In his last video, Moore law's is dead, a reliable leaker about Intel and Amd, confirms that Apple is making its server chips for internal use, so he believes that Apple may enter the server market again.
"Server" chips or Hyperscaler/Cloud chips?
Amazon is rolling out more Graviton2 (Neoverse0 nodes than x86_64 nodes. Baidu is working on an AMD Neoverse solution. Microsoft appears to be working on a Neoverse solution. Oracle's cloud has Ampere nodes . etc.
Who isn't a top end hyperscaler than doesn't have some ARM Neoverse /Ampere / etc ARM nodes in deployment at this point?
If Apple is building (or contracting semi-custom work ) cores for the back end of iCloud , email , photos , etc. (i.e., primarily all Linux workload now on x86_64 ) then Apple could have a hyperscaling chip of their own , but that would have nothing to do with macOS or retail sales at all. Aamazon isn't selling servers. They sell time on servers but not the servers themselves. Google, Facebook , Azure , etc.
So just because it is a hyperscale server chip wouldn't necessarily mean a retail product would fall out of that.
The Jade2c and Jade4C 20 / 40 core chips serving as "server" chips for XCode Cloud. Perhaps . But somewhat doubtful that is actually going to cut the "grade' as a server chip for general purpose (e.g., non macOS ) workloads that invovle substantive I/O throughtput. Apple probably isn't going to scale macOS past 64 cores ( a cap that it has now). There are zero 3d party GPGPU drivers. Again look at major cloud services vendors who don't offer any AI/ML backend coverage. Apple macOS cloud licensing model doesn't really support mass , multitenant servers either.
MacStadium/MacColoc runs 10's of thousands of Minis as servers. It is a "server business" but it isn't the same as what is being talked about in the vast majority of that video.
The bulk of that video is about AMD Bergamo ( 128 core AMD Epyc variant for Cloud/Hyperscalers) . Supposedly the AMD (or Intel) competitive intelligence for said that the competition was "Apple/ARM". If actually look at what Ampere is doing with
Any tech startup that wants to live beyond is seed and venture funding rounds and make it to either an initial public offering or an acquisition by a
www.nextplatform.com
They are moving to their own custom cores from baseline ARM Neoverse ones. That is mainly to do customer SoC for different Cloud/Hyperscalers. Those cores could run the current backend of iCloud, Apple email , etc. just fine at lower power than the x86_64.
The roadmaps for AMD and Intel for high end server cores TDP is just up, up and up. 500W ... 600W ( noted in the video). That is actually a problem for folks that need to do "carbon neutral" computing for various workloads. That is the problem that Bergamo is suppose to help solve. Zen4D cores ...toss the single threaded drag racing low perf/watt chase and go for more effective mutliple thread for specific workloads.
Bergamo is suppose to come in 2023 with 128 cores. Look at the chart above. Ampere is shipping 128 cores now. 2022 probably go to 5nm can crank the core count again. 2023 ... they are playing catch up.
Apple's 40 core , only supports boots macOS , no 3rd party GPU chip likely won't be the primary issue. Apple M1 sereis as development workstaiotns/laptops to deploy to ARM Neoverse N and V series
Arm is hosting its annual Tech Day shindig, virtually (again) thanks to the coronavirus pandemic, and is providing a lot more insight into the future
www.nextplatform.com
One of the chicken-egg thing with server ARM is that developer tool on same general platform. ( e.g., Xeon SP in HP z8 deployed to Xeon SP server ). Apple delivering more developer client systems to deploy from is a bigger threat than Apple flipping some semicustom on their own back end.
To do a minor tweak on Neoverse N2 or V2 Apple doesn't need a huge design team fork from their own fully custom work.
How could those chips be? Same as Mac Pro chips? Does it make sense to Apple enter the server market again?
At a big enough point it makes sense for a Hyperscaler /Cloud vendor to do their own server boards. They don't have to be fully custom but once at the point have internal hardware team to specific nodes , then can contract spec own tweaks to "Open Compute " designs to meet minor variations needs for server cluster designs.
For XCode Cloud yes. But those really don't have to be "servers" in classic sense. There is large macOS CI market done with Mac Mini and Mac Pros now. The Mac Pro 2019 has a "rack model" for the case but it isn't a classic hyperscale box. A "half sized" Mac Pro with much fewer slots would probably rack better at a higher density. So yeah they could do that but it isn't going to put AWS , Azure , or any major cloud services vendor out of business.
P.S. MLID seems to ramblie in and out of talking about consumer/gaming chip rumors and into some of the more esoteric stuff that Intel and AMD have coming in the server spaces . Often trying to maps the latter back down into products or at least cast them into "sizzle" for the enthousiant/gamer folks to consume as buzz.
About 18 mins into video he makes a comment about being "less worried AMD can compete with Intel in the server space". Generally He is still on the Intel takes back over in 2024-5 theme. More than a decent chance this "Apple server" thing is being thrown out there to get clicks. It is likley a
HUGE tangent from the market that Bergamo (and Intel's and other ARM vendors ) are trying to cover as a market segment.
forums.macrumors.com
linustechtips.com
