So it’s confirmed these scores are for the base 14” MacBook Pro.
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.
First Geekbench for base M3
- Thread starter APCX
- Start date
- Sort by reaction score
M1 Pro is 2,376 single-core & 12,180 Multi-core
M3 is 3,030 single-core & 11694 Multi-core
M3 is 3,030 single-core & 11694 Multi-core
Did you expect frequency to stay the same?? That's how speed increases
It’s only ok for everyone else. With Apple it’s a scandal, an outrage!
Where is the joke? I don't get it. These are very good improvements.
3nm was advertised with 15% higher performance at the same power, so this is what we are getting. Did you expect something else?
M3 OpenCL:
M2 OpenCL:
Maybe in Metal there’s more difference. I’m still looking those scores up.
Nah. Not over-hyped. You just fail to grasp what a big leap it is to be producing working, sellable, apparently cost-effective v1 of a chip on an ~3 nm process. M3 approaches theoretical maximum densities, and it is working. Just wow.
Me, I am hella impressed. From here Apple/TSMC move forward on ~3 nm.
Wow, that multi core score. Interesting to see it run at 4 ghz though.
I don't actually think this is a bad result (albeit perhaps not quite what people, myself included, were hoping for.)
In this test, the 16 core M3 Max basically matches the 24 core M2 Ultra with 4 (33%) less P-Cores and 4 (50%) less E cores. Granted GeekBench 6 is not the best indicator of multithreaded performance, but still, it's something.
IMHO, the OpenCL Compute benchmarks are where things are a bit more worrying.
M3 Max: ~93K
vs
M2 Max: ~88K
M1 Max: ~71K
So ~31% faster than M1 Max and ~6% faster than M2 Max.
Obviously OpenCL is a semi-deprecated API on macOS at this point so I think the Metal results, and more importantly, actual application benchmarks, will be far more indicative of real performance, but still, not a great start...
(To be fair however, it's clear that the focus of the new GPU design was adding new, forward looking features like ray tracing and mesh shaders, so judging it purely on raster/compute is arguably unfair)
You were hoping for a better multi core result than this? I’m pretty sure @Homy was being facetious.
Agree the gpu results are disappointing though.
I think the result is great for a 16-core M3 Max, almost as fast as 24-core M2 Ultra. I was thinking of the rumor about M3 Max beating M2 Ultra before these actual scores came out. The GPU is not disappointing either because Metal is the one that counts. OpenCL has always been very slow on Apple Silicon since it was deprecated.
Hmmm not so sure. We have one Metal score for a base M3 and it’s only a 7 or 8% increase.The GPU is not disappointing either because Metal is the one that counts. OpenCL has always been very slow on Apple Silicon since it was deprecated.
https://browser.geekbench.com/v6/compute/1209070
Yes I saw that but I'm waiting for M3 Max. It should be at least 20% faster according to Apple.
As I said, I think the multicore results are perfectly fine, and if they hold up in application benchmarks, good (I added the part about expectations last minute.) I think people, myself included were hoping for a larger single core uArch improvement (that might be reflected in the multicore score,) but that didn't happen.
Like I said, Metal and app benchmarks matter much more than OpenCL, but the performance scaling we see here doesn't exactly inspire confidence.I think the result is great for a 16-core M3 Max, almost as fast as 24-core M2 Ultra. I was thinking of the rumor about M3 Max beating M2 Ultra before these actual scores came out. The GPU is not disappointing either because Metal is the one that counts. OpenCL has always been very slow on Apple Silicon since it was deprecated.
Case in point, if the Max scales like this that would be a real disappointment. Still, we need more real data before we can draw any meaningful conclusions, and as I said, the focus on this year's GPU was clearly forward looking feature support not raw raster/compute.
Oh yes I was hoping for more single core improvement too. I think there is a chance that these scores undersell the actual performance. Frequently new machines are indexing and doing other maintenance tasks for a couple of days. After that scores do improve.
I do think the multicore scores are more than fine, they are spectacular .
Fair points.
Is it though? GHz is arbitrary between different chips and generations.
The fact Apple is putting a "4GHz chip" in their entry level computers means nothing. The actual performance, efficiency, that's what matters.
Also the Metal tests on geekbench do not measure RT.
Wait for Blender/Redshift tests with v4.0 for results.
Wait for Blender/Redshift tests with v4.0 for results.