Why can’t Apple be clear about the performance of the M1 Ultra?
Apple’s new M1 Ultra will be an impressive processor.
There’s no doubt about it since the M1 Ultra is, well, pretty much two impressive M1 Max chips joined at the hip. That makes it a 20-core CPU (16 performance, 4 efficiency), plus a 64-core GPU and a nutty 800 GB memory bandwidth thanks to its unified LPDDR5/6400 memory for all cores.
But even with all those impressive specs and the M1’s known pedigree, it’s still maddening to understand why Apple will never be clear about how fast its new computers will be.
For the M1 Ultra reveal, Apple showed off another round of its mud-clear performance charts, like the one below. Of course, the main advantage is that the 20-core M1 Ultra outperforms a “16-core desktop” while using less power, but the lack of clarity on speed and speed how is it faster is infuriating.
Apple does the same for graphics, claiming that the M1 Ultra’s 64-core GPU is slightly faster than the “highest-end discrete GPU” while using less than 200 watts less power. For the record, while Apple didn’t specify in its presentation what the components are, the company did specify in the footnotes that it was shooting on Intel’s 12th Gen Core i9-12900K and Nvidia’s GeForce RTX 3090.
What motivates me even more is that despite these charts being about as generic as legally possible, there are some legit and impressive numbers that Apple quoted, but didn’t mention in their presentation.
In Apple’s M1 Ultra press release, it says the new processor can transcode 6K ProRes video 5.6 times faster than the current Mac Pro’s 28-core Intel Xeon processor paired with its fancy Afterburner card. . Sure, the old Mac Pro Xeon is now old and musty (whose fault is that?), but Apple’s Afterburner card features a custom-designed ASIC designed solely to accelerate ProRes media on the desktop. ‘Apple. The fact that the tiny Mac Studio can smoke it by 5.6x is indeed worth fighting for. But Apple didn’t in its M1 Ultra reveal, instead focusing on those smoky, undefined graphics that make people doubt claims more than believe them.
So what are we to believe? Obviously Apple’s performance claims of ProRes transcoding for one (by the way, ProRes is Apple technology). The jaw-dropping first leaks don’t surprise us either: the M1 Ultra will indeed be fast.
For example, in an apparently programmed error, someone leaked a Geekbench 5 score of the M1 Ultra, putting it at 24,055 for multi-core performance and 1,793 for single-core performance. Geekbench, if you don’t know, uses a few dozen performance “loops” modeled on modern workloads like text compression, cryptography, and JPEG decompression. It is also famous for people who do not pay attention to the warning signs that the results are uploaded to the internet.
Is the M1 Ultra faster than Intel? Yes and no).
Obviously, assuming we accept that the leaked performance is close to the actual performance of the M1 Ultra, we can see two things from the Geebench scores: the 20-core M1 Ultra seems to be around 29% faster than the score of 18,551 we saw in our original 12th Gen Core i9-12900K test here. Given that we’re looking at 20 cores versus 16 cores, that probably fits.
In Geekbench 5 single-threaded performance, although it is reversed. The M1 Ultra, remember, is basically two M1 Max chips with much better cooling. With the 12th Gen Core i9-12900K scoring 2,001, that gives the Intel processor an 11% win over the leaked M1 Ultra score.
If you’re wondering how the M1 Ultra compares to the M1 Max, we saw MacWorld’s score for the 10-core M1 Max at 12,671 for multi-core Geekbench 5, which is what you’d see if you broke the 20- M1 Ultra core in half. In single thread, it’s also basically within the margin of error with a score of 1,788.
Seeing the gigantic cooling that Apple put into the Mac Studio for the M1 Max, we expect the final scores to be faster, but at first glance it looks like the M1 Ultra basically gives you, well, 2x the M1 Max.
If we were to play it even further by just doubling the performance of the M1 Max, for example, Maxon’s Cinebench R23, that would give the M1 Ultra a score of 24,750 (2 x 12,375) in multi-threaded performance and probably the same performance at 1,529 in single-threaded performance. How does this compare to Intel’s Core i9-12900K? For multi-threaded performance, we recorded a score of 27,275, which would mean the 16-core Core i9 outperforms the 20-core M1 Ultra by 10%. And with the Core i9’s score of 1,984, the Intel processor would easily outperform the M1 Ultra by almost 30%.
We still don’t totally believe that, as it’s hard to believe that the giant copper cooler Apple uses in the Mac Studio wouldn’t (we hope) pay more dividends. None of this lives up to Apple’s power claims, but actual performance results are best left to independent reviewers.