A developer profiling Apple's M3 Neural Engine found an RTL performance bug that cuts DRAM weight-streaming speed from a normal 45-60 GB/s down to just 17-19 GB/s whenever a model's total weight size lands on an exact multiple of 1 MiB, affecting 7 of the 15 models in the ANEMLL project. By rerouting around the flawed speculative prefetch path in the kernel's DMA engine, the developer restored full throughput, more than doubling token generation speeds for Llama 3.2 1B and Qwen3-8B.
eiln.github.io
· 2026-09-10
Asahi Linux developers have officially declared support for Apple's M3, M3 Pro, and M3 Max Macs, bringing them close to parity with earlier M1/M2 support—except for the Mac Studio M3 Ultra, which remains unsupported. Key limitations remain, including weak GPU acceleration, no sleep mode, and non-functional HDMI ports on M3 MacBooks due to missing DCP support.
phoronix.com
· 2026-09-06
Apple has released the M5 Ultra chip for the new Mac Studio, built by fusing two dual-die M5 Max chips into a quad-die design, a first for the company's silicon lineup. The chip offers up to 36 CPU cores, a jump in inter-die connection speed to 4.4TB/s, and unified memory bandwidth of 1.2TB/s, both notable increases over the M3 Ultra.
engadget.com
· 2026-08-25
GoKawiil