New Intel SapphireRapids vs AMD Epyc Genoa in the Server space

[SpyderTracks]

So this is well worth a look, AMD still absolutely dominate in the Server space with their new Epyc Genoa lineup in both price and performance

These new Intels are supposed to be "scalable" in the fact that you can get server boards with up to 9 CPU sockets, so that's up to 540 Cores in one server chassis

Problem is, Intels 60 core headline chip costs $17000 vs AMD Epycs 96 core beast costing under $12000 and you'd need 2 of the Intels to keep up with one of the Epycs, plus using about half the power

Plus Intel have locked certain features of the CPU behind a paywall, so you have to call them and pay them to unlock further performance that is already on the chip, but locked away by software limitations

DLCs for datacenters: Intel Software Defined Silicon in upcoming Xeon Sapphire Rapids locks CPU features behind a paywall

Intel could be introducing Software Defined Silicon (SDSi) with upcoming server and datacenter processors such as Xeon Sapphire Rapids. SDSi enables unlocking processor features on a pay-as-you-go basis to minimize upfront costs and enable organizations to add new features without a hardware...

www.notebookcheck.net

People ask me why I'm anti Intel.... THIS IS WHY, they've turned uber greedy, only interested in the bottom line, they don't care about their loyal customers.

 

[Scott]

I would love to see the counter argument for this. Surely we're missing something. On the face of it you would be off your rocker to invest in that platform.

 

[SpyderTracks]

I would love to see the counter argument for this. Surely we're missing something. On the face of it you would be off your rocker to invest in that platform.

From what I'm reading, Intel are purely banking on the AI accelerators baked on the die on these. Outside of AI, performance is utterly useless, but for AI workloads, it beats the pants off the Epyc lineup (skip to 12:15)

So they're hoping that early adopters will buy these up now and due to the "scalability" of them, they'll adopt them with long term upgradeability in mind by dropping in further CPU's when they need added performance.

It's a huge gamble, especially when you factor in most larger enterprises would get dedicated GPU's for AI acceleration

To me it's a sign they just have nothing else to offer currently and IIRC the big data warehouses including google upgraded to Epyc chips only recently

https://www.amd.com/en/press-releases/2022-02-10-google-cloud-chooses-3rd-gen-amd-epyc-processors-to-power-new-compute

And IBM are launching their own dedicated AI hardware: https://research.ibm.com/blog/ibm-artificial-intelligence-unit-aiu

Apple use Amazon Web Services, so they're pretty much out

I dunno, I don't see it, but then again, Intel never fail to surprise me