specs. And probably by Supermicro, if we had to guess.

Nvidia benefits from a shorter time frame to cycle learnings about manufacturimg back into the chip biz, also from that the mfg and chip sides have incentive to transparency with eachother. There is a synergy in doing both.

If AMD is sucessful in challenging Nvidia then there will be plenty of demand to go around. I think AMD should keep a mfg arm as long as it is profitable. What is a bunch of mfg engineers without factories to play with – a sinking resource IMO.

I agree open systems designed for composability/reusability require more engineering effort to achieve good performance across use cases and it’s important to get expertise in those use cases but either direct hiring to acquire it or placing AMD engineers on customer design teams seems much better than trying to figure out what to do with 1100 experienced system engineers (which ideally would each be strategically placed). I’m sure there are a dozen or more engineers who would pivot easily from rack-level design to board or chip level design (or design for test, which seems even more of a reach) but how good will AMD be at finding those (without just assigning everyone the new training/tasking and seeing who can hack it)?

I think you make a good point. The Ramapo fault is not that far away!