Geekbench METAL scores

[TripleYoThreat]

Just for reference: MacBook Pro M1 Max - Geekbench 5 Metal score:

View attachment 1873070

MacBookPro18,2 - Geekbench

Benchmark results for a MacBookPro18,2 with an Apple M1 Max processor.

browser.geekbench.com

Did you test this yourself?

We're saying better than the Radeon VII and 5700 XT?! That's outrageous.
I've felt this since the benchmarks for the initial M1 came out too; I believe Apple is doing some faking or spoofing when benchmarks are running.
Anyone remember when many years back, I believe Qualcomm was caught doing the same thing? When benchmarks ran, it kicked on some super turbo mode (or something to that effect) and it was achieving crazy scores in benchmarks. There was a lot of talk about it - some said, well hey, it still reached that score, even if by some super turbo mode. Others said hey, it's not engaging this mode at any other time, so we would never see/use such power in our usage.

I remember seeing that video of someone at the Apple Store who literally opened all stock applications on the M1 and they popped right up. I was pretty impressed. I bought an M1 air a month or two ago and it was nothing like that lol. It was great, but not a game-changer. It was effective, and the battery life was definitely bar none. Solid build, and all that. But I didn't notice crazy processing speeds as I'd expect. Everything loaded and ran just about the same as my 2015 MBP. The minute difference was not enough to justify the purchase and I returned it. It was Apple refurbished so I would hope, like sincerely hope, that played a part in it not being all that. But I don't think so.
I would click an app, and it would open in a few seconds, just around the time the bounce animation on the dock finished - just like my 2015 MBP, my brand new 11700k custom Mac, and even my custom Mac with a 4690k.

Thoughts?

 

[EugW]

Did you test this yourself?

We're saying better than the Radeon VII and 5700 XT?! That's outrageous.
I've felt this since the benchmarks for the initial M1 came out too; I believe Apple is doing some faking or spoofing when benchmarks are running.
Anyone remember when many years back, I believe Qualcomm was caught doing the same thing? When benchmarks ran, it kicked on some super turbo mode (or something to that effect) and it was achieving crazy scores in benchmarks. There was a lot of talk about it - some said, well hey, it still reached that score, even if by some super turbo mode. Others said hey, it's not engaging this mode at any other time, so we would never see/use such power in our usage.

I remember seeing that video of someone at the Apple Store who literally opened all stock applications on the M1 and they popped right up. I was pretty impressed. I bought an M1 air a month or two ago and it was nothing like that lol. It was great, but not a game-changer. It was effective, and the battery life was definitely bar none. Solid build, and all that. But I didn't notice crazy processing speeds as I'd expect. Everything loaded and ran just about the same as my 2015 MBP. The minute difference was not enough to justify the purchase and I returned it. It was Apple refurbished so I would hope, like sincerely hope, that played a part in it not being all that. But I don't think so.
I would click an app, and it would open in a few seconds, just around the time the bounce animation on the dock finished - just like my 2015 MBP, my brand new 11700k custom Mac, and even my custom Mac with a 4690k.

Thoughts?

No I didn't test this myself. It's on their database now, and I'm 100% sure nothing is faked, although Geekbench isn't necessarily a good test to assess real-world application performance here.

The reviewers got their units recently. They are not supposed to publish benchmarks, but of course they're going to run them anyway, but every single release somebody forgets to turn off auto-publishing of those scores. But I'm sure Apple is OK with it, because it gets people talking.

BTW, that app loading thing is true with fresh installs. It's not so true once the drive gets gunked up with real applications and data. I have sometimes fresh installed macOS on my Intel iMac when doing major OS upgrades, just to speed things up. But it inevitably slows back down so these days I usually don't bother.

I had heard a rumour that the Apple Store regularly does a fresh OS install on the display units. Or maybe a reset or something.

 

[TripleYoThreat]

No I didn't test this myself. It's on their database now, and I'm 100% sure nothing is faked, although Geekbench isn't necessarily a good test to assess real-world application performance here.

The reviewers got their units recently. They are not supposed to publish benchmarks, but of course they're going to run them anyway, but every single release somebody forgets to turn off automatic benchmark publishing.

BTW, that app loading thing is true with fresh installs. It's not so true once the drive gets gunked up with real applications and data.

100% sure? Isn't it possible that the chip could recognize the code of a benchmark and perform differently or operate in a different manner? Just like hash rate limiters on the new Nvidia cards - when a certain algorithm is detected, it could reduce or improve performance.

Also which app loading thing exactly? Even my 2015 mbp on mojave opened apps nearly as fast as the M1 air.

For the M1 Pro GPU to beat the Radeon VII....
Ah you know what, I get it. This is metal, it's written by Apple. Of course it'll beat others at their own game.

Though if premiere and fcpx speak metal, guess that's all that matters.

I wonder how it would score on Bruce X

 

[h9826790]

100% sure? Isn't it possible that the chip could recognize the code of a benchmark and perform differently or operate in a different manner? Just like hash rate limiters on the new Nvidia cards - when a certain algorithm is detected, it could reduce or improve performance.

Also which app loading thing exactly? Even my 2015 mbp on mojave opened apps nearly as fast as the M1 air.

For the M1 Pro GPU to beat the Radeon VII....
Ah you know what, I get it. This is metal, it's written by Apple. Of course it'll beat others at their own game.

Though if premiere and fcpx speak metal, guess that's all that matters.

I wonder how it would score on Bruce X

The Radeon VII can't perform in GB5 tests. This is Apple's problem, they intentionally skipped SMU Firmware loading for the Radeon VII in the driver.

CMMChris developed RadeonBoost to fix that, however, that fix doesn't work in Big Sur anymore.

I showed the difference in this post

Tired of low Geekbench scores? Use RadeonBoost!

This is the detail scores comparison from my Radeon VII. Apparently Something like Sobel can benefit from this kext a lot. But Stereo Matching perform pretty much the same. And NO, this kext shouldn't infringe the PowerPlay kext for downvolting. I checked that on my cMP.

forums.macrumors.com

My Radeon VII can achieve almost 100,000 scores with the SMU firmware (via Radeon Boost). This is way above the M1 Max can do.

So, the Radeon VII can do 13.4 TFLOPS, ~100,000 in GB5

M1 Max has ~10 TFLOPS, can do ~70,000 in GB5

Both shows the M1 Max is just about 70% of a Radeon VII's performance. The numbers match (when the software works without specific artificial restriction).

I haven't seen the Luxmark result yet, but it expect that can score 35000.

IMO, the M1 Max isn't something has super power, but definitely a very nice laptop (if the software works).

 

[TripleYoThreat]

The Radeon VII can't perform in GB5 tests. This is Apple's problem, they intentionally skipped SMU Firmware loading for the Radeon VII in the driver.

CMMChris developed RadeonBoost to fix that, however, that fix doesn't work in Big Sur anymore.

I showed the difference in this post

Tired of low Geekbench scores? Use RadeonBoost!

This is the detail scores comparison from my Radeon VII. Apparently Something like Sobel can benefit from this kext a lot. But Stereo Matching perform pretty much the same. And NO, this kext shouldn't infringe the PowerPlay kext for downvolting. I checked that on my cMP.

forums.macrumors.com

My Radeon VII can achieve almost 100,000 scores with the SMU firmware (via Radeon Boost). This is way above the M1 Max can do.

So, the Radeon VII can do 13.4 TFLOPS, ~100,000 in GB5

M1 Max has ~10 TFLOPS, can do ~70,000 in GB5

Both shows the M1 Max is just about 70% of a Radeon VII's performance. The numbers match (when the software works without specific artificial restriction).

I haven't seen the Luxmark result yet, but it expect that can score 35000.

IMO, the M1 Max isn't something has super power, but definitely a very nice laptop (if the software works).

Thank you for this insight, very helpful! On the other hand, it matching the 5700 XT is still outrageous. It's performing at about the level of a Pro Vega 56, which is somehow better than the Pro Vega 64. That doesn't make any sense lol. There's good reason why I find these benchmarks to be very skewed and arbitrary

 

[randy85]

If anyone's interested here's Metal score for my Gigabyte 6800 XT vs my W5700X.

I'm definitely interested in an M1 Max Macbook Pro for working away. It'll be a huge upgrade to my 2016 MBP.

There's probably going to be another wave of hype saying M1 Max kills the Mac Pro or whatever, but it'll probably still make sense to use both if you can. Do the real heavy lifting on the Mac Pro. A lot of the Youtube crowd only use mirrorless cameras which skew their perception of how good the M1 machine are.

 

[EugW]

There's probably going to be another wave of hype saying M1 Max kills the Mac Pro or whatever, but it'll probably still make sense to use both if you can. Do the real heavy lifting on the Mac Pro. A lot of the Youtube crowd only use mirrorless cameras which skew their perception of how good the M1 machine are.

Depends on the workflow, but it should be noted that M1 Pro and M1 Max both have hardware ProRes and ProRes RAW acceleration in hardware, including 8K support.

 

[randy85]

Depends on the workflow, but it should be noted that M1 Pro and M1 Max both have hardware ProRes and ProRes RAW acceleration in hardware, including 8K support.

That is very exciting. Also the software optimisation for AS will make it a good experience I'm sure.

I suppose raw GPU performance is still an advantage for things like noise reduction and so on.