The top-end AMD and Xeon server chips alone value over $7000 they usually carry out round 10x sooner than M1. With Apple M1 on 3nm, these Intel/AMD chips would drop to 5x sooner.
Huh? I understood that die shrinks have been extra about energy consumption than efficiency (additionally extra chips per wafer). You would possibly get a bit of additional efficiency due to elevated thermal headroom, however since thermals aren’t a lot of a problem with the M1 anyway I doubt that’d translate to a 2x efficiency leap.
They do each decrease energy and improve efficiency. Apple A14 (5nm) for instance was as much as 40% sooner than A13 (7nm 2nd-gen N7P), transistor rely went from 8.5b to 11.8b (round 40% extra transistors):
If they decrease the clock speeds to cut back energy whereas rising the variety of transistors, it will not be double however on the similar clock pace, doubling the transistor rely would supply as much as double the efficiency. Not all apps take full benefit of it however extremely parallel apps do. For issues like video processing, it might probably course of extra pixels on the similar time or a number of frames on the similar time.
M1 is on TSMC 5nm. 5nm+ this yr can be used for M1X/M2 and needs to be round 15% increased density. 3nm in 2022 will permit for greater than double the transistors in M1.
Good, 5 extra folks can now use the m1 Mac.
Just had to answer this. Linux powers a lot of the web. This very web site is probably going operating on a Linux server, so YOU are utilizing Linux proper now, not directly.
I’m wondering if that is the form of factor Intel is apprehensive about. Intel and AMD each cost important premiums for Xeon and Epyc server chips as a result of neither are full field producers, the chips go into Dell/HP/Lenovo {hardware}. Apple is a full field producer they usually can now make the highest-end chips in the most cost effective method on extra superior nodes than all of them. Apple may fully personal server {hardware} performance-per-dollar and per watt in the event that they wished to and if the {hardware} was absolutely suitable with server methods like Linux.
The top-end AMD and Xeon server chips alone value over $7000 they usually carry out round 10x sooner than M1. With Apple M1 on 3nm, these Intel/AMD chips would drop to 5x sooner. If Apple constructed a 2x sooner CPU for mainstream Pro customers after which a 4x sooner choice for high-end, this might be the quickest chip of any producer and Apple can construct them and ship them for revenue at a fraction of the value of Intel and AMD.
If they got here out with a revolutionary method of managing servers and made them like managing iPhones, they may make a big effect on cloud internet hosting and reseller internet hosting. That enterprise is value billions to Microsoft and Amazon.
Which cloud server is Apple utilizing for iCloud? I do know they give up utilizing Xservers a very long time in the past. What makes the AMD and Intel server chips so highly effective? More cores, extra busses, ??? I’ve at all times believed Apple was utilizing OTS server {hardware} as a result of it was inexpensive but you are saying it is far more costly. Their pace has to return from multi-cores or cluster computing. I keep in mind Dauger Computing arising with Pooch for cluster computing 20 years in the past. I do know it is extra concerned than simply assembling a bunch of servers however Apple may do that in the event that they noticed a strategy to generate profits from it. –or get monetary savings whereas doing it.
Some of the expertise Apple makes use of server-side is listed right here:
https://www.macgasm.net/news/interesting-technologies-included-apples-data-center/
More transistors and better clock speeds makes the AMD/Xeons extra highly effective. The Threadripper chip has round 32 billion transistors for the CPU components operating at 2.9-4.3Ghz (280W):
https://www.tomshardware.com/reviews/amd-threadripper-3990x-review
Apple has 16 billion in the M1 for the entire chip and that is break up between CPU/GPU/IO and runs at round 20W.
https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive
OTS {hardware} was inexpensive than Apple’s Intel server {hardware} – they hardly ever used the newest chips – however Apple Silicon is less expensive and extra energy environment friendly than something accessible simply now.
The Threadripper chips use a number of CPU chiplets 8×8 for 64-cores linked with Infinity Fabric (42GB/s).
Apple may do the identical with their very own chips. 8x M1 (32-core) on 5nm+ would carry out near the quickest chips accessible they usually can construct it for a fraction of the value.
They may not make a lot inroads into server corporations as a result of they do not have frequent {hardware} refreshes (it will likely be round 2 years they’ve taken for the 16″ MBP when it arrives) however the worth will be considerably decrease. They will even allocate a number of area to the GPU cores, which are not used a lot in servers however that is gaining popularity with AI and cloud rendering.
