Ars Technica Tests M5 Ultra Mac Studio, Finds Uneven Gains at a Higher Price

Apple’s pricier Mac Studio lost a video-encoding test to its cheaper sibling. In a hands-on review published September 24, Ars Technica found that the M5 Ultra finished behind the M5 Max on an H.264 encode and led it by only about 26% in a Geekbench multi-core test. Those new results put a sharp limit on what buyers can assume from the Ultra’s doubled chip specifications.

Strong gains, except where they weren't

The exception stands out because the Ultra was much faster elsewhere. Against the outgoing M3 Ultra, Ars Technica measured roughly 50% faster multi-core CPU performance and GPU gains of 33% to 66%, depending on the test. In many general-purpose tests against the M5 Max, the Ultra’s CPU was 80% to 90% faster and its GPU 50% to 80% faster.

The review also found the Max was barely slower than the Ultra on a separate H.265 encode. These are results from specific tests, not a measure of how every task will run. Apple’s August launch figures emphasized selected peak and “up to” performance comparisons; the review shows why those figures cannot predict every workload.

The premium has grown

The M5 Ultra Studio starts at $5,499, versus $3,999 for the cited base M3 Ultra configuration when it launched. The review puts an M5 Ultra with 256GB of memory and 1TB of storage at $9,499, compared with $5,599 for a comparable configuration about 18 months earlier. Those are launch-era comparisons, not prices for identical chips.

The less expensive M5 Max has risen too: it starts at $2,499, up from $1,999 for the cited M4 Max model at launch. The cost of extra Ultra capacity therefore matters alongside its speed advantage, especially when the advantage changes by task.

Why memory still makes the Ultra attractive

For local AI, memory capacity determines how much room a model and a coding agent’s working context have. Bandwidth affects how quickly the system can move data and generate output, though it is not the only factor. Apple lists the M5 Ultra at up to 1.2TB per second of memory bandwidth and plans a 512GB memory option; that capacity is distinct from the review’s 256GB price comparison.

The odd encoding result remains unexplained. Both M5 machines drew about 75 watts on average during the review’s transcoding test. The reviewer considered a power limit, but observed unused CPU capacity and Ultra clock speeds matching the Max’s; a bug remained possible. Ars Technica sent the findings to Apple. An answer could help buyers judge whether that shortfall is fixable or a lasting constraint.