Actually there are a handful of "Mac13,5" M3 results up on geekbench now. They're... not awesome. They claim high 2900s or low 3000s for 4.05GHz.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
I think that is impressive since it takes a 6Ghz 13900KS to match the score.
Huh.
The MT benchmarks on reported on GB are not within 0.2% of the M2 Pro, but they're REALLY close.
10c M2 Pro (6P4E) = ~12100
8c M3 (4P4E) = ~11700
That's less than 4% lower with two fewer P cores. Very impressive if accurate!
Last edited:
What’s the criteria for “awesome”?
That would put it at a higher score than a 13900k, right? The fastest or within spitting distance of the highest single thread of any current cpu.
I’d like it to be higher sure, but I’m not sure I’d be disappointed with that result.
Edit: call me crazy but this speculation that somehow the lower spec M3 pro will have less performance than the M2 pro seems too pessimistic. I think it’s perfectly reasonable to be disappointed if there’s only a little performance uplift sure, but I highly doubt that there’d be a performance regression.
That's more than the Qualcomm Oryon clocked at 4.3GHz gets under Windows and not the canned Linux score.
Wonder what the max clocks are for the clusters in a multicore load. At least the M1 didn't run all the P cores at 3.2 Ghz I think it dropped to 3.0.
No one is going to use Linux for that chip. There are even less professional apps than there is for macOS.
I think these results are not bad at all, especially considering that they will likely improve over time (the machines are probably still running the initial background bookkeeping jobs). The 4Ghz clock for the base M3 is also not too bad, it's 0.5 increase over M2. M3 Max is likely clocked at 4.2 Ghz.
I said preliminary GB6 ST results were "not awesome".
Don't get me wrong. It's an impressive score. But it's not awesome. I had hoped for a better result given the clocks.
The M2 hits ~2650 at 3.5GHz. If it could run at 4.05GHz, the score would be ~3066. So at least for GB6 ST, the M3 is showing very little in the way of IPC improvement. This is not hugely surprising, and it was no doubt a bunch of work for Apple to redesign it to support higher clocks. I'm not saying they screwed this up. But I'd hoped for a bit more.
That said, the MT results are VERY promising. And if I'm right about the desktops running higher clocks when they finally arrive, that's going to look quite good even for ST.
I won't bother quoting a couple other replies.
Don't get me wrong. It's an impressive score. But it's not awesome. I had hoped for a better result given the clocks.
The M2 hits ~2650 at 3.5GHz. If it could run at 4.05GHz, the score would be ~3066. So at least for GB6 ST, the M3 is showing very little in the way of IPC improvement. This is not hugely surprising, and it was no doubt a bunch of work for Apple to redesign it to support higher clocks. I'm not saying they screwed this up. But I'd hoped for a bit more.
That said, the MT results are VERY promising. And if I'm right about the desktops running higher clocks when they finally arrive, that's going to look quite good even for ST.
Don't you mean it the other way around?
Regardless, it is very well known that Geekbench produces better results under Linux. Folks say it's because of compiler settings and CPU power management differences (Linux is more likely to run the CPU unlimited and in case of Oryon reviewers did observe that the fan was running full blast all the time due to missing power management drivers).
I see what you mean. It does concern me a little that there’s not much improvement in ipc, but not having the whole scope of the situation, I’d hazard a guess that there was some engineering tradeoff made there.
Maybe. There was one much lower score, which I ignored for that reason.
That's how they did it for M2 but I think they might make it more dependent on chassis than chip now. We'll see. The really interesting question is what happens in the Studio and Pro.
If they do 4.2GHz on the Max in the 14" (or even in the 16") I'll be impressed. That would get you to 3150, and that would be seriously good stuff. Close to the i9 14900K, in a frickin' laptop!
Well of course, we talked about this a couple weeks ago in the context of the A17. They put in some IPC improvements, then had to sacrifice some IPC to get it to run at higher clocks. Net result was very little IPC change. This isn't surprising. Like I said, I just hoped for a little bit more.
Remember also how the initial batch of A17 Pro scores were between 2750 and 2900, but if you look at them now the bulk is over 2920. I am sure we will see a similar effect as more M3 scores are coming in.
The more I think about this the more I'm convinced the GB6 results are real. If so, the MT performance is a real triumph. The M3 Pro will be much better than skeptics feared, and the Max will be a monster.
GB6 doesn't scale multi-core linearly, for those of us with tasks that scale well across cores it isn't as good of a measure as measuring our real workloads - I'm waiting to see the large codebase compilation tests.
dear confused-user
he lives in a bubble.
no one uses MacBooks to do 3d rendering, such that they will see the difference between a m2 and m3.
It works both ways.
1 - ARM actively contributes to the Linux kernel making sure to include it's optimizations for ARM.
2- And Linux has added deep kernel support for ARM64 Architecture , including performance counters, pointer authentication, kernel page table dump, and various extensions like Scalable Matrix Extension and Scalable Vector Extension among others.
Does Cinebench 2024 measure your workload better?
Cinebench 2024: Reviewing the Benchmark
Maxon’s Cinebench is a perennial benchmark favorite. It’s free, easy to run, and scales across as many cores as you can give it. Its $0 cost allows the internet to provide plenty of res…
chipsandcheese.com
That will change with M3 as blender supports hardware RT now