Daily AI news, sourcedUpdates / Chips & infra
Updates Chips & infra · 115 items · newest issue first · page 2 of 6
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- audio.cpp adds Drama3 text-to-speech, running up to about 5x faster than real time on an RTX 5090Unconfirmedr/LocalLLaMAOct 16Chips & infra
- User reports ~79 tok/s running DeepSeek-V4-Flash-0731 on four Radeon AI PRO R9700s via Affinity’s refused four-card layoutUnconfirmedr/LocalLLaMAOct 16Chips & infra
- Developer releases a reverse proxy that wakes or shuts down a local LLM server on incoming requestsUnconfirmedr/LocalLLaMAOct 16Chips & infra
- Codex Day 11 cuts 'Selected model is at capacity' interruptions and auto-continues when they still occurOfficialOpenAI (Codex)Oct 16Chips & infra
- Local Inference Lab publishes a ~98 GiB QAD NVFP4 build of Qwen3.8-Flash-Next that is smaller than NVIDIA's 133GB NVFP4 checkpointOfficialLocal Inference LabOct 16Chips & infra
- LocalLLaMA user offloads speculative-decoding drafters to an AMD Strix machine over llama.cpp RPC to stretch two RTX 4060s from 150K to 195K contextUnconfirmedr/LocalLLaMAOct 16Chips & infra
- Microsoft adds experimental llama.cpp support to Windows ML so apps can run GGUF models locally through new task-specific APIsOfficialMicrosoft Foundry on Windows BlogOct 16Chips & infra
- Apple shows a new MLX release running Muse Spark Open at over 36 tokens per second across four M5 Ultra Mac Studios linked by RDMA over ThunderboltOfficialApple Machine Learning ResearchOct 16Chips & infra
- llama.cpp v0.7.0 moves the server's default port to 9931 and adds LiliCorr speculative decoding, K2 Horizon support and a web UI models manager
- Ollama v0.40.5-rc1 ports MLX Metal custom kernels to CUDA and accepts custom tool calls in Responses history
- LocalLLaMA user measures KLD and top-1 agreement of Unsloth Qwen3.8 Flash Next quants on English and Finnish textUnconfirmedr/LocalLLaMAOct 15Chips & infra
- First RTX Spark tests find the Surface Laptop Ultra finishes local agent runs about 25% faster than an M5 Max MacBook Pro but lags in games, per Tom's HardwareReportedTom's HardwareOct 15Chips & infra
- NVIDIA details Nordic AI buildout, from Sweden's Sferical One supercomputer to a Stockholm AI Lab with NorrskenOfficialNVIDIA BlogOct 15Chips & infra
- DeepSeek reports a partial outage of chat and V4 Pro and V4.1 Flash APIs while a DeepSeekHarness Bilibili account appears
- Nvidia's share of China's AI chip market fell to 34% in the first half as Huawei rose to 28%, IDC data shows, per ReutersReportedReuters (via U.S. News)Oct 15Chips & infra
- Developer's PulsarLM inference engine drops old, low-attention tokens to keep long chat and agent contexts boundedUnconfirmedr/LocalLLaMAOct 15Chips & infra
- LocalLLaMA user builds three four-GPU Radeon AI Pro R9700 machines on $127 used EPYC CPUs for about $10,700 each and 384GB of VRAMUnconfirmedr/LocalLLaMAOct 15Chips & infra
- LocalLLaMA user benchmarks AMD's MI350P against Nvidia's RTX Pro 6000 Blackwell after tuning AMD kernels to double its out-of-box speedUnconfirmedr/LocalLLaMAOct 15Chips & infra
- Alexandr Wang says Meta is fixing a batch of inference bugs in Muse Wonder's deployment and apologizes for its uneven performanceStatementX @alexandr_wangOct 15Chips & infra
- LocalLLaMA user reports 176 tokens/s decode and 6,450 tokens/s prefill with Q4 Qwen3.8-Flash-Next on two Radeon AI Pro R9700s using the Radiance engineUnconfirmedr/LocalLLaMAOct 15Chips & infra