So M1 Max with 32 Core GPU matches RTX 3080 Mobile.

[ElfinHilon]

Right, but it is still very likely Apple would provide the maxed out 32c version to reviewers

I am actually really unsure about this. Apple has provided reviewers with less than the top end devices in the past. I really think we should wait a bit and see what comes out.

I see a few scenarios playing out:
1. The GPU is actually the 24 core.
2. The GPU is actually the 32 core, but geekbench has some sort of scaling issue with the cores.
3. The GPU is actually the 32 core, no scaling issues, but this benchmark is the only benchmark where we receive (much) lower scores than anticipated.

Given what's been stated, I'm expecting either 1 or 3 to happen.

 

[mi7chy]

GFXBench is probably best case since it's garbage, not as garbage Geekbench mid case and real world workload like hashcat worst case. Got hashcat 6.1.1 working again uninstalling/reinstalling so I'll see if I can install/run it on store display MBP M1 Max.

 

[dugbug]

GFXBench is probably best case since it's garbage, not as garbage Geekbench mid case and real world workload like hashcat worst case. Got hashcat 6.1.1 working again uninstalling/reinstalling so I'll see if I can install/run it on store display MBP M1 Max.

Here is a gfxbench m1max compared to m1. I can't speak to what model.

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

 

[EugW]

Here is a gfxbench m1max compared to m1. I can't speak to what model.

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

It's the 32-core GPU M1.

Scaling ranges mostly from 3.17X to 3.99X, with the average looking to be about the middle of that around 3.5Xish, if we exclude one outlier test that scales at 2.4X (Car Chase offscreen).

 

[Serban55]

https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

 

[Natrium]

I was right; the benchmark was done on the 32c gpu M1 Max.

More results have since appeared on Geekbench that explicitly mention 32c: https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

 

[ElfinHilon]

I was right; the benchmark was done on the 32c gpu M1 Max.

More results have since appeared on Geekbench that explicitly mention 32c: https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

As we've been saying, either the processor or Geekbench may be misreporting/labeling what the GPU cores are. If these are all actually 24 core and not 32 core, you'd expect all of them to report in the way shown.

That being said, these very well could be the 32 core GPU's, but it'd be a little disappointing to see the scores this low. I would personally doubt that the geekbench scores listed are actually indicative of performance.

 

[luvmango]

I was right; the benchmark was done on the 32c gpu M1 Max.

More results have since appeared on Geekbench that explicitly mention 32c: https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

There's only one metal benchmark and it doesn't list the compute units

 

[JimmyjamesEU]

I was right; the benchmark was done on the 32c gpu M1 Max.

More results have since appeared on Geekbench that explicitly mention 32c: https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

Not convinced that proves anything. Only two models exist on the database at the moment: 18,2 (MacBook Pro 16) and 18,3 which is the 14. Until we see either a score between this supposed 32 core and the 16 core, which would mean you are correct, or we see a higher score, meaning this is the 24 core.

 

[jinnyman]

Here is a gfxbench m1max compared to m1. I can't speak to what model.

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

If this is 32 gpu one, I have to say I'm disappointed. Of course, I have to wait for a real world experience, but still disappointed.
I was planning to purchase a highend one, but now I'm contemplating whether I shoulld wait another year (waiting for M2!)

 

[Natrium]

Not convinced that proves anything. Only two models exist on the database at the moment: 18,2 (MacBook Pro 16) and 18,3 which is the 14. Until we see either a score between this supposed 32 core and the 16 core, which would mean you are correct, or we see a higher score, meaning this is the 24 core.

There IS a M1 Pro 16 core result with the score 38.359:

MacBookPro18,3 - Geekbench

Benchmark results for a MacBookPro18,3 with an Apple M1 Pro processor.

browser.geekbench.com

 

[JimmyjamesEU]

If this is 32 gpu one, I have to say I'm disappointed. Of course, I have to wait for a real world experience, but still disappointed.
I was planning to purchase a highend one, but now I'm contemplating whether I shoulld wait another year (waiting for M2!)

Seems pretty good to me. Here is the Max compared to a 3080 laptop gpu.

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

 

[JimmyjamesEU]

There IS a M1 Pro 16 core result with the score 38.359:

MacBookPro18,3 - Geekbench

Benchmark results for a MacBookPro18,3 with an Apple M1 Pro processor.

browser.geekbench.com

I said there was. Not sure I understand your point. 1 14 inch model and 5 16 inch results. If anything that 14 inch model emphasizes my point. Perfect scaling from 8 to 16 cores.

 

[Serban55]

I was right; the benchmark was done on the 32c gpu M1 Max.

More results have since appeared on Geekbench that explicitly mention 32c: https://browser.geekbench.com/v5/compute/search?utf8=✓&q=MacBookPro18

and like others had said to you
"
Nope... apparently Geekbench is not reporting correct stats...

reported cache is wrong
max frequency is wrong
and many suspect compute units is wrong also"

 

[ElfinHilon]

There IS a M1 Pro 16 core result with the score 38.359:

MacBookPro18,3 - Geekbench

Benchmark results for a MacBookPro18,3 with an Apple M1 Pro processor.

browser.geekbench.com

Yes, and that's the 16 core if you scroll down.

EDIT: And the reason why aren't doubting that one, is because that score is in line with the 2x performance we expect to see over the 8 core M1. It's almost exactly double.

EDIT 2: A Reddit user has managed to find 4 different identifiers for the macbook pro:
18,1
18,2
18,3
18,4

If you look at the results from the CPU bench, there's TWO different pro maxes, but ALL of the given GPU benchmarks we have from geekbench come from 18,2, which is likely to be the 24 core. Only other thing I can think of is that 18,2 could be the 14 inch with 32 core GPU, while the 18,4 is the 16 inch 32 core GPU. I'm rather doubtful on that, but we don't know yet.

 

[JimmyjamesEU]

With the caveat that it's apparently not a great benchmark, more results coming in

Apple M1 Max Catches up to RX 6800M, RTX 3080 Mobile in GFXBench 5

The M1 Max is flexing its graphics muscle in new benchmarks.

www.tomshardware.com

Not definitive proof but something to consider as more results appear.

 

[JimmyjamesEU]

Yes, and that's the 16 core if you scroll down.

EDIT: And the reason why aren't doubting that one, is because that score is in line with the 2x performance we expect to see over the 8 core M1. It's almost exactly double.

EDIT 2: A Reddit user has managed to find 4 different identifiers for the macbook pro:
18,1
18,2
18,3
18,4

If you look at the results from the CPU bench, there's TWO different pro maxes, but ALL of the given GPU benchmarks we have from geekbench come from 18,2, which is likely to be the 24 core. Only other thing I can think of is that 18,2 could be the 14 inch with 32 core GPU, while the 18,4 is the 16 inch 32 core GPU. I'm rather doubtful on that, but we don't know yet.

I asked someone who found out the codes on twitter

https://twitter.com/i/web/status/1450193863587635204

It seems the model names dont reveal what gpu is used, just screen size and whether it's the M1 Pro or M1 Max.

So 18,1 is the 16" with the M1 Pro, 18,2 is the 16" with M1 Max, 18,3 is the 14" with M1 Pro and 18,4 is the 14" with M1 Max.

 

[ElfinHilon]

I asked someone who found out the codes on twitter

https://twitter.com/i/web/status/1450193863587635204

It seems the model names dont reveal what gpu is used, just screen size and whether it's the M1 Pro or M1 Max.

Interesting! That's fascinating. Thanks for sharing that!

EDIT: I'm also reading several reports from people that state that Geekbench shows full GPU core count, even on cores that have been disabled. This apparently happened quite commonly with the M1 upon lauch. This is getting spicy.

 

[JimmyjamesEU]

Apple just released a new RC of Monterey. From 21A558 --> 21A559. It's a long shot, but perhaps the incorrect core count was being reported! ?

 

[crazy dave]

It's also quite low due to having around 2/3rd's of the compute units of the 3080, of which, Apple directly compared it to.

Yes that what I’m referring to when I say Apple’s TBDR design gets away with less compute for the same graphics output. However they still quote how fast it is relative to their other chips and this doesn’t match it. So if accurate and is indeed the one on the 32 core something is wrong with:

A) Apple’s scaling
B) the test

If A) it can be that they lowered clocks to keep the GPU cooler. Which is pretty standard, but then marketing and their executives screwed up. Or something esoteric is going wrong, a bottleneck in their compute design.

 

[ElfinHilon]

Yes that what I’m referring to when I say Apple’s TBDR design gets away with less compute for the same graphics output. However they still quote how fast it is relative to their other chips and this doesn’t match it. So if accurate and is indeed the one on the 32 core something is wrong with:

A) Apple’s scaling
B) the test

If A) it can be that they lowered clocks to keep the GPU cooler. Which is pretty standard, but then marketing and their executives screwed up. Or something esoteric is going wrong, a bottleneck in their compute design.

Yeah, that's partially why I'm so hopeful that this is the 24 core and not the 32 core.

 

[ElfinHilon]

Apple just released a new RC of Monterey. From 21A558 --> 21A559. It's a long shot, but perhaps the incorrect core count was being reported! ?

What is this about? What's it mean? :O

 

[JimmyjamesEU]

What is this about? What's it mean? :O

It's just a joke. There is a new Release Candidate that has been released by Apple a few moments ago. Due to it being a small update (seemingly), I was joking that perhaps the os was misreporting how many gpu cores were active. To be clear, I dont think that's the case.

 

[ElfinHilon]

It's just a joke. There is a new Release Candidate that has been released by Apple a few moments ago. Due to it being a small update (seemingly), I was joking that perhaps the os was misreporting how many gpu cores were active. To be clear, I dont think that's the case.

Ah got it. At this point, unless we see another geekbench entry in there with a score between the 16 and "32" core, we are kind of at an impass. If we do see another score that's between those two though, I'm going to be quite sad, at least for the geekbench front.

 

[EPO75]

Yes that what I’m referring to when I say Apple’s TBDR design gets away with less compute for the same graphics output. However they still quote how fast it is relative to their other chips and this doesn’t match it. So if accurate and is indeed the one on the 32 core something is wrong with:

A) Apple’s scaling
B) the test

If A) it can be that they lowered clocks to keep the GPU cooler. Which is pretty standard, but then marketing and their executives screwed up. Or something esoteric is going wrong, a bottleneck in their compute design.

Funny part is 32x0,25 = 8 cores = 24 cores and the score matches what you would expect from a 24 cores higher clocked version. Ah well we see Monday..