SemiconductorsBRIEF

Apple Unveils M6 and M5 Ultra Chips, Refreshing Mac mini and Mac Studio for Desktop AI

E
EffectStory 編輯部Editorial Team
Published · Updated
Apple unveiled the M6 and M5 Ultra chips on August 25, 2026, alongside new Mac mini and Mac Studio desktops. M6 is Apple's first 2-nanometer chip; M5 Ultra's four-die design lifts peak AI GPU performance up to 4.5x over M3 Ultra.

What did Apple announce with M6 and M5 Ultra, and which products use them?

Apple announced the M6 and M5 Ultra chips on its website on August 25, 2026, alongside new Mac mini and Mac Studio desktopsCITE:E1. Apple said the launch is meant to strengthen on-device artificial intelligence (AI) processing and graphics performance across its desktop lineupCITE:E1.

What core innovations define the M6 chip?

M6 is Apple's first chip built on a 2-nanometer process, pairing a 12-core CPU and 12-core GPU with dual 16-core Neural EnginesCITE:E2. Apple said this configuration lifts peak AI GPU performance by nearly 30% over the M5CITE:E2.

How much do M6's memory bandwidth and multithreaded performance improve?

M6 supports up to 32GB of unified memory with 170GB/s of memory bandwidthCITE:E3. Apple said multithreaded performance rises as much as 1.2x over M5 and as much as 2.4x over the original M1CITE:E3.

How does M5 Ultra's multi-die architecture enable higher AI performance?

M5 Ultra, positioned for professional workstations, links two dual-die M5 Max chips into a four-die package using Apple's UltraFusion technologyCITE:E4. In its top configuration, M5 Ultra offers a 36-core CPU, 80-core GPU, 1.2TB/s of unified memory bandwidth, and up to 512GB of unified memory, with peak AI GPU performance rising as much as 4.5x over M3 UltraCITE:E4.

How do the new Mac mini's M6 and M5 Pro configurations upgrade AI and specs?

The new Mac mini ships in M6 and M5 Pro versions, with the M6 model adding a Neural Accelerator to every GPU core for the first timeCITE:E5. Apple said this lifts AI performance by up to 4x and graphics performance by up to 2x over the M4 Mac mini, with 16GB of unified memory as standardCITE:E5. The M5 Pro configuration tops out at an 18-core CPU and 20-core GPU with up to 64GB of unified memory and a Thunderbolt 5 port, and the new Mac mini overall adds Wi-Fi 7, Bluetooth 6, and 2.5Gb Ethernet supportCITE:E6.

How much does Mac Studio's AI performance jump with the new M5 chips?

The new Mac Studio offers M5 Max and M5 Ultra configurations with up to 512GB of unified memoryCITE:E7. Apple said AI performance rises more than 4.3x over the previous generation, targeting professionals who train large AI models, render complex 3D scenes, or edit high-resolution videoCITE:E7.

Chip specifications at a glance

ChipCPU coresGPU coresUnified memoryMemory bandwidthAI performance gain
M61212up to 32GB170GB/sGPU AI peak up to ~30% over M5CITE:E2
M5 Ultraup to 36up to 80up to 512GB1.2TB/sAI GPU peak up to 4.5x over M3 UltraCITE:E4

Product configurations at a glance

ProductChip optionsUnified memory (max)Other specsAI/graphics gain
Mac mini (M6)M616GB standardNeural Accelerator in every GPU coreCITE:E5AI up to 4x, graphics up to 2x vs. M4 Mac miniCITE:E5
Mac mini (M5 Pro)M5 Pro (18-core CPU, 20-core GPU)64GBThunderbolt 5, Wi-Fi 7, Bluetooth 6, 2.5Gb EthernetCITE:E6
Mac StudioM5 Max / M5 Ultra512GBAI performance up over 4.3x vs. prior generationCITE:E7

What this means: Apple's own figures point to two distinct upgrade paths rather than one uniform leap. M6's gains center on per-core efficiency — a 2-nanometer process, dual Neural Engines, and a 170GB/s, 32GB memory ceiling — consistent with on-device inference workloads in the Mac miniCITE:E2CITE:E3CITE:E5. M5 Ultra's gains instead come from scaling the package itself: fusing two M5 Max dies raises memory bandwidth to 1.2TB/s and capacity to 512GB, which lines up with Mac Studio's reported 4.3x AI performance jump and its stated target of large-model training and 3D rendering workloadsCITE:E4CITE:E7. The Mac mini's split between a 16GB M6 tier and a 64GB, Thunderbolt-5-equipped M5 Pro tier shows Apple segmenting AI capability within a single desktop line rather than reserving it for one chip aloneCITE:E5CITE:E6.

📊 Evidence

FAQ

When did Apple announce the M6 and M5 Ultra chips?

Apple announced both chips on its website on August 25, 2026<CITE:E1>.

What is M5 Ultra's maximum unified memory and bandwidth?

M5 Ultra supports up to 512GB of unified memory with 1.2TB/s of bandwidth<CITE:E4>.

What connectivity upgrades come with the new Mac mini?

The new Mac mini adds Thunderbolt 5, Wi-Fi 7, Bluetooth 6, and 2.5Gb Ethernet support<CITE:E6>.

📎 Sources

  1. cna.com.tw

Related data

Author's TakeEffectStory 編輯部

The headline number is architecture, not raw core count. M5 Ultra's 4.5x AI GPU gain over M3 Ultra comes from fusing two dual-die M5 Max chips into a four-die package via UltraFusion, which is what pushes memory bandwidth to 1.2TB/s and capacity to 512GB — the two figures that matter most for training large models on a desktop rather than in the cloud. M6's gains look different: a 2-nanometer process and dual 16-core Neural Engines lift AI GPU throughput by about 30% over M5, while memory stays capped at 170GB/s and 32GB, a profile suited to on-device inference rather than training. Worth watching next is how the Mac mini's two tiers diverge in practice: the M6 configuration standardizes at 16GB with per-core Neural Accelerators, while the M5 Pro configuration offers 64GB and Thunderbolt 5 — Apple is now segmenting AI capability within one desktop line rather than across separate product categories.

E
EffectStory 編輯部Editorial Team

Related

BRIEF

SpaceX and NVIDIA Plan to Launch Orbital AI Data Centers Starting Next Year, Musk Says

SpaceX plans to launch its first NVIDIA-chip-powered AI satellites, Starmind AI1, in the fourth quarter of 2027, reaching large-scale deployment by 2028 — at least a year ahead of the original schedule. The company has also filed with the FCC for a network of up to 1 million satellites, while Taiwanese suppliers Unitech and Sesoda report rising satellite-related revenue and order shares tied to the buildout.

林紀旭 James Lin ·
BRIEF

NVIDIA Launches Jetson Orin Nano 2 Robotics Computer, Targets Entry-Level Edge AI

NVIDIA announced the Jetson Orin Nano 2 robotics computer on August 25, 2026, packing 78 TOPS of AI compute, 8GB of memory, and an 8-core Arm CPU while doubling inference performance and cutting power draw 40% versus its predecessor. Modules and developer kits ship in the first half of 2027, targeting a robotics developer base NVIDIA says already tops 3 million.

Nathan ·