Apple's M1 Max GPU is at least 3x faster than M1, Metal benchmark shows

Posted:
in Current Mac Hardware edited October 20
Supposed M1 Max benchmark results reveal the chip's GPU is at least three times faster than the first-generation M1 included in 2020's 13-inch MacBook Pro.

M1 Max


An unconfirmed Geekbench post on Wednesday shows an M1 Max with 64GB of unified memory clocking a Metal score of 68870. That result compares to an average score of roughly 21800 for the 13-inch MacBook Pro with M1 chip.

The M1 Max also smashes Metal scores logged for AMD's Radeon Pro 5600M, the most performant GPU offered with last year's 16-inch MacBook Pro, outpacing the former MacBook torch bearer by 62%, according to Geekbench.

It is not clear what M1 Max configuration is represented in the Geekbench Compute post, but Apple during its "Unleashed" event on Monday touted maximum performance on par with the discrete GPU in a high-end PC laptop. Apple offers the M1 Max in 24- and 32-core GPU configurations, both of which can be accompanied by up to 64GB of unified RAM.

As noted by developer Steve Troughton-Smith, the score posted today falls short of the more than 90000 points achieved by the laptop version of Nvidia's class-leading RTX 3080, suggesting they belong to an M1 Max with 24-core GPU.

Today's Metal score arrives two days after the first single- and multi-core CPU score hit the Geekbench Browser, which illustrated a 50% increase over the 8-core M1. Subsequent tests have been posted and corroborate the initial results.

The M1 Max joins the M1 Pro as Apple's new "pro" level silicon designs. The M1 Pro features an 8- or 10-core CPU and a 14- or 16-core GPU, while the M1 Max comes with a 10-core CPU and a 24- or 32-core GPU. Most configurations are available on both 14- and 16-inch MacBook Pro models.

Read on AppleInsider

Comments

  • Reply 1 of 16
    fastasleepfastasleep Posts: 5,843member
    Oh weird, didn't realize there was a 24 core option. I'm super interested in seeing the real world tests this coming Monday!
    williamlondonrepressthiswatto_cobra
  • Reply 2 of 16
    (...) I'm super interested in seeing the real world tests this coming Monday!

    mee too! it'll be also interesting to see them repeated a couple of months later as these will only be the first flutters
    edited October 21 watto_cobra
  • Reply 3 of 16
    dk49dk49 Posts: 185member
    What about single core performance? I am sure the difference won't be that much between Max and the M1.
    nadriel
  • Reply 4 of 16
    I did comparison with some Core i9 11 generation and it is interesting to see some subtests differences in single and multicore. It seems sometimes as Intel would run multicore on 2 or so cores max or is very inefficient.
    Especially AES-XTS, Machine Learnin, Speech Recognition
    The same is for Apple with Text Compression.






    magman1979
  • Reply 5 of 16
    dk49 said:
    What about single core performance? I am sure the difference won't be that much between Max and the M1.
    For CPU yes - the single core CPU score will be around the same as M1.

    Metal is GPU not CPU. 

    These new chips are better at multi-core for both CPU and GPU.

    williamlondonrepressthiswatto_cobra
  • Reply 6 of 16
    I’d really like one of these outlets to show the performance gains of the baseline 14-inch to get a reference of how much faster the binned chips are compared to the M1.
    watto_cobra
  • Reply 7 of 16
    KITAKITA Posts: 382member
    The Apple M1 scores ~18,000 in OpenCL (1) and ~20,000 in Metal (2) on Geekbench.

    The Macbook Pro with M1 Max 32 core scored ~60,000 in OpenCL (3) on Geekbench. The ~68,000 Metal score above is likely the 32 core version as well.

    The RTX 3080 scores much higher than 90,000. A 130W version of the RTX 3080 in a laptop scores ~120,000 in OpenCL (4) on Geekbench.

    Apple used a 165W version of the RTX 3080 in their comparison:


    muthuk_vanalingamelijahgmagman1979nadrielOutdoorAppDeveloperwilliamlondonnetroxrepressthis
  • Reply 8 of 16
    Interesting. The Geekbench 5.4.1 result for my 2017 iMac Pro with a Radeon Pro Vega 64 is 61440.  So if this benchmark result is legit, it outperforms my iMac Pro in both CPU and GPU.  It's too bad they did not also announce a M1 Max Mac Mini along with the laptops.  I think it will still be a while until they release a Pro desktop (either iMac or headless) pro model.
    vedelppawatto_cobra
  • Reply 9 of 16
    keithw said:
    (...) It's too bad they did not also announce a M1 Max Mac Mini along with the laptops.  I think it will still be a while until they release a Pro desktop (either iMac or headless) pro model.

    I feel the same :'( :'( :'( :'( :'( :'( Now I have to spend money on a Mac Mini I will never like but will use for a couple of years :neutral: 
    edited October 21 keithwwilliamlondonwatto_cobra
  • Reply 10 of 16
    mjtomlinmjtomlin Posts: 2,490member
    KITA said:
    The Apple M1 scores ~18,000 in OpenCL (1) and ~20,000 in Metal (2) on Geekbench.

    The Macbook Pro with M1 Max 32 core scored ~60,000 in OpenCL (3) on Geekbench. The ~68,000 Metal score above is likely the 32 core version as well.

    The RTX 3080 scores much higher than 90,000. A 130W version of the RTX 3080 in a laptop scores ~120,000 in OpenCL (4) on Geekbench.

    Apple used a 165W version of the RTX 3080 in their comparison:




    Apple stopped supporting and optimizing OpenCL (and OpenGL) on Macs a few years ago when they started designing their own GPU's optimized for their Metal APIs. I'm not sure those are good benchmark comparisons unless you're comparing "like" systems.
    tmaymagman1979OutdoorAppDeveloperwilliamlondonwatto_cobra
  • Reply 11 of 16
    XedXed Posts: 1,113member
    Oh weird, didn't realize there was a 24 core option. I'm super interested in seeing the real world tests this coming Monday!
    The M1Max looks like the M1Pro from the 16-core GPU up. The Max adds 16-cores to the GPU,along with with doubling the SLC and memory, as that looks like a copy of that half of the SoC.

    it’s very impressive how Apple designed the chip to be scalable in this way.

    my guess is the 24-core will still the other 8 cores but they’ll be disabled.
    watto_cobra
  • Reply 12 of 16
    ppietrappietra Posts: 278member
    KITA said:
    The Apple M1 scores ~18,000 in OpenCL (1) and ~20,000 in Metal (2) on Geekbench.

    The Macbook Pro with M1 Max 32 core scored ~60,000 in OpenCL (3) on Geekbench. The ~68,000 Metal score above is likely the 32 core version as well.

    The RTX 3080 scores much higher than 90,000. A 130W version of the RTX 3080 in a laptop scores ~120,000 in OpenCL (4) on Geekbench.

    Apple used a 165W version of the RTX 3080 in their comparison:


    This is an individual score so you shouldn’t use an average value as reference because averages are brought down by people running the benchmark while other things are also running.  Also the M1 score in that table is brought down because there is a version of the M1 with just 7 cores.
    A more regular OpenCL score should be around 19,500, and that means a ratio of 3.08 vs the M1 Max score,
    Also, M1 Pro OpenCL score is 38,400, and that means a ratio of 2. 
    These two results makes it extremely unlikely that the M1 Max score is for a 32-core GPU.
    watto_cobra
  • Reply 13 of 16
    mjtomlinmjtomlin Posts: 2,490member
    KITA said:
    The Apple M1 scores ~18,000 in OpenCL (1) and ~20,000 in Metal (2) on Geekbench.

    The Macbook Pro with M1 Max 32 core scored ~60,000 in OpenCL (3) on Geekbench. The ~68,000 Metal score above is likely the 32 core version as well.


    Also...

    There's a MacBook Pro 16-core that scored ~38,000 in OpenCL...

    8 - 18,000
    16 - 38,000
    32 - 60,000

    Something is way off with that last score...

    8 -> 16, 100% gain
    16 -> 32, 50% gain

    Me thinks that's a 24-core Max, not a 32-core.

    Some have suggested that Geekbench is not showing proper statistics...

    Every M1 MacBookAir shows 8 compute units, even though there are models with only 7 units.
    Every M1/Pro/Max shows maximum GPU frequency being 1000MHz, we know it's 1278MHz for the M1, and is supposedly a little higher for Pro and Max.

    So that 32 compute units might not be accurate.
    edited October 21 williamlondonwatto_cobra
  • Reply 14 of 16
    Laptop GPU performance scores should be viewed with suspicion regardless of whether you are looking at a laptop with a RTX 3080 or a M1 Max. Apple highlighted the issue in their presentation. That heavy, hot, oversized $5000 Windows laptop with a top end desktop GPU may run fast when plugged into a wall socket (sometimes with two separate power supplies taking up two outlets) but on battery it runs extremely slowly and not for long. Being able to use the M1 Max at full speed for hours on battery while still getting near desktop performance is a game changer for mobile pros.

    Too bad not many people are traveling these days. The thing I hear over and over again (and in my own case) is that these new MacBook Pros would be instant buys except that there is no need for them currently.
    GG1jas99docno42watto_cobra
  • Reply 15 of 16
    canukstormcanukstorm Posts: 2,557member
    Oh weird, didn't realize there was a 24 core option. I'm super interested in seeing the real world tests this coming Monday!
    There's also a 14-core GPU option
    watto_cobra
  • Reply 16 of 16
    BeatsBeats Posts: 2,646member
    Did the M1 MacBooks get a price drop?
    watto_cobra
Sign In or Register to comment.