Another benchmark, M1 vs Ryzen 5800X on single threaded workloads. Ryzen wins the benchmarks, but the M1 has better perf/watt.https://github.com/tuhdo/tuhdo.github.io/blob/master/emacs-tutor/zen3_vs_m1.org …
-
Show this thread
-
Package power delta for 0 to 1 core loading on M1 is ~10W peak, ~25W for the Ryzen, though the author undervolted it so it might be lower. Core power deltas seem to be closer to 6W for M1, 17W for the Ryzen (guessing from the non memory intensive test for the M1).
1 reply 0 retweets 7 likesShow this thread -
But that Ryzen is definitely not optimizing for perf/watt. The 5600X is only ~5% slower on clock but ends up on ~17W package, ~12W core delta. So again I'm getting a bit of a "meh" feeling here.
1 reply 0 retweets 11 likesShow this thread -
I think if AMD moved to 5nm and put out an APU with the same design targets as the M1, they could hit comparable results. But Apple has been improving their chips at a ridiculous speed... So the real question is what will happen on their *next* generation.
3 replies 5 retweets 49 likesShow this thread -
Replying to @marcan42
TSMC 7nm+ (introducting EUV) is indeed optimised for performance and not really for power efficiency and AMD are definitely designing for that. TSMC 5nm EUV that the M1 uses has a massive 30% power improvement over 7nm.
3 replies 0 retweets 1 like -
Replying to @0xGRS
I bet Apple has better power management too. x86 systems have always kind of sucked at that, and that does a lot for the "battery lasts forever" results.
1 reply 0 retweets 0 likes -
Replying to @marcan42
Definitely. The problem AMD currently has is the 12nm I/O die power usage and whatever the heck is going on with the X570 chipset. That's kinda why I can see the I/O die going 7nm and K12 being introduced in the chipset. More so when one thinks of Epyc in the future.
1 reply 0 retweets 0 likes -
Replying to @0xGRS
Don't forget the RAM. On-package dies save a lot of power. Apple isn't doing that to screw over repairs folks. Multiple cm of a 128bit (2xSODIMM) multi GHz bus plus a connector? That's not free.
1 reply 0 retweets 0 likes -
Replying to @marcan42
Absolutely. I think we could see it with Zen 5. We learned with Vega that HBM2 packaging is just too expensive, but with 5nm EUV, they could just stack it on die for APUs. They have everything in place to do it on their end, and TSMC's. Just depends on the yields I suppose.
1 reply 0 retweets 0 likes
Just LPDDRx on the side would help vs desktop designs anyway, doesn't need to be *excessively* fancy.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.