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.
So M1 Max with 32 Core GPU matches RTX 3080 Mobile.
- Thread starter iBug2
- Start date
- Sort by reaction score
Because GeekBench isn't testing rasterization, it is testing compute. The results will differ.
If you sort by GPU versus by device the extras go away. Then you get the "best score" for the card by the "best api". Which is weird because the 6900XT tops the list if you do it by device but the 3090 tops if you do it by GPU.
Also the main money makers are console games which all the new ones support ray tracing. The pc games are ports.
Okay I am not sure how I feel about this. I ran a quick test on my 6900xt and got vastly different scores depending on API chosen.
I wouldn't say the 3080 Laptop scores are suspect, but they could be missing results (if you sort it appears the Vulkan and DX12 tests were not ran).
I wouldn't say the 3080 Laptop scores are suspect, but they could be missing results (if you sort it appears the Vulkan and DX12 tests were not ran).
That's how Fusion's been doing since the days of eyeon, the original developers. It needs as much as RAM as you can give it to it if you want real time or close to real time playback. Of course it all depends on complexity of the comp but add some particles, some 3D render to 2D and post processes such as motion blur, aberration, shake and my private PC with 128GB of ram has over 75% eaten up by the app. More and more tools now are GPU accelerated and app defaults to GPU even though it says Auto in the options. Maybe hard switching tools to CPU in Preferences will help the RAM issue.
API overhead and implementation details have a non-trivial impact at these high framerates. That’s also why GFXbench is not very good, it’s simply not demanding enough. But it can be used to approximate things to a limited degree. It’s best to compare the best score to the best score.
I’m not sure I understand the last comment about “appearing to not having been run” but the first part is expected.
GPU benchmarks like these test: how optimally the benchmark was coded in the API; how optimally the driver for the API was written for the hardware; and the underlying hardware.
Creating tests for just the last part is almost impossible. So you just accept that there’s going to be more variation across different tests and more variables underlying the result. And, ultimately, when you are using the GPU all three are indeed what you care about: how well the program was coded for the API, how well does the API run on your GPU, and how good is the hardware at that specific task.
So I've stated this before, but I didn't provide any evidence.
Here's the 16core GPU compared to M1:
MacBookPro18,3 (16-Core GPU) vs Mac mini (Late 2020)
Here's the supposed 32core GPU compared to M1:
MacBookPro18,2 ("32"-core GPU) vs Mac mini (Late 2020)
Seeing these results, I do believe that Geekbench is reporting this as an error in the GPU core, when the reality is that the GPU being tested is indeed the 24core GPU.
Here's the 16core GPU compared to M1:
MacBookPro18,3 (16-Core GPU) vs Mac mini (Late 2020)
Here's the supposed 32core GPU compared to M1:
MacBookPro18,2 ("32"-core GPU) vs Mac mini (Late 2020)
Seeing these results, I do believe that Geekbench is reporting this as an error in the GPU core, when the reality is that the GPU being tested is indeed the 24core GPU.
Last edited:
See my post directly above this. I genuinely suspect we are looking at the 24 core GPU option and geekbench is reporting incorrectly.
Even further, we know that the 5600M is supposed to be about par with the 16 core GPU. I guess we will have to wait and see what happens with the benchmarks. This is only one benchmark, so I wouldn't hold too much wait in it, but the more and more I look into figures directly provided by Apple, I think this is the 24 core GPU just being miss-labeled in Geekbench.
EDIT: Alternatively, we could be actually looking at the 32core GPU, but it just doesn't scale well for the M1 Max due the depreciated graphics API being used here. That's also entirely possible. However, if we are indeed looking at the 24core here, oh boy this is gonna be a wild ride.
Last edited:
I say that because on the score browser the 3080 laptop gpu is missing Vulcan/DX12 scores just like the 6900XT is missing those scores as well. It shows up as failed/not supported which isn't true (clearly I ran the test).
A14 only does half FP32 rate, M1 has full FP32 enabled.
GFXBench is more garbage than Geekbench. Still prefer real world workload and games.
It’s curious. The M1 score is 19000. Divided by 8 equals around 2400 approximately. The 16 core yields 38000. Again around 2400 per core. one Would think the 32 core would be around 72000 opencl score. 60000 must be the 24 core.
That may be the case if the M1 Max was about 1.5x faster than the M1 Pro across the various subtests. But the M1 Max is twice faster in some, and not even faster in others.
The plot thickens.
EDIT: some have suggested that the M1 Max could be hitting some power/thermal limit in that test, if it runs on battery or if some macOS setting limits power/fan speed.
Last edited:
Yeah, that's what I am personally leaning towards. I'm confused as to why the 32core would see such terrible scaling in relation to the rest.
Yep, good point. I'm honestly thinking we are looking at the 24 core here. It's possible this is also just extremely poor scaling for OpenCL or something.
My guess would be something with the different core complexes or another. I'm not an engineer, so I'm just making educated guesses here. All I know, is that OpenCL doesn't do too great for Macs.