News
Short write-ups of new developments, in our own words, with sources and an as-of date. Facts are checked against the sources; interpretation is labelled separately.
Today in 3 lines
-
Nvidia Explains NVFP4 and KV Cache Tiering as Responses to Memory Shortage (blogs.nvidia.com)
On November 5, Nvidia said in an official blog post that memory supply remains tight and prices are rising, and described NVFP4 checkpoints and KV cache tiering as its responses. The figures are Nvidia's own claims.
-
SK hynix Q3 Revenue at 104.6 Trillion Won, Demand Expected to Outpace Supply Through 2030 (news.skhynix.com)
SK hynix announced on October 28 its Q3 revenue of 104.6155 trillion won and operating profit of 83.3338 trillion won. It expects demand to keep exceeding supply through 2030, and more than half of its production capacity for 2027 and 2028 is already under long-term contracts.
-
Apple Unveils MacBook Pro with Touch OLED 'PROLED XDR' and Up to 128GB Memory on M5 Max (apple.com)
On October 29, Apple unveiled 14- and 16-inch MacBook Pro models with a touch-enabled tandem OLED display. The chips are M5 Pro and M5 Max, unified memory goes up to 128GB, and sales start November 6.
-
Apple M6 14-inch MacBook Pro unveiled: up to 32GB memory, 170GB/s bandwidth (apple.com)
Apple unveiled the 14-inch MacBook Pro with the M6 chip on October 27. Unified memory starts at 16GB and goes up to 32GB, with bandwidth of up to 170GB/s. All performance figures are Apple's own comparisons.
-
Rokid Color Full-Color Glasses Unveiled in China… Overseas Release Set for CES 2027 (idevice.com)
Rokid unveiled Rokid Color, full-color microLED glasses, starting at 4,999 yuan. Reportedly weighing 55g with a no-wristband control method, the overseas launch is scheduled for CES in January 2027.
-
Samsung and Warby Parker 'Intelligent Eyewear' Pre-orders to Open in the US on November 12 (sammobile.com)
Warby Parker's co-CEO reportedly said at the earnings call that pre-orders for the glasses made with Samsung and Google will start in the US on November 12. Pricing will be announced that day, and shipping is set for mid-December.
-
LTX-2 Code v1.6.0, Code Moves to Apache 2.0… Weights Keep a Separate License (github.com)
Lightricks updated the LTX-2 repository to v1.6.0 (November 4) and v1.6.1 (November 5). The code is now Apache 2.0, but model weights are not covered by that license, and several legacy APIs were removed.
-
Sol-H3-VAE, a Fast Decoder for MiniMax-H3, Adds Support in ComfyUI 0.44.0 (huggingface.co)
Sol-H3-VAE, which replaces H3's video VAE decoding, has been released, and ComfyUI 0.44.0 added support on November 4. The production team reports that decoding a 5-second clip on an RTX 5090 dropped from 12.5 seconds to 5.0 seconds.
-
llama.cpp b12180 adds text-only Kimi-K3.1-Pro support… Vulkan detects ReBAR (github.com)
On November 6, llama.cpp build b12180 added text-only Kimi-K3.1-Pro support. The same day, the Vulkan backend began detecting whether ReBAR is supported.
-
Ollama 0.40.12~0.40.15: LFM3 Expansion, MLX Speculative Decoding and Sampling Improvements (github.com)
Ollama released four times between October 27 and November 3. LFM3 expanded to Linux and Windows, DFlash speculative decoding became supported on MLX, and top-k sampling got faster.
-
65B Dense Model Ripple-65B Preview Released Under Apache 2.0 (huggingface.co)
Ricursive Intelligence has released the 65B dense model Ripple-65B Preview under Apache 2.0. It supports a context of up to about 1 million tokens, and all benchmarks are the developer's own figures.