M3 core counts and performance

tomO2013 said:
This is a very interesting breakdown over at anandtech ….
www.anandtech.com

Apple Announces M3 SoC Family: M3, M3 Pro, and M3 Max Make Their Marks

www.anandtech.com www.anandtech.com

They speculate that the reason we are seeing different TOP numbers for A17pro and M3 is down to the precision….

A17pro apple communicated an 35 TOPS number at INT8 precision. Anandtech speculates that M3 with a figure of 18 TOPS is likely quoting INT16/FP16 precision. M1 and M2 communicated results based off of INT16/FP16.

Anandtech also asks (what I feel will be a very interersting thing to explore with live production samples) in the hands of real users - does M3 also support INT8 and allow to trade precision for throughput?!?!

This was probably the biggest head scratcher for me of the entire keynote!
Click to expand...

Jimmyjames said:
I think @leman might have said that already? Also I believe Qualcomm’s npu is int4.
Click to expand...

It was @Andropov
Andropov said:
I found this surprising too: https://x.com/iancutress/status/1719176613445742855?s=46&t=AVo4Ae4rwcqD3xOi5WvXyg

Looks like the M3 doesn’t have the A17’s Neural Engine.
Click to expand...
However Ryan is being a bit more circumspect as to whether the NPU is architecturally the same as in the A16 vs A17, stating it could be inconsistent messaging rather than using the old NPU.
 
Jimmyjames said:
GFXBench score for M3; 52.2fps. M2 was 48fps approximately. Seems like gfxbench is problematic lately.
gfxbench.com

GFXBench - Unified cross-platform 3D graphics benchmark database

The first unified cross-platform 3D graphics benchmark database for comparing Android, iOS, Windows 8, Windows Phone 8 and Windows RT capable devices based on graphics processing power.
gfxbench.com gfxbench.com
Click to expand...
Can we stop using gfx bench for graphics. It's not a great benchmark for graphics performance especially with RT and mesh shaders now.

Nvidia GPUe also perform bad at GFX bench and they are the best GPUs available
 
exoticspice1 said:
Can we stop using gfx bench for graphics. It's not a great benchmark for graphics performance especially with RT and mesh shaders now.

Nvidia GPUe also perform bad at GFX bench and they are the best GPUs available
Click to expand...
As always people (including Apple) choose the benchmark that best displays their favourite device.
 
Here's an OpenCL score for the base M3 (10 GPU cores); it's the highest of the three I saw listed. The 15" M2 Air (also 10 GPU cores) is listed at 27,884 on the GB site. 30,615/27,884 = 1.1

1698870176160.png
 
Andropov said:
Curious about the M3 Pro/Max and whether those are clocked higher and have higher single core scores.
Click to expand...
Me too. Though, thermally, the one that could be clocked the highest is the base model, since it's got significantly fewer cores than the Pro and Max, yet is now in the 14" chassis. [Unless they reduced the number of fans on the 14" base from two to one, like they on the base iMac.]
 
theorist9 said:
Me too. Though, thermally, the one that could be clocked the highest is the base model, since it's got significantly fewer cores than the Pro and Max, yet is now in the 14" chassis. [Unless they reduced the number of fans on the 14" base from two to one, like they on the base iMac.]
Click to expand...
They did. That much is confirmed. It's just got one fan.

And you can always have dynamic clocking, such that using just one core intensely clocks it at, let's say 4.4 and using 6+ cores then goes down to 4GHz or whatnot.
 
theorist9 said:
Me too. Though, thermally, the one that could be clocked the highest is the base model, since it's got significantly fewer cores than the Pro and Max, yet is now in the 14" chassis. [Unless they reduced the number of fans on the 14" base from two to one, like they on the base iMac.]
Click to expand...
Apple Silicon chips have had little thermal limitations so far, clock is not limited by thermals, particularly on short tests like Geekbench. In fact the fastest single core scores on previous generations have been from Pro/Max chips (which have more cores).
 
casperes1996 said:
And you can always have dynamic clocking, such that using just one core intensely clocks it at, let's say 4.4 and using 6+ cores then goes down to 4GHz or whatnot.
Click to expand...
Yeah, I've wondered why Apple didn't do that in the M1/M2, particularly on their desktop devices -- whether it's because they haven't been able to scale those chips to those higher clocks, or if there's some other reason.
 
Andropov said:
Apple Silicon chips have had little thermal limitations so far, clock is not limited by thermals, particularly on short tests like Geekbench. In fact the fastest single core scores on previous generations have been from Pro/Max chips (which have more cores).
Click to expand...
I meant if they wanted to significantly increase clocks--such that thermals would matter—the 14" base would have been the best candidate (if it had the same cooling, which it doesn't).

Further, I recall that, under heavy sustained loads, the 16" Max performed somewhat better than the 14" Max (even for the M1 generation, where the 16" didn't have a higher clock), indicating that they are operating at the edge of where thermals would matter.

Plus the fact that Pro/Max chips (which are in more thermally capable machines) have higher SC performance than base M's, even with the same clocks, indicates thermals do matter somewhat. Indeed, if you add even more thermal capability, going from the 16" MBP to the Studio and MP, the SC scores increase further, even with the same clocks, indicating that thermals are having an effect even here:

1698872431561.png


To be sure, these thermal effects are small compared to what is seen in Intel laptops, but they are there nonetheless, supporting the view that Apple is operating at the edge of where thermals matter....which, as you know, is where you'd want to be in a well-designed device.
 
Last edited:
Metal score for base M3. 47587

Mac15,3 - Geekbench

Benchmark results for a Mac15,3 with an Apple M3 processor.
browser.geekbench.com

Very little graphics or compute improvement which is strange as that is what they stressed the most.
 
exoticspice1 said:
I guess 3NB can't clock high. The reason why M2 Max was higher clocked because 5NP was highly optimized by theat point.
Click to expand...
I suppose we can fantasize that the real reason Apple decided not to release the Mini and Studio this week is because they plan to increase the clocks on the desktop devices, and can only do it with N3E 😜.
 
Jimmyjames said:
M3 Max scores. Multi core = 21084!!!!

Mac15,9 - Geekbench

Benchmark results for a Mac15,9 with an Apple M3 Max processor.
browser.geekbench.com

Frequency still 4 Ghz. Hmmmm, wonder if that will change.
Click to expand...
3151 in single core too. Very solid upgrade.
Oh nvm it was reported in other site as being 3151 single core 🤔 And Geekbench was down.
 
Last edited:

Similar threads

T
9to5Mac rumor: "New 13- and 15-inch MacBook Air models will both use M3 chips"
Replies
15
Views
966
Clix Pix
Clix Pix
T
ARM-based NVIDIA Grace server chip with 500 GB/s UMA claims lower power and higher performance than competing Intel & AMD designs
2
Replies
32
Views
2K
mr_roboto
M
Eric
Meta signals more layoffs could be coming after 'thousands' of workers are reportedly given low performance reviews
Replies
11
Views
386
Nycturne
Nycturne
Back
Top