Shysters gonna shys. How's Intel's strategy of securing Apple as a foundry customer going?
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.
First M2 benchmarks
- Thread starter EugW
- Start date
- Sort by reaction score
wow that's insane .though i'm guessing we are talking about the 10c unbinned version right ?Geekbench 5 Metal scores: 29832
Mac14,7 - Geekbench
Benchmark results for a Mac14,7 with an Apple M2 processor.browser.geekbench.com
Gfxbench 5 results
GFXBench - Unified cross-platform 3D graphics benchmark database
The first unified cross-platform 3D graphics benchmark database for comparing Android, iOS, Windows 8, Windows Phone 8 and Windows RT capable devices based on graphics processing power.gfxbench.com
what the hell
i have a s20fe 5g and it's definitely not that fast i can tell .
but for real,how ?
In the fan powered 13' you'd basically get real close to the base 14.2 M1 Pro performance.
Sure, in relative terms. But GB5 underestimates the compute performance of Apple GPUs.
GB5 compute benchmarks are just too short. Apple GPUs have around 10ms warmup period, that’s how much time you need to be doing work to trigger high performance power state. But Gb5 tests are usually done quicker than that, meaning they execute in a low/medium power state with only up to 60-70% of possible speed. GB5 compute scores for all M1 models would probably be at least 40-50% higher if the benchmark did a proper warmup.
Just from the numbers. Efficiency = performance/energy used. Increase performance without increasing the used power and the efficiency goes up. Basically, you use the same amount of d eggy to do more work.
Some posters in this thread seem to think that for the efficiency to go up, the energy usage has to go down, which is a very weird stance given the fact that Apples energy usage is already ridiculously low. I mean, their performance cores use less power than Intels efficiency cores while performing better than Intel performance cores at your time higher energy usage.
I wonder if John Poole(?) and the rest of the folks over at Geekbench would be amenable to doing this?
Yes. It’s the 13” MacBook Pro which doesn’t offer the lower binned GPU.
For Geekbench 5 Metal on Apple Silicon, all I’m interested in here is relative performance.
I doubt you will see any big jumps. I'm just not sure if benchmarks are even worth anything at this point.
Last edited:
I remember seeing a tweet about an upcoming GB update bringing longer benchmark duration options from John himself.
Slight but important nitpick, the equation is actually:
efficiency = work done / energy used
If you want to base this calculation on a performance number, since performance is a rate unit (work done per unit time), you need to divide it by another rate unit, energy used per unit time (AKA power). The per-time units cancel out leaving you with a ratio that can be compared with work done per unit energy ratios.
Fully agree that if you use the same energy but get more work done for that energy expenditure, you've improved efficiency. This means efficiency can go up even when power also goes up; with good power management what matters is how many Joules got burned doing the work the user needed to be done. More watts for fewer seconds can be less joules total if the number of seconds goes down enough.
In addition to the reports that he's expressed interest, I'd add that for a program like Geekbench, changing methodology is fraught with difficulty. How do you avoid invalidating all test results collected so far? Every so often GB does a big step change which invalidates comparisons, and that's when Poole bumps the major version number - you can't directly compare GB4 and GB5 scores. But you are supposed to be able to compare all GB5 scores to each other, regardless of the minor GB5 version number.
Fixing the runtime issue for GPU compute tests in GB5 should be trivial, but they're going to have to take a lot of care to make sure it still produces scores worth comparing to earlier GB5 GPU compute scores.
RE: these GPU scores
By way of comparison, I have a 2018 mac mini running the BlackMagic RX 580 eGPU (only Apple endorsed egpu other than its discontinued pro counterpart) This machine scores ~34300 on metal.
So, presumably the new Air can almost perform at the level of the base BM eGPU at least in bursts for general compute.
Does that sound right?
I suspect the optimizations for common graphics workflows make the base M2 more useful overall though.
By way of comparison, I have a 2018 mac mini running the BlackMagic RX 580 eGPU (only Apple endorsed egpu other than its discontinued pro counterpart) This machine scores ~34300 on metal.
So, presumably the new Air can almost perform at the level of the base BM eGPU at least in bursts for general compute.
Does that sound right?
I suspect the optimizations for common graphics workflows make the base M2 more useful overall though.
That’s nuts!
ProRes benchmarks are gonna rock.
Cuz M2 gains hardware 8K ProRes acceleration.
In the end, the M2 will be used mostly in the MacBook Air where it thermal throttles. So these benchmarks are “cute” so to say.
It is my understanding that the M1 MBA only thermal throttles on longer, heavy use as happens in some benchmarks, exporting videos, gaming, etc.. For people who do that then that is important, of course. For people like me who don't do those things and instead tend to do stuff with occasional spikes of heavy use such as running Topaz Denoise AI or Sharpen AI on a single photo, exporting a single photo from Lightroom, etc. then my guess is that thermal throttling would rarely or maybe never happen.
It will also be used in the Mac mini and also the MBP 13".
I will buy an M2 Mac mini. I'm pretty sure it will have a fan. Plus the second most popular Mac with M2 will be the 13" MacBook Pro, and that has a fan.
EDIT:
@exoticSpice beat me to it.
The thermal throttling in the M1 Air only happens after several minutes of sustained load. The M2 Air will very likely be a “bursty” champion at least, and a mid range workhorse unless they have sorely missed the mark.
Plus even if it does throttle, it will still be hella fast.
The results I got with my fanless m3 MacBook aren't quite the same thing, but it does give you an idea:
This is how the 2017 MacBook Core m3, i5, i7 behave during sustained CPU load: 10+ runs of Cinebench
As I suspected, the table material makes a difference. The first time I ran the test on my 2017 MacBook Core m3, I ran it on a granite counter, and turned off WiFi and Bluetooth. I just ran it again, but this time ran it on a wood table (which would act more as an insulator, keeping the heat...
forums.macrumors.com