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
Excluding the iMac Pro, every Intel version of the iMac since it has been released have used mobile CPU chips. The slim form factor of the machine meant that it was more practical and cost effective to design using mobile CPUs than full size desktop CPUs... as such Mac Desktops have been crippled for a very long time (which is why professionals demanded pro machines, which Apple delivered but not well).
Now with the M1 Pro/Max, I am not sure where Apple is going to go with the larger iMac or a Mac Pro replacement.
Mac Mini's are always going to remain entry-level Mac budget desktops or low-cost server machines, and so they will keep with the similar chips used in MacBook Air and MacBook Pros. Mac Mini can still be used for server farms and cloud computing devices without beefing up their CPU/GPU.
This is something people are really convinced of, but it's untrue. Have a look at Everymac's long list of iMac models/CPUs: https://everymac.com/systems/apple/imac/index-imac.html
The earliest Intel iMacs did use mobile segment chips, but in 2009, when Apple switched to i3/i5/i7 CPUs, they switched almost exclusively to desktop segment chips. (I'm using Intel's definition of segment here - if you look one of their chips up on ark.intel.com, it will usually have a line for "Vertical Segment" in the specs, and that can be "Mobile" or "Desktop".)
Today you can order an i7-10700K or i9-10910 in the 2020 revision of the 27" iMac. They're slightly restricted when installed in the iMac because Apple's non-Pro iMac cooling system is limited to about 85-90W iirc, but other than that they're standard high spec desktop CPUs.
Rumors have the forthcoming new 27" iMac with the same M1 Pro / M1 Max SoCs, and the Mac Pro (and Mac Pro Cube) with a choice of dual or quad M1 Max SoCs...
Apple is still selling an Intel Mac mini (which can be had with up to 64GB RAM), if they intended to keep the Mac mini to only entry-level (meaning only the low-end M1 SoC) they would have discontinued the Intel model...
As for "without beefing up their CPU/GPU", Mac mini with a maxed-out M1 Max SoC would make an excellent personal workstation for 3D / DCC by day and rendernode (on 10Gb Ethernet) by night...
I could see these Mac mini Pro models filling CGI / Animation classroom labs, used by students in the day & grinding out frames on the overnight...
Or a big-time production house doing the same, but workers by day, renderfarm by night, and all with dual SoC MAc Cubes or quad SoC Mac Pros (and these units would have dual 10GB connections)...! ;^p
M1 Max Metal (Geekbench 5)
MacBookPro18,2 - Geekbench
Benchmark results for a MacBookPro18,2 with an Apple M1 Max processor.
browser.geekbench.com
Really surprised to see this coming from the 32 core. I'm really hoping that this is just a mislabeled 24 core. All the figures we are getting out of this match so closer to the 24 core, not the 32.
EDIT: OK actually, this very well could be the 24 core. The GPU core count isn't listed. Let's hope this is the 24 core and not the 32!
Last edited:
Unlikely, since that is more than 3X as fast as M1.
Also, I’d be quite surprised if Apple sent out 24-core M1 Max models for review.
This is just one result of course and the number will likely increase somewhat but don’t expect miracles.
Its, at best, 10% more than 3x. Given the upgraded RAM size AND the increased bandwidth, I think it's in the realm of possibility. This is also one example I've linked.
Well, this is just one M1 Max result. This is 10% more than 3X when compared to some of the best M1 results. However, when compared to the average M1 result, this is about 3.3X M1.
Mac mini (Late 2020) - Geekbench
Benchmark results for a Mac mini (Late 2020) with an Apple M1 processor.
browser.geekbench.com
Not sure how you are arriving at those numbers to be honest. Most of the scores hover around 21-22k for metal. This is more or less right in line with 3x.
You're not looking close enough. The scores range from below 19000 to just over 22000, and Geekbench itself says it's 20581-21168.
We don't know where 68870 lies on the M1 Max 32 scale, but if we compare that to say 21000, then it's 3.18X scaling. If we compare that to 20000, it's 3.44X scaling.
apple said m1 max is 4x faster than 5600M, not sure how they get this result is close to 3090.
Yes, for OpenCL. For Metal they hover around what I said. You're looking at everything, not just the Metal score, which is what this is based off. OpenCL and Metal are different scores.
EDIT: Even this agrees with me (scroll down to M1 GPU). https://browser.geekbench.com/metal-benchmarks
Ah yes, it seems the scores below 20000 are only OpenCL. Noted.
If you look below that it also says Apple M1 is 20581. I was looking around and some of the sub 21000 Metal scores are MacBook Pros. There are even more MacBook Airs below 21000 but presumably some of those are 7-core.
But like I said, if you compare this M1 Max score vs 21000, that's about 3.2X scaling.
I'd like you to be right on this, but I'm not optimistic, esp. since I wouldn't expect Apple to ship out 24-core Maxes to reviewers.
I'm not sure on the math you're doing here. I'm getting ~9% difference between the two.
Yes, but can you show me your math on how you are arriving at that?
I'm doing (21,000 x 3)/69,000, which accounts for a difference of around ~9% of the total.
In other words (Expected/Actual)
Oh I see what you mean. Anyhow, you posted as I was editing. I'm simply dividing the score by 21000 to get 3.28X scaling. If you choose 22000, then it's 3.13X scaling.
Both are more than 3X. Since we don't know where 68870 sits in the range, perhaps can just guess it's probably around 3.2X scaling. Given that it's more than 3X and furthermore it's rather unlikely Apple would send out the gimped versions of the MacBook Pro Max for review, I think it's being overly optimistic to think it's the 24-core model being tested. Added to that, Geekbench was actually identifying the chip as the 32-core variant.
You could be right, but I'd guess that isn't the case here. For you to be right, you're counting on 24-core Max to scale better than 3X, Apple intentionally sending out gimped 24-core Maxes for review, and Geekbench misidentifying the chip.
I was suggesting 3X = 300% + 10% = 310%. That's if you use 22000 as a baseline for comparison - 3.13X.
However, if you use 21000 as a baseline for comparison, it's actually 3.28X or about 3.3X.
I think we should split the difference and call it 3.2X for now, until we have more data. The point is though, to expect that...
1. Apple shipped out 24-core models to reviewers
2. Geekbench is misidentifying 24-core models as 32-core
3. 24-core models are scaling better than 3X vs 8-core models at the same clock speed
...is reaching.
I won't say it's completely impossible, but the KISS principle suggests to me that it's more likely a 32-core Max that isn't scaling perfectly, which is OK actually.
Ah I see the math you’re doing. Even geekbench doesn’t calculate it that way for what’s it’s worth.
That being said, I am unsure. Apple could be sending out the 24 core due to supply constraints, and for various other reasons. Apple doesn’t always send out the best hardware to reviewers. We’ve seen this a few times in the past with the M1 and other launches.
Also, like I’ve stated previously, I believe it’s reporting as an error. We saw the CPU benchmarks being listed as 24MHz instead of whatever the proper speed was. Errors happen. I don’t think this is totally out of the realm of possibility.
Lastly, youd REALLY expect the 32 core to be 4x performance, not 310%, like is showing in geekbench.
I was guessing it would be more like 3.5X scaling, just because. (Yes, I just pulled that number out of my... err.. thin air...)
Looking back through this thread, it seems @leman was guessing ~3.2X to ~3.3X scaling, and we're right in that ballpark.
We'll see as more numbers come in. The more scores that show up higher than that one 68870 score, the more unlikely it will be the 24-core variant. Let's see if it hits @leman's 70000 guesstimate.
Then there's also the fact that the 32core has 2/3rd's of the compute units that the 3080 does (which they were using to compare in their own keynote), which if the given numbers are true, genuinely doesn't make sense. This would mean there's some sort of severe bottleneck in geekbench for compute. If we assume the score of ~80K for the actual 32 core, this would be in line with the numbers given for the razor 3080 laptop (which Apple directly quoted) and has a score of ~120K.
I genuinely don't believe this to be the 32 core. I think if this was the case, Apple wouldn't have been so adamant about their presentation. Granted, this IS just one benchmark, but the numbers are so far off for the 32 core, I have a hard time believing this isn't the 24 core.
Also, in the Metal Score posted above, the core count isn't listed. So we don't actually know for a fact if this is either or.
I genuinely don't believe this to be the 32 core. I think if this was the case, Apple wouldn't have been so adamant about their presentation. Granted, this IS just one benchmark, but the numbers are so far off for the 32 core, I have a hard time believing this isn't the 24 core.
Also, in the Metal Score posted above, the core count isn't listed. So we don't actually know for a fact if this is either or.
so ig if is 32 core it will be 8 core more than 24 core one that's about ～33% improve that would be 20869 more in metal compare to 24 core one so 20869+68870=89739 that's about 90000 and 3080 max q is about 12000 if apple is true then the problem could be Geekbench....