Apple M1 Ultra Benchmarked, Gets Destroyed by the RTX 3090
Apple recently unveiled its fastest SoC yet: the M1 Ultra. Its new flagship silicon fuses two M1 Max chips into a single component with “Ultrafusion” technology, and powers the new Mac Studio. When Apple unveiled it at its Peek Performance event recently, it included several charts to demonstrate its prowess. Now, we hope you are sitting down, because it turns out its charts weren’t telling the whole story.
As we explained in our primer on the chip when it launched, when Apple displayed its performance charts it left out a few key details. Notably missing was the actual tests it performed to arrive at its conclusions. The charts were reminiscent of the kind we see from tech companies that show “relative performance” and then vague numbers like “1.5x.” Particularly notable were its comparisons to a “Highest-end discrete GPU,” which it noted was the RTX 3090 in the footnotes. In the chart, shown below, Apple claimed the M1 Ultra offered more performance than an RTX 3090, while using 200w less power. The quote from Apple is, “At its peak, M1 Ultra delivers faster performance than the highest-end GPU available while using 200 watts less power.”
Now, what Apple is saying here might be partly true, but it’s not the whole story. In The Verge’s testing, the RTX 3090 annihilated the M1 Ultra, both in gaming and compute benchmarks. In Shadow of the Tomb Raider, the RTX 3090 reached 142 fps at 4K compared to the Ultra’s 108 fps. “Of course Apple will suck at gaming,” you say. Point taken. However, in Geekbench’s compute test, The RTX 3090-equipped PC more than doubled the M1 Ultra’s scores, both in Metal and OpenCL. So where did Apple come up with the numbers for its chart? We have no idea, as the footnotes on the product page only state, “Performance measured using select industry‑standard benchmarks.”
“But The Verge only ran a few benchmarks,” you say. Ars Technica’s full battery of testing showed similar results, and they concluded it’s roughly equal to an RTX 3070. YouTuber Dave Lee found similar results in his tests as well, with the Nvidia GPUs destroying the M1 Ultra in Blender. To be fair to Apple though, the M1 Ultra was the fastest in Premiere Pro. This all seems to indicate the line for Apple’s M1 Ultra stops on the chart where the chip itself is “at its peak.” So it stopped the RTX 3090 line at the same spot, but the Nvidia GPU’s line could keep going even higher. It’s a confusing chart even by PR standards.
What’s insane about this situation is the M1 Ultra is a truly revolutionary chip in terms of efficiency. In Ars’ testing they said a Core i9-12900 beat the M1 Ultra in Handbrake, but it used 300W to do so. Meanwhile, the M1 Ultra was hovering in the mid-80W range the entire time. That’s impressive. Apple clearly has something special on its hands here, no doubt about it, but more powerful than an RTX 3090? That’s doesn’t seem likely, at least in most real-world workloads, and especially not in gaming.