Apple's M5 Ultra with an 80-core GPU will power through your AI models and 8K video
Engadget · C · trust 58/100
The M6 has also arrived with all-new 2nm process technology.
Apple With the launch of the latest Mac mini and Mac Studio , Apple introduced some brand new chips with cutting-edge technology. The M6 is the company's first processor built on a 2-nanometer process, while the the M5 Ultra is debuting a new "quad-die" configuration that marries four M5 processors into a single, powerful chip ideal for AI or content creation.
Let's start with the Apple's most powerful processor ever, the M5 Ultra, coming to new high-end Mac Studio machines. Though based on the M5's older third-generation 3nm process, it fuses two dual-die M5 Max chips to form the quad-die architecture, "a first for Apple silicon," according to the company.
Joining four chips might sound clunky, but the M5 Ultra's interconnects run at a record 4.4TB/s, allowing the quad dies to operate very much as a single processor. It's available with up to 36 cores (12 super cores and 24 performance cores) that deliver up to 25 percent higher single-threaded performance and 30 percent better multithreaded performance than the M3 Ultra.
The M5 Ultra supports up to 512GB of high-bandwidth unified memory with 1.2TB/s of bandwidth, up 50 percent over the M3 Ultra. Given current sky-high memory prices, that could push your Mac Studio with M5 Ultra into five-figure territory, since the base model starts at $5,499. That may be worth it for some AI power users, though. The extra memory will let them "store huge datasets entirely in local memory, increase the tokens-per-second speed, and run huge LLMs with hundreds of billions of parameters entirely on device," Apple explains.
Boosting those speeds further is a next-gen 80-core (max) GPU with a neural accelerator in each core that bumps peak AI compute 4.5 times compared to the M3 Ultra, while boosting graphics performance by 40 percent. It should handle whatever video creators throw at it as well, thanks to an updated Media Engine with four ProRes encode and decode engines, along with hardware-accelerated AV1, H.264 and HEVC codecs. Along with the extra compute power, Apple promises "industry leading energy efficiency."
Next is the M6 that will feature in some Mac mini models. It will be a strange duck by Apple's standards. As we learned earlier this year from Bloomberg 's Mark Gurman, Apple reportedly only plans to build an M6 and not release M6 Pro or M6 Max versions at all. Instead, it will skip straight to the M7 by mid-2027, apparently due to technological leaps with that chip in the area of AI processing.
That's not to say that the M6 isn't advanced. It uses TSMC's latest 2nm process tech, "packing greater transistor density into a smaller die for a major leap in performance and power efficiency," Apple states. It has a new 12-core processor complex (two more than the M5), with two super cores, four performance cores and six efficiency cores. Unified memory is quite limited at 32GB, but the 170GB/s bandwidth is up 10 percent over the M5.
With all that, the M6 "delivers the world's fastest single-threaded performance and up to 1.2x faster multi-threaded performance as compared to M5," according to Apple.The company is also promising a 30 percent uplift in neural processing, along with improvements to the shader core architecture, dynamic caching and hardware-accelerated ray tracing. Those updates will help it deliver faster gaming frame speeds and "50 percent increased geometry rates for complex graphics," Apple said.
Still, this is a chip that will likely only appear in low-end Mac, iMac and MacBook machines (if Apple doesn't release Pro or Max versions as rumored). Its greatest feature will probably be power efficiency, which will make it ideal for productivity laptops. Stay tuned for Engadget's reviews to find out how much faster and more efficient the new chip really is.
Read the original at Engadget →
Open in TruthVane →