Apple announced the M5 Ultra on August 25 as its most powerful chip to date. The design links two dual-die M5 Max chips through an updated UltraFusion interconnect to form the first quad-die M-series processor. The move supplies a concrete hardware step for users who already run heavy local workloads on Mac Studio and Mac Mini systems.
Context
Earlier Ultra chips stopped at a dual-die layout. The M5 Ultra is the first time Apple has packaged four dies together in one M-series part. The new UltraFusion link provides more than 4.4 TB/s of inter-die bandwidth and raises connection density by more than six times, so the four dies appear to software as a single processor. Apple paired the chip announcement with refreshed Mac Mini and Mac Studio models and mentioned an M6 processor in the same briefing, though no specifications for the M6 were released.
Details
The M5 Ultra CPU contains up to 36 cores: 12 super cores and 24 performance cores. Apple claims single-threaded performance rises by as much as 1.25 times and multithreaded performance by as much as 1.3 times versus the M3 Ultra. The GPU scales to 80 cores, and each GPU core now includes a Neural Accelerator—an addition not present in prior Ultra chips. These figures come directly from the MacRumors account of the announcement.
TechCrunch reported that Apple presented the M5 Ultra and the M6 together with the updated desktop hardware. The same report notes that the UltraFusion interconnect is the central technical change enabling the quad-die configuration. No release dates, pricing, or third-party benchmark data appear in either source. The MacRumors summary cuts off before listing further GPU or memory details.
The sources contain no statements from independent developers, no power-consumption figures, and no comparison against current Intel or AMD workstation parts. They also omit any mention of memory bandwidth or cache hierarchy changes that usually accompany a die-count increase.
Reactions and counterpoints
Neither source includes quotes from third-party developers or competing vendors. The limited information released so far leaves open questions about sustained thermals under full load and about how the new Neural Accelerators in the GPU cores will be exposed to developers.
Why it matters
Engineers and technical founders who already use Mac Studio or Mac Mini for local model training, video encoding, or large-scale compilation now have a defined next hardware target. The shift to four dies and the placement of Neural Accelerators inside every GPU core show Apple continuing to raise on-device AI capacity while still scaling traditional CPU and GPU counts. The stated 1.25× and 1.3× gains are measured only against the M3 Ultra; whether those margins hold against current AMD Threadripper or Intel Xeon systems will require independent testing that is not yet available. Buyers planning workstation purchases in the coming quarters therefore have clearer specifications to track, but they still lack data on power draw, memory configuration, and real-world throughput that would let them compare the M5 Ultra directly with non-Apple alternatives. The announcement supplies a specification milestone rather than a finished product review.
{"word_count": 612, "sources_used": 2}---
Sources:
No comments yet