News Apple's M1 Max SoC Allegedly Makes Shaky Geekbench Debut

[Admin]

These early Geekbench performance numbers for the M1 Max seem a bit soft.

Apple's M1 Max SoC Allegedly Makes Shaky Geekbench Debut : Read more

 

[jridder]

Hmmm, I don't know about trusting these numbers. The build numbers for macOS seem a little suspicious. Build 21A558 was released to developers today and it's supposed to be RC. The build on this bench says 21E140a. Something is off.

 

[hotaru251]

Could be soemthing unexpected (or a bad sample)

either that or M1 chip does not properly scale.

 

[JamesJones44]

This could also be the bottom end 6 performance 2 efficiency version found in the bottom end 13" MBP and being miss reported as 10 core. The single thread score still looks shaky with that, but the multi core score would make sense in that scenario. Definitely a take with a grain of salt result.

 

[joneskind]

Alright guys, I did the math.

Be A a performance core and B an efficiency core.

• We know from the M1 benches that 4A + 4B = 7590
• The allegedly M1 Max benches says that 8A + 2B = 11542

If the second statement is true :

2 x (8A + 2B) = 23084
12B = 23064 - 7590 = 15894
B = 1291
A = 606

Which means that a performance core is a little more than twice as powerful as the efficiency one.
It seems consistent with Apple's description of the M1

 

[joneskind]

either that or M1 chip does not properly scale.

Apps running on M1 chips can use both the efficiency cores and the performance cores at the same time. So do Geekbench.

The M1 is a 8 cores chip (4-4) while the M1 Max is a 10 cores chip (8-2)

Those benches seems fairly consistent to me, without having to invoke scalability problems.

I did the math in a comment above

 

[joneskind]

Some new benches pop up, with better scores

https://browser.geekbench.com/v5/cpu/search?q=Apple+M1+Max

12693 for the best one

 

[bkkcanuck8]

Take all them with a grain of salt (some of them made no sense i.e. version 12.4 build number out of whack... but any case any or all of them can be spoofed). That being said, I had guestimated 12750 approximately (probably a rough calculation based on change in configuration of cores and e-cores running around 70% as powerful as e-cores (when all cores used) and of course a slight penalty for all cores - per core speed.... So 12,693 seems reasonable to me (11,750ish seemed off - and I would have been disappointed).

 

[bkkcanuck8]

Alright guys, I did the math.

Be A a performance core and B an efficiency core.

• We know from the M1 benches that 4A + 4B = 7590
• The allegedly M1 Max benches says that 8A + 2B = 11542

If the second statement is true :

2 x (8A + 2B) = 23084
12B = 23064 - 7590 = 15894
B = 1291
A = 606

Which means that a performance core is a little more than twice as powerful as the efficiency one.
It seems consistent with Apple's description of the M1

I vaguely remember hearing that e-cores are about 70% as powerful as a p-core... but when you run all p-cores, they won't run the same performance as a single p-core, and the average of 4 p-cores is not going to be the same as the average of 8 p-cores.

That said, I was expecting around 12,750ish.... I was hoping I was underestimating but it seems reasonable. 11,750 sounds low. I don't know if any of these are from real runs though.

 

[bkkcanuck8]

My bad, I actually under estimated the improvement... I found my response on a youtube video.... (it was made when it was expected to be an 8/4 mix). It is beginning to look like it is around 12,750 (about 3,000 more than your top of the line gaming laptop). GPU is a bit of a outstanding question (geekbench does a compute benchmark not necessarily a video based benchmark).

"I expect the CPU portion of M1X to max out at 11,750 on Geekbench (about 58% more performance) (probably a bit less maybe 50%). The calculation I used takes what I believe is the performance profile for an efficiency core of 70% of the performance core. Assuming that - without taking any thermal throttling down on adding 4 more performance cores (likely there will be some) would give the Geekbench score of 11,750. [i.e. x4 + x40.70 = 7399; then x = 1088; therefore x8 + x40.70 = 11,750 or around 58%... though that is likely higher due to clock speed of performance of all core execution dropping some amount] "