Any benchmark that can not account for ALL hardware (or popular/usable hardware) doesn't benefit any cross-MFG comparisons.The context was a poster saying that SME should be excluded from GB results on Apple. What's good for the goose is good for the gander.
Well, AMD/Intel also get a boost in that subtests due to AVX-512, so if you want to exclude Apple SME from the comparison, you'll have to exclude AVX-512 from x86 parts to keep the comparison fair.
Any benchmark that can not account for ALL hardware (or popular/usable hardware) doesn't benefit any cross-MFG comparisons.
Yes, but the power consumption of M4 has more than doubled compared to M1. One P-core in M4 running at 4.5 GHz consumes 7.2 to 9W.
If you have 12 of them (in M4 Max), the total power consumption only for the CPU part will be nearly 100W.
Of course, but the generally accepted IPC comparison has been SPECint for a while now.
Geekbench supports the AVX512-VNNI extension, which is available only on Saphire Rapids. The more generic AVX-512 is used only in one test (Background Blur). SME is used in three tests (Background Blur, Photo Library, and Object Detection).
So yes, there's a clear reason why Geekbench states that you can't compare those results, but who reads those boring release notes and specs?
It was around 5 watts for M1, it is around 7 watts for M4. Let's say 50% increase. Looks bad on paper, until we remember that these are sub 10 watt cores. I'd say they have a problem once they need to raise the wattage past 12 watts per core. Until then they are fine.
Which CPU runs at peak clock MC? They will run around 3.7Ghz, where they consume much more manageable 3-4 watts.
I don't have access to good quality SPECint results for all the platforms. I do have access to low-quality Geebbench results — but a lot of them, which allows me to do at least rudimentary statistical analysis. I see no ISO-clock peformance improvements in 7 our of 15 GB6 tests from M1 to M4, and 10-20% improvements in the rest.
Yeah there's ondevice cope-pilot exclusivity for Qualcomm.Another analyst, Patrick Moorehead affirms it in the replies.
We've already seen differences in the results Geekerwan uses for its comparisons. He might have explained the reason, but I have no time to watch videos.Anandtech has SPECint tests for Apple cores up to M3 I believe. Geekerwan also has a lot of SPEC tests for Apple and ARM cores though I tended to put a bit more weight on Anandtech's results in the past.
Interesting tweet from analyst Ryan Shrout;
View attachment 100353
Another analyst, Patrick Moorehead affirms it in the replies.
Yes, yes, now I think about it.. it perfectly aligns with leaks from XDA/WindowsCentral several months ago.Interesting but probably not a big deal in the end. I mean, AMD is marketing these as Copilot PCs and Microsoft and OEMs are as well, so it will probably be a slightly delayed update after launch.
As far as AnandTech goes, I'm not sure we had anything after M1
I think it’s worth keeping mind this was entirely about peak performance and AMD has other issues (incompetencies and sloth) they don’t pay enough attention to which was going to give Arm an opening with good to good enough perf — and mobile is most client volume, not desktop. But now? Now you can’t even make that case really, and Oryon + Arm will have further updates before Zen 6 — nevermind Apple but that’s kind of separate.Well, well, well, my friends...
The Zen5 hype train crashed spectacularly.
The SME isn’t adding as much as you think. IPC gains minus SME are still 5-6% over M3 which is in a single year and on top of an already high IPC. This isn’t amazing except for the fact that it’s in a single year and the absolute performance gain from Apple driving 5-6% is higher than it is for say Zen 4 to 5 with the same % — Apple still leads.Apple M4 supports the SME, but all other platforms do not.
I have good news.As a result, you can't compare those scores. That's what Geekbench says in their release notes for 6.3. From my perspective, GB6 is a terrible benchmark. The previous version is closer to the truth.
Anandtech has SPECint tests for Apple cores up to M3 I believe. Geekerwan also has a lot of SPEC tests for Apple and ARM cores though I tended to put a bit more weight on Anandtech's results in the past.
I'm not a fan of using iPhone SoC as a basis for comparison against laptop Mx chips. The gap between A14 and M1 was >10%.True, I was only thinking of 1t tests and looking now, we didn't get anything for M3 either, so M2 (A15) was the last time we got a review from Anandtech.
It was around 5 watts for M1, it is around 7 watts for M4. Let's say 50% increase. Looks bad on paper, until we remember that these are sub 10 watt cores. I'd say they have a problem once they need to raise the wattage past 12 watts per core. Until then they are fine.
Hope for nothing less than 20-23 months from Zen 5 launch.I think, in Zen 6, AMD will switch to N3 and polish the existing architecture to get an additional 10-15% IPC increase. I hope the competition from other companies will make this process faster, and we won't need to wait another two years.
Does he explain how he measures power?I have posted a screenshot from Geekerwan. There's 3.43W for M1 and 7.21W for M4 in SPEC INT and 3.92W vs. 8.95W for FP.
I have posted a screenshot from Geekerwan. There's 3.43W for M1 and 7.21W for M4 in SPEC INT and 3.92W vs. 8.95W for FP.
If all of those flopped, then did any of them really flop though?Arrow Lake: Flop
Zen5: Flop
Apple M4: Flop
Cortex-X5: ? (Flop, obviously, we know how this story ends)
I'm not a fan of using iPhone SoC as a basis for comparison against laptop Mx chips. The gap between A14 and M1 was >10%.
EDIT: and for that matter, even iPad vs MBP isn't really fair IMHO.
This slide on iGPU performance is a bit disconcerting.
Did they have the same memory BW and/or memory? The latency curves were different.As far as I know, that gap was due to M1 clocking higher, but no architectural change. Same is true for A15 and M2.