M2 Ultra benchmarks show performance bump over M1 Ultra
The first benchmark results for the M2 Ultra have started to surface, with the latest Apple Silicon chip appearing to have an as-expected bump in performance over its predecessor.

M2 Mac Studio
Apple introduced the M2 Ultra on Monday as the M2 equivalent of its M1 Ultra. Less than a week after its introduction, benchmark results for the chip are now appearing online.
The first wave of Geekbench 6 results, spotted by MacRumors, seems to peg the single-core score of the M2 Ultra at around 2,800, and a multi-core score in excess of 21,000.
According to Geekbench's browser, a Mac Studio with an M1 Ultra was able to reach 2,379 for the single-core mark, 17,565 for multi-core. If the new results are genuine, the single-core performance of the M2 Ultra would be in the region of 17% more than the M1 Ultra, while multi-core performance will be up almost 20%.
During the launch of the updated Mac Studio, Apple said the M2 Ultra can "deliver 20% faster performance than M1 Ultra." It also claimed the 76-core GPU was 30% faster, and the 32-core Neural Engine 40% faster.
The results are also a major bump from the last of the Intel Mac Pro models. Benchmark scores for the 128-core Xeon W-3257M used in the highest-spec Mac Pro only manages 1,377 and 10,382 in the single- and multi-core results.
In effect, the as-found M2 Ultra results appear to be more than double the scores of that Intel variant.
Read on AppleInsider
Comments
Double the Intel performance — ouch, indeed!
Way to go, Apple. Keep up the good work.
browser.geekbench.com/metal-benchmarks...
https://wccftech.com/apple-m2-ultra-soc-isnt-faster-than-amd-intel-last-year-desktop-cpus-50-slower-than-nvidia-rtx-4080/
And even the Intel 13900K and AMD 7950X are 2022 chips. Also, neither are workstation chips. A comparison with the AMD Threadripper 7000 and 7000 Pro that gets released in September, which will be made on the same node as the M2 Ultra, won't be particularly favorable. The M3 Extreme - which I am betting that Apple is going to launch on TSMC's 2nd gen 3nm process in 2025 - will be needed to compete, except that in 2025 the AMD Threadrippers made on TSMC's 1st gen 3nm process will be out, as will - and this is a worst case scenario for them - Intel's 5nm desktop and workstation chips.
Going forward Apple is likely going to de-emphasize direct comparisons between Apple Silicon and Intel - note that they have avoided mentioning AMD altogether - in favor of comparisons with previous generations of Apple Silicon. In a way that will be appropriate. The software that most people are going to run on the Mac Pro and Mac Studio are going to be so different from the software that most people are going to run on Threadripper and Xeon-W workstations and servers that direct comparisons will be impossible anyway.
PS: But since you mentioned these better” chips how is their performance per watt compared to the M2?
co-processor, Intel, AMD, Qualcomm and Nvidia in the near future will need to work even harder on wattage used, versus performance.
It’s not just the performance of the chips used. It’s also the OS and how the software is implemented with those chips, to achieve performance. you need to look no further than the Apple Vision Pro.
Now we know why Apple started to add LiDAR to their iPhones and iPads, we also now know why they started to make a lot of their in house programs share a closer look and feel across the Mac, iPhone, and iPad over the last few years too, which comes in handy for the Apple Vision Pro.
Djay a popular DJing app, has already extended its iPad app into VisionOS. From now to early next year, you will hear testimonials from developers daily, Apple has just created a whole new ecosystem for developers and it doesn’t hurt to have very good versatile in-house SOC chips.
https://developer.apple.com/news/?id=h3qjwosp
Additionally it's been 6 years since Apple introduced a wonderful AR demo at their Apple Store outside of their new main campus in Cupertino. Maybe today that might seem simple, but back then it was pretty crazy how that could render in real time.
https://www.youtube.com/watch?v=De73_NVWmWM
There is a reason the vast majority of render farms for example aren’t Mac Pros, and it is all Apple’s fault. This new Mac Pro will not change anything, being a Mac Studio in a bigger box costing one third more with a few limited function previous gen PCIe expansion slots (no GPU expansion) and proprietary SSD slots (why? why? WHY?). Hopefully the real new AS Mac Pro will come with the M3 variants next year, but this version does not bode well and I think it might be all too late.
Listed Tech specs for the Mac Pro. So you have two x16 slots that can support video cards. Granted the AUX power is only 300W total.
Again the same Aux power cables that are used on the 2019 Mac Pro are also compatible with the 2023 Mac Pro.
Even without video cards, the expansion is highly wanted and useful. Just having internal storage is highly useful, because it is much faster than external storage.
https://www.blackmagicdesign.com/products/intensitypro4k
https://www.blackmagicdesign.com/products/decklink/techspecs/W-DLK-34
I think it probably would be possible to support a PCIe GPU in some way, even if just for compute but there may be a reason why they will try to avoid it. 3rd party GPUs have a different rendering architecture and the Metal API has to support both.
https://developer.apple.com/metal/sample-code/
https://developer.apple.com/documentation/metal/metal_sample_code_library/rendering_a_scene_with_deferred_lighting_in_swift
"Some macOS GPUs have an immediate mode rendering (IMR) architecture. On IMR GPUs, a deferred lighting renderer can only be implemented with at least two render passes. Therefore, the sample implements a two-pass deferred lighting algorithm for the macOS version of the app."
If Apple eventually deprecates immediate mode support in Metal, 3rd party GPUs wouldn't be usable with it.
There's not much point in 3rd party GPUs anyway. This is clear with what happened with Nvidia. Despite the Mac Pro having slots, there was never Nvidia support. AMD was supported because those GPUs were used all through the lineup. Nobody will write drivers and offer software support for the 0.01% of Mac users who want to add an AMD GPU on top of the Ultra GPU.
The geekbench scores suggest meaningful speed as of today however the lack of upgradability may tip the scale for some...
Is the latest Ultra is the first meaningful GPU boost from a hopped up 5,1 pro from 2010/2012, thanks to the 5,1 upgradability...
... and how might Apple justify this as serving the customer ...?
What do you mean by memory buffer? The CPU, GPU, ML and all other processing cores have direct access to the same memory space simultaneously. Unless data is coming from the SSD, once its in RAM is does not have to be shuttled back and forth, as a traditional GPU card would have to do. This is why a Mx chip can do more with less power.