Geekbench AI: Indication of Topaz, LrC speed up?

[gradi]

Someone on another forum ran Geekbench AI 1.0 on his M1 Max MBP 64gb with Sonoma and then with Sequoia. The AI results for the Neural Engine, GPU, and CPU improved substantially with Sequoia.

The Geekbench AI scores for the M1 Max, Sonoma vs. Sequoia, for the Neural Engine, GPU, and CPU is very impressive. I have Ventura 13.7 on my 16" M2 Pro 12/19 MBP 32gb/1tb and have seen no new features in Sonoma or Sequoia that interest me, but if there is a nice performance boost for photo AI stuff such as in Topaz, LrC, etc. then, of course, I would welcome that and would likely upgrade (maybe wait for 15.1).

Has anyone here noticed a good speed boost with photo programs that use AI after updating to Sequoia?

 

[crazy dave]

@jo-1 also wrote about the same pattern for the M4 iPad Pro here:

beware - there will always be progress - that's an intrinsic factor in technology.

BUT there are times when that progress makes quantum leaps like with the M4 in the iPad PRO 2024 or the A12X in the iPad PRO 2018 or the M1 MAX in the MacBook Pro or the S10 in the Apple Watch 10 ...

It is important to understand these leaps in technology for doing an educated decision when buying technology relevant things.

Just completed the Geekbench AI tests on my M4 iPP and the leap in Software from iPadOS 17.6.1 to 18.0 is very impressive - we'll probably see that kind of progress not only in HW but also in future SW while Apple is ironing out the routines to address ML in the new devices.

Here're my results for the M4 ( which is kind of the first generation of the new chips found in future devices )
View attachment 2421247

We can expect not only SW but also HW to improve significantly with the 2 nm process and that will be definitely my next iPhone.

Apple has just begun to optimize SW and HW for ML and with the next generation of iPhone SoCs the leap will be very significant again - probably the biggest leap in efficiency for a long time

edit: The A18 in the current new iPhone shares the NE with the M4 and the new NE will just get better

I haven't seen any others verify runs of this. I'm still on Sonoma, I'll try to remember to test my Mac before upgrading.

 

[gradi]

Please report back to this thread what you discover. I do not recall reports that Sonoma was much faster than Ventura (which I am currently running). But if Sequoia provides a real performance boost for my Topaz Photo AI program and Lightroom Classic then I would likely update to Sequoia. Sometimes benchmark results don't always get mirrored in real life app results though.

 

[dmccloud]

I had to hunt for my results under Sonoma since the GB results browser still doesn't let you look up past results, but I found them in an old post. Based on my results, performance is a bit of a mixed bag overall...


Sonoma Results:

CPU:
4152 Single Precision / 6702 Half Precision / 5707 Quantized

GPU:
12632 Single Precision / 13815 Half Precision / 10733 Quantized Score

Neural Engine:
4132 Single Precision / 20807 Half Precision / 23059 Quantized

Sequoia Results:

CPU:
4110 Single Precision / 6579 Half Precision / 5190 Quantized

GPU:
12572 Single Precision / 13741 Half Precision / 12676 Quantized

Neural Engine:
4103 Single Precision / 24542 Half Precision / 27376 Quantized

 

[gradi]

I had to hunt for my results under Sonoma since the GB results browser still doesn't let you look up past results, but I found them in an old post. Based on my results, performance is a bit of a mixed bag overall...


Sonoma Results:

CPU:
4152 Single Precision / 6702 Half Precision / 5707 Quantized

GPU:
12632 Single Precision / 13815 Half Precision / 10733 Quantized Score

Neural Engine:
4132 Single Precision / 20807 Half Precision / 23059 Quantized

Sequoia Results:

CPU:
4110 Single Precision / 6579 Half Precision / 5190 Quantized

GPU:
12572 Single Precision / 13741 Half Precision / 12676 Quantized

Neural Engine:
4103 Single Precision / 24542 Half Precision / 27376 Quantized

Thank you. Weird that the other guy saw such a massive performance boost using Geekbench AI going from Sonoma to Sequoia. But maybe unique to his 64gb M1 Max or maybe he somehow made a mistake in the testing.

 

[crazy dave]

Thank you. Weird that the other guy saw such a massive performance boost using Geekbench AI going from Sonoma to Sequoia. But maybe unique to his 64gb M1 Max or maybe he somehow made a mistake in the testing.

I just upgraded. Weirdly did get some error messages, out of memory from my system while running it, but I'm not sure if that's relevant. The app finished and ran just fine and the memory seemed okay when I watched. Unsure what the cause of the error message was.

However, yeah the 14" M3 Max had a massive performance boost in Geekbench AI going from Sonoma to Sequoia:

Sonoma GB AI 1.1 M3 Max 14" 14-core	Single Precision Score	Half Precision Score	Quantized Score
CPU	3835	5398	4752
GPU	13347	13619	4229
NPU	2609	13243	14543

Sequoia 15.1 GB AI 1.1 M3 Max 14" 14-core	Single Precision Score	Half Precision Score	Quantized Score
CPU	4385	7490	5987
GPU	15245	19752	17673
NPU	4588	32894	35772

 

[gradi]

I just upgraded. Weirdly did get some error messages, out of memory from my system while running it, but I'm not sure if that's relevant. The app finished and ran just fine and the memory seemed okay when I watched. Unsure what the cause of the error message was.

However, yeah the 14" M3 Max had a massive performance boost in Geekbench AI going from Sonoma to Sequoia:

Sonoma GB AI 1.1 M3 Max 14" 14-core	Single Precision Score	Half Precision Score	Quantized Score
CPU	3835	5398	4752
GPU	13347	13619	4229
NPU	2609	13243	14543

Sequoia 15.1 GB AI 1.1 M3 Max 14" 14-core	Single Precision Score	Half Precision Score	Quantized Score
CPU	4385	7490	5987
GPU	15245	19752	17673
NPU	4588	32894	35772

Thank you for this update. You certainly got a big benchmark performance boost with Sequoia. I am still on Ventura 13.7.1 on my M2 Pro 12/19 MBP. It certainly seems like for Apple Intelligence that work has been done in MacOS to improve the performance of AI stuff. What I really care about is whether Topaz Photo AI and Lightroom Classic that use the AI stuff (Neural Engine, etc.) would also see a performance boost. Maybe I will update to Sequoia in the next few weeks and find out. Anyone here already tried it with photo apps?

 

[crazy dave]

Thank you for this update. You certainly got a big benchmark performance boost with Sequoia. I am still on Ventura 13.7.1 on my M2 Pro 12/19 MBP. It certainly seems like for Apple Intelligence that work has been done in MacOS to improve the performance of AI stuff. What I really care about is whether Topaz Photo AI and Lightroom Classic that use the AI stuff (Neural Engine, etc.) would also see a performance boost. Maybe I will update to Sequoia in the next few weeks and find out. Anyone here already tried it with photo apps?

There I cannot help you I'm afraid. Also some of my boosts are insane to the point where I have to wonder if something is wrong with either the original run or the new one - e.g. GPU quantized going from 4000+ to 17500+. But my numbers are my numbers and while I can't rerun the old OS numbers, I've rerun the new ones and gotten the same result - also since a restart, haven't gotten the same error messages with the numbers being the same.