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
doesnt explain the obsession over ONE benchmark in this thread
clearly m1 max owners are very well endowed if you look at many of the other benchmarks.
It is odd that GB has no issue running Metal code with AMD GPU's. You would think it would be the same, unless they hand tuned the Metal code path for AMD (which seems odd).
its not all we have...what about all the other benchmarks that show a clear difference. You guys are obsessing over geekbench . I have read this entire thread and the two main people very upset about these score seem to be about 12-14 years old by the posts and questions they ask. I have noticed that the people with actual knowledge dont seem too worked up over a geekbench number.
Monday, but don't count on benchmarks.*
*Often Apple refuses to allow benchmarks in its reviews from people it gives the review units to. I don't know if this time it will be any different.
Actually my understanding is ProRes export is typically a GPU task that Apple made dedicated hardware for, so at least that’s nixed.
I felt like an idiot because I forgot to write AI and was editing the post when you posted your follow up question. Neural nets are often accelerated on GPUs, though Apple has dedicated hardware for that and AMX blocks on the CPU. Not sure how it compares, pros and cons to running the algorithm on a modern Nvidia GPU with tensor cores. Unfortunately a lot of open source AI projects are only just being ported to Apple’s framework and there aren’t many great standardized benchmarks for NN.
A lot of scientific modeling and simulation are done on GPUs, you can see some of them actually in the GB subtests. While some such applications require 64-bit levels of precision found on pro GPUs, others do not, 32-bit like what is found on consumer and prosumer hardware is fine. Of course that segment is pretty much entirely captured by CUDA and Nvidia GPUs. Metal could do it, but Nvidia has a massive head start that even other cross platform APIs that are also geared towards hardware in clusters (which Apple is unlikely to ever compete in) and workstations (which Apple has historically only had one product, the Mac Pro) will find it hard to compete without a massive advantage.
A big one for Apple however is rendering a scene for animation. Such workloads tend to stress compute and memory heavily and that’s a segment Apple takes very seriously.
Dear Lord, if you are this upset about what interests other people then you need to relax. The majority of benchmark results are from geekbench, and they are frequently used by reviewers to judge general performance. It's why there are 23 results currently. Your angry ad hominem attacks aren't warranted or welcome. "12-14 years old". What a ridiculous thing to say. I haven't ever claimed to have the most knowledge, and if those that do don't worry about geekbench then great. I won't be policed by you. If I'm within the rules then I'm entitled to post about what interests me. Don't like it, don't engage.
Interesting, thank you.Actually my understanding is ProRes export is typically a GPU task that Apple made dedicated hardware for, so at least that’s nixed.
I felt like an idiot because I forgot to write AI and was editing the post when you posted your follow up question. Neural nets are often accelerated on GPUs, though Apple has dedicated hardware for that and AMX blocks on the CPU. Not sure how it compares, pros and cons to running the algorithm on a modern Nvidia GPU with tensor cores. Unfortunately a lot of open source AI projects are only just being ported to Apple’s framework and there aren’t many great standardized benchmarks for NN.
A lot of scientific modeling and simulation are done on GPUs, you can see some of them actually in the GB subtests. While some such applications require 64-bit levels of precision found on pro GPUs, others do not, 32-bit like what is found on consumer and prosumer hardware is fine. Of course that segment is pretty much entirely captured by CUDA and Nvidia GPUs. Metal could do it, but Nvidia has a massive head start that even other cross platform APIs that are also geared towards hardware in clusters (which Apple is unlikely to ever compete in) and workstations (which Apple has historically only had one product, the Mac Pro) will find it hard to compete without a massive advantage.
A big one for Apple however is rendering a scene for animation. Such workloads tend to stress compute and memory heavily and that’s a segment Apple takes very seriously.
I can't ever remember reviews omitting benchmarks from Macworld, Gruber, Arstechnica, Anandtech. Do you have any references for this?
Ars and AnandTech don't get review units. Neither does Gruber AFAIK.
Those reviews come later, after they have purchased their own units. Dunno about Macworld.
Luckily the retail units arrive Tuesday, so if we don't get a full set of benchmarks on Monday, the benchmarks will come Tuesday anyway. BTW, for AnandTech, we sometimes don't see reviews until several days or even weeks later.
Geekbench is not the only compute heavy task that shows weird scaling issues. Again this is not a case of, the GPU is suddenly bad, oh noes!, but why aren’t the workloads scaling like they should. Could some of it be a GB issue? Sure! For instance in the land of weird scaling, in Windows the Geekbench CPU benchmark suffers from a weird scaling issue for high core count CPUs. Something that is acknowledged and is promised to be fixed. And that’s for Windows a rather common platform. Unlike the M1 GPU which is a less mature platform for most applications, which means other tests like Redshift are still being optimized.
Could it be that these benchmarks are being run without high power mode on? Did Apple quote some numbers with the mode on and off? Maybe, maybe not.
People are just interested as to why these results are the way they are. For the most part, it’s just fun speculation with yes a tinge of disappointment that some of the numbers don’t line up as expected. Obviously we’ll find out when reviewers can not only share the benchmarks but also put them context.
Last edited:
Gruber absolutely does get review units. As does Macworld, Sixcolors etc. They all run benchmarks. Can you point me to any who you believe do get review units and are told not to benchmark?
"i wont be policed by you"..??
calm down there guy..i never told you what or what not to do. Nobody attacked you..
I pointed out the fact that apples claims arent disproven by geekbench results.
I love how my pointing out a fact that you got incorrect has now devolved into "I wont be policed by you"
smh
You're right. I stand corrected. In previous years they were left out in the cold.
Perhaps it's because M1 was a new architecture, and they knew that few could properly provide a technical review like AnandTech. I guess we can hope Apple will allow the same for M1 Max.
Yes, the YouTubers for example who get units who are allowed to publish in advance of the official release date are often told not to post any benchmarks, so you'll see them gush about colours and smooth video scrubbing but not actually provide of list of performance measures. I remember one even saying that he couldn't.
Don't Rene Ritchie, MKBHD and iJustine etc publish benchmarks? I could swear they do.
Did Rene Ritchie even get a review unit last time? I don't know about the M1 launch, but for previous releases I know he was left out.
iJustine did but it looks like she published the same day as AnandTech, which was the day of release.
You came into a thread saying everyone in here was acting like 12-14 year olds. I'm not sure how you can spin that as not an attack. We are discussing theories on why the scaling may not be in line with what we expected it to be. Is it rather pointless discussion when we will know the results in 3 days? Yeah, kinda. Is it fun? Absolutely it is.
That’s true, typically Anandtech at least has to purchase Apple hardware. I suspect that they’ll get review units though simply because as you said for the M1, new SOCs in a new device that they’re trying pump up, they’re going to want it reviewed by them. But we won’t know for sure until the embargo lifts.