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

[Serban55]

I believe it'll be as good as 3080 unless you are doing pure compute.

Maybe even better..on pure power for couple a seconds, maybe minutes my former Razer laptop with mobile 3080 from around 81k in benchmark settle down to around 56k
If this Mbp with their unified , power efficient SoC along side with a proper cooling system...they can perform over 60k all the time
Again , maybe Apple did that chart to show the performance under load for a longer period of time...no wonder the 3080 from MSI that can show the full potential is better with proper cooling system and the razer didnt

 

[diamond.g]

Do you have a good source for that, or is it personal speculation? I checked ULs benchmark descriptions and the general web and have seen no indication of this. Is my google-fu failing me?

Honest speculation, UL has no other reason to hide mobile results from the website result browser otherwise right? Which also explain why we do not see M1 Mac results (there is no official Mac version of this benchmark right?)

 

[EntropyQ3]

Honest speculation, UL has no other reason to hide mobile results from the website result browser otherwise right? Which also explain why we do not see M1 Mac results (there is no official Mac version of this benchmark right?)

No idea truthfully.
But they make no mention of such, nor does my naked eye see any visual difference between running the benchmark on my 12.9” iPad Pro and my PC. (The phone screen is too small to make any real statents about rendering quality.)
I can’t really see how the absense of a MacOS native version of the benchmark matters either way for this line of speculation. They don’t have that for their benchmark suite in general.

 

[diamond.g]

No idea truthfully.
But they make no mention of such, nor does my naked eye see any visual difference between running the benchmark on my 12.9” iPad Pro and my PC. (The phone screen is too small to make any real statents about rendering quality.)
I can’t really see how the absense of a MacOS native version of the benchmark matters either way for this line of speculation. They don’t have that for their benchmark suite in general.

I mean how easy is it to tell the difference between FP16 and FP32 rendering?
It was more for the concern that if you cannot search for Mac results, that we know they have, then you can't use the average/median results instead having to rely on single run variances.

 

[leman]

Honest speculation, UL has no other reason to hide mobile results from the website result browser otherwise right? Which also explain why we do not see M1 Mac results (there is no official Mac version of this benchmark right?)

Does the mobile benchmark app even have the option to upload the results?

 

[EntropyQ3]

I mean how easy is it to tell the difference between FP16 and FP32 rendering?

Properly applied it should be impossible.
It’s a totally valid way of reducing hardware demands for games rendering.

 

[diamond.g]

Does the mobile benchmark app even have the option to upload the results?

You can share the result, I assume UL gets all the results because otherwise how do they generate the comparison graphs. Sharing in this case isn't "the same" as uploading though.

Properly applied it should be impossible.
It’s a totally valid way of reducing hardware demands for games rendering.

Wild Life engine

3DMark Wild Life uses a cross-platform graphics engine optimized for mobile devices and lightweight notebooks. The engine was developed in-house with input from members of the UL Benchmark Development Program. The rendering, including scene upda...

support.benchmarks.ul.com

Their support page doesn't speak to the precision used. But it wouldn't be surprising, and would be interesting to see if Windows honors the lower precision (older GPUs don't support FP16) or it if just forces FP32 everywhere (which would imply the work performed isn't the same).

 

[JimmyjamesEU]

Strangely, the last 10 results on the Geekbench gpu database for the new MacBook Pros were all posted within 30 minutes. Perhaps someone doing multiple runs? Still waiting for some scores inbetween the 40,000 and 60,000 to see how a 24 core model will do.

 

[dugbug]

lots of 67-68K metal benchmark scores now. I'm reminded of just how bad my 2016 is with an AMD Radeon Pro 460 (16310).

Metal and OpenCL list:

macbookpro18 - Geekbench 5 Compute Search - Geekbench

-d

 

[diamond.g]

lots of 67-68K metal benchmark scores now. I'm reminded of just how bad my 2016 is with an AMD Radeon Pro 460 (16310).

Metal and OpenCL list:

macbookpro18 - Geekbench 5 Compute Search - Geekbench

-d

I have the same notebook, lol. I planned on upgrading my wife's and mine to Apple Silicon, early next year. I think I am going to keep the larger size, and will get an Air for my wife instead of keeping her in the pro line.

 

[hefeglass]

I was just looking at this, paused and screen captured:
View attachment 1870282
Searched the corner and found it was for a laptop with a 3050 TI, not the best discrete laptop GPUs. Went off googling 3050 TI vs 3080. So thought just about 1/2 the 3080 *roughly*. Sneaky, sneaky marketing graph. Then realised that's the pro, not the max!!! ??‍♂️? So yeah with the 32-core Max, not 16-core Pro it's gonna be ball-park. Nice!

(All with a good pinch of salt at least for now)

Maybe someone will finally make us some decent AAA games now it can match up with the graphics performance of a laptop that sounds like a vacuum cleaner! ?

and looks like one

 

[bcortens]

People should compare the sub-tests - there are a few tests that are dragging the M1 Max down, it looks like histogram equalization and face detection are only about doubled vs the M1 but most other things are between 4 and 3.5x faster than the M1.

M1 Mini vs M1 Max

 

[EugW]

Strangely, the last 10 results on the Geekbench gpu database for the new MacBook Pros were all posted within 30 minutes. Perhaps someone doing multiple runs? Still waiting for some scores inbetween the 40,000 and 60,000 to see how a 24 core model will do.

I wouldn't be surprised if none of the review units are 24-core. This is the launch of one of Apple's most important products in its history after all.

However, luckily we only have to wait a few extra days for the non-review units to show up (Tuesday).

 

[hefeglass]

If you look at the breakdown of tests in Geekbench you can see the incredible variance.

On the SFFT test the 5600M actually beats out the M1 Max with 593 Gflops to 585, also they're fairly close on other tests like the Gaussian Blur, Stereo Matching, and Histogram Equalisation tests.

But the M1 Max absolutely stomps it in Particle Physics by over 3x, and a bunch of other tests where the M1 Max is over twice as fast.

I still think there must be a reason why Apple compared the M1 Max to a 3080, and we're going to see that the Geekbench scores don't reflect real world performance.

ive noticed my m1 macbook pro running blender 3.0 alpha (arm version) rips through fluid and smoke simulations quite a bit faster than my ryzen 9 5900hs 3070 laptop. I ended up doing a side by side test with the same projects and its almost twice as fast in some instances. Rendering is obviously a different story..but not for long once 3.1 drops with metal backend
(although my tests with simulations are likely just testing the cpus)

 

[Pressure]

People should compare the sub-tests - there are a few tests that are dragging the M1 Max down, it looks like histogram equalization and face detection are only about doubled vs the M1 but most other things are between 4 and 3.5x faster than the M1.

M1 Mini vs M1 Max

You would normally use a neural engine for those two things ... which the M1 has

 

[JimmyjamesEU]

From here https://www.reddit.com/r/apple/comments/qdic01
more of a real world test. The Max seems to do pretty well

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

compared to the score for an Alienware with a 3080

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

 

[EugW]

You would normally use a neural engine for those two things ... which the M1 has

Yeah, one thing some people keep forgetting is that Apple has incorporated a ton of hardware accelerated features, etc. into their chips. The one example I bring up is the hardware ProRes and ProRes RAW acceleration. I'm no video editor, but from my understanding, for some workflows this will completely transform how MacBook Pros can be used in video editing. It's a really, really big deal, and arguably much more so than the raw general purpose CPU performance (which of course is also excellent).

 

[hefeglass]

Game over. 15 new scores have been posted. All below 70,000. I’d believe apple would give out some 24 core versions, but not all. These scores are for the 32…unfortunately.

Unfortunately? Is your day to day use case just running geekbench over and over? If that isnt the case, then i wouldnt be too upset about a benchmark number.

 

[Pressure]

From here https://www.reddit.com/r/apple/comments/qdic01
more of a real world test. The Max seems to do pretty well

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

compared to the score for an Alienware with a 3080

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

I don't know the subtests of the Puget benchmarks for Adobe Premiere Pro but that Alienware x17 R1 has a processor and GPU with a TDP of 165 Watt each.

Have fun running that on battery ?

 

[JimmyjamesEU]

Unfortunately? Is your day to day use case just running geekbench over and over? If that isnt the case, then i wouldnt be too upset about a benchmark number.

It’s unfortunate because it undermines Apple’s claims.

 

[EugW]

I don't know the subtests of the Puget benchmarks for Adobe Premiere Pro but that Alienware x17 R1 has a processor and GPU with a TDP of 165 Watt each.

The subtests are listed, with the individual scores.

One thing you'll notice is that the ProRes scores for Apple M1 Max are excellent... because it has hardware ProRes acceleration.

 

[mi7chy]

From here https://www.reddit.com/r/apple/comments/qdic01
more of a real world test. The Max seems to do pretty well

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

compared to the score for an Alienware with a 3080

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

www.pugetsystems.com

Good hardware video encoding performance but that's distinct from iGPU performance. Often people see, for example, 5x faster hardware video encoding/decoding performance and automatically assume iGPU or CPU performance is also 5x better but they're all distinct from each other.

 

[Pressure]

The subtests are listed, with the individual scores.

One thing you'll notice is that the ProRes scores for Apple M1 Max are excellent... because it has hardware ProRes acceleration.

Do you mean the following: Standard Export Score, Standard Live Playback Score, Effects Score and GPU Score?

They don't really say much though. What are we exporting? What effects are applied?

I'm sure they have an explanation of the subtests somewhere though

 

[JimmyjamesEU]

Good hardware video encoding performance but that's distinct from iGPU performance. Often people see, for example, 5x faster hardware video encoding/decoding performance and automatically assume iGPU or CPU performance is also 5x better but they're all distinct from each other.

The gpu score is a little less, the others are on par. Not sure what you’re looking at.

 

[leman]

It’s unfortunate because it undermines Apple’s claims.

Not really. Those machines are likely to perform much better on real world tasks. For video editing, even the base M1 manages to compare very favorably to much faster hardware.

I would still like to know what's the problem with Geekbench, and why Apple and AMD score lower than expected, while Nvidia is doing so well. Frankly, I start thinking that all of these benchmarks should post the kernel and setup code, because otherwise we have no idea what they are doing.