Joeman2026-10-03 06:41:32Joeman says maxed-out M5 Ultra Mac Studio brings local text and image generation close to cloud speedTaiwanese YouTuber Joeman said his fully configured M5 Ultra Mac Studio, priced at NT$437,900, delivered striking local AI performance after a week of testing across benchmarks, video editing, gaming, and on-device model workloads. In his test results, local text generation and image generation ran close to cloud services, while video generation still lagged behind top cloud models. The machine he bought used a 36-core CPU, 80-core GPU, 256GB of unified memory, and 4TB of storage. Joeman also compared export times against an M4 Max MacBook Pro, showing the M5 Ultra cut Premiere export time from 6 minutes 27 seconds to 3 minutes 2 seconds. In AI tasks, he said a Qwen image model produced the same prompt in 27 seconds on the M5 Ultra, versus 49 seconds on an M3 Max and 3 minutes 4 seconds on an M2 Max. A 5-second video generated locally still took 6 minutes, compared with 5 minutes 7 seconds on cloud Wan 2.2 and 3 minutes 52 seconds on Kling 3.0. Joeman said the biggest value of the machine is keeping sensitive data on-device, and argued local and cloud computing will coexist with different roles.350
OpenAI2026-08-31 07:15:38OpenAI and Anthropic are driving Mac shortages as AI firms pile into Mac mini and Mac StudioA report from The Information says OpenAI has bought tens of thousands of Mac mini and Mac Studio systems for reinforcement learning workloads and for training AI agents that can operate computers. Anthropic is taking a different route, renting Mac compute through Amazon Web Services to run similar jobs. The demand is now feeding into Apple’s supply chain, with higher-end Mac configurations reportedly staying out of stock for months. The report argues that AI companies are not choosing Macs because of brand preference. The main draw is Apple’s M-series unified memory architecture, which fits memory-heavy workloads that do not depend as heavily on large-scale parallel processing as model pretraining does. These include computer-use agent training loops that repeatedly observe a screen, act inside a real operating system, and receive feedback. The shift is notable because it opens a separate lane in the AI hardware market. NVIDIA still dominates large-scale training infrastructure, and high-end GPU clusters remain central for pretraining frontier models. But Apple’s desktop Macs appear to be gaining ground in reinforcement learning, agent training, and some local enterprise deployment cases where memory efficiency matters more than raw parallel compute.1170
OpenAI2026-08-31 02:27:00OpenAI Buys Tens of Thousands of Macs for AI Training; Apple Mac Revenue Jumps 29%OpenAI has bought tens of thousands of Mac mini and Mac Studio machines for reinforcement learning training and Computer-use Agent tasks, according to The Information, which cited people inside OpenAI. Anthropic has also been renting large amounts of Mac computing capacity through Amazon Web Services. Apple's Mac business generated $10.4 billion in revenue in its most recent quarter, up 29% year over year. That growth rate surpassed iPhone, making Mac the fastest-growing hardware segment at Apple. The OpenAI purchase spans the Mac mini and Mac Studio. Anthropic's Mac compute rental runs through AWS, the report said. Apple's Mac numbers come from its latest quarterly earnings. The Mac revenue gain was 29%, while iPhone grew at a slower pace in percentage terms. The report did not say exactly how many Macs OpenAI bought, only that the number is in the tens of thousands. The machines are used for reinforcement learning training and for Computer-use Agent tasks. Both OpenAI and Anthropic are among the AI companies using Apple's desktop hardware for part of their computing needs.1050
Apple2026-08-25 15:21:23Apple unveils M5 Ultra, says AI peak compute reaches 4.5x M3 UltraApple on Aug. 25 introduced the M5 Ultra, a new high-end chip the company said is the most powerful in its lineup to date, with a clear focus on AI computing and graphics performance. According to Apple’s newsroom announcement, the chip debuts exclusively in the new Mac Studio released the same day. The M5 Ultra marks the first quad-die design in the M-series. Apple said it combines two dual-die M5 Max chips through its UltraFusion packaging technology, delivering more than 4.4TB/s of inter-die bandwidth and over six times the connection density of the prior generation. The chip can scale up to a 36-core CPU, made up of 12 super cores and 24 performance cores, with Apple claiming up to 1.25x faster single-threaded performance and up to 1.3x faster multi-threaded performance than the M3 Ultra. On the graphics side, the M5 Ultra supports up to an 80-core GPU, with each core equipped with a Neural Accelerator. Apple said peak AI GPU compute is up to 4.5x that of the M3 Ultra and more than 6x higher than the M1 Ultra, while graphics performance is up to 40% faster than M3 Ultra. The chip also includes a 32-core Neural Engine, up to 512GB of unified memory, and 1.2TB/s of memory bandwidth. Mac Studio models with M5 Ultra start at $5,499, with shipments scheduled to begin on Sept. 22.1030
Apple2026-08-25 13:30:26Apple unveils new Mac mini and Mac Studio with a stronger push into AI workloadsApple has introduced new Mac mini and Mac Studio models as it leans harder into AI computing, saying more users are turning to Apple computers to build and run AI agents. In a Tuesday announcement, the company said its chips for iPhone and Mac include a dedicated Neural Engine for running AI models, along with a RAM architecture designed to reduce some memory bottlenecks. The new Mac mini comes in M6 and M5 Pro versions, priced in mainland China from RMB 6,999 and RMB 12,999. Apple described the desktop as suitable for around-the-clock agent computing and said it can stay on a desk for long-running AI agent tasks. The M6 is Apple’s first 2nm M-series chip, with both the CPU and GPU moving to 12 cores, while each GPU core adds a neural acceleration unit for the first time. Apple said AI performance can reach up to 4x that of the M4 version. Memory starts at 16GB and goes up to 32GB on the M6 model, with bandwidth rising to 170GB/s. Apple also said prompt processing speeds for large models in LM Studio can be as much as 4.8x faster than the M4 version. The M5 Pro model supports up to 64GB of unified memory and 307GB/s of bandwidth. Both versions add Wi‑Fi 7, Bluetooth 6 and a 2.5Gb Ethernet port, with sales starting on Sept. 22.1150
Erik Voorhees2026-08-24 15:40:45Erik Voorhees says the ROI on local AI hardware is getting worseShapeShift founder Erik Voorhees said the return on investment for local AI hardware is becoming less attractive, pointing to a $20,000 Mac Studio as an example during a discussion with Kain Warwick and Taylor Monahan. The conversation focused on the economic trade-offs facing individuals and companies that are considering buying high-performance on-premise AI computing equipment. According to Voorhees, the rapid rise of cloud-based AI services and the faster pace of hardware iteration are making it harder to justify expensive local machines for AI workloads. In his view, the cost recovery case for buying premium local hardware is weakening as cloud offerings improve and new devices arrive more quickly. The remarks were cited in a post referenced as coming from @laurashin.490
Apple2026-07-22 14:35:00Apple said to be preparing a broad Mac refresh from this fall through next yearApple is preparing new versions across its current Mac lineup from this fall into next year, according to a market report cited by BlockBeats on July 22. The planned refresh is said to cover both desktops and laptops, pointing to one of the company’s widest Mac update cycles in the period described. Products mentioned in the report include an entry-level 14-inch MacBook Pro equipped with a new M6 chip, a new iMac, and higher-end 14-inch and 16-inch MacBook models featuring a redesigned chassis and OLED touch displays. Apple is also said to be planning updates for the MacBook Air and MacBook Pro lines more broadly. The report added that new Mac mini and Mac Studio models with upgraded chips are also in the pipeline. At the same time, release timing for some of these devices may be affected by supply challenges tied to memory chips. No more specific launch dates were provided beyond the window spanning this fall to next year.1420
