1143 articles tagged with AI Hardware
Sudo su (@sudoingX)로컬 AI를 여러 GPU 등급에서 실험한 결과, Qwen 3.5 9B를 RTX 3060 12GB에서 128K 컨텍스트와 초당 50토큰으로 구동해 단일 프롬프트로 전체 게임을 자율적으로 만들었다는 사례를 공유했다. 엔트리급 GPU에서도 강력한 로컬 AI 활용 가능성을 보여준다.https:/...
When the #AI bubble pops, it’s going to be interesting to figure out how to run those cheap data center GPU’s in the #homelab for #selfhosting.
RoundPipe: Full Fine-Tune 32B Models on a Single 24GB GPURoundPipe fine-tunes 32B models on a single 24GB GPU with 1.5-2.2× speedups via round-robin pipeline dispatch.https://genti...
Inference shift opens door for AI chip startups to challenge NvidiaInference shift from training to serving creates opportunities for AI chip startups. Nvidia's $20B Groq acquihire...
AI-chip startups tar upp kampen mot Nvidia i inferensens nya erahttps://redaktionen.net/artikel/832#ai #svtech
🌘 GitHub - systalyze/utilyze➤ 從負載監控邁向效率優化：深入剖析 GPU 真實算力✤ https://github.com/systalyze/utilyzeUtilyze 是一款專為 GPU 效能監控設計的開源工具，旨在解決傳統工具（如 nvidia-smi）僅能顯示「負載率」而非「實際運作效率」的問題。透過直接讀取 GPU 效...
Oh, joy! Another groundbreaking GitHub tool nobody asked for – now you can finally measure how well your GPU is doing "useful" work… because, apparently, keeping up with the latest...
Born ML Framework migrates to gogpu/wgpu — pure Go WebGPU, zero CGO, zero shared libraries. True single binary ML with GPU. Plus: DeepSeek V4 roadmap.
📰 C++ CuTe/CUTLASS vs CuTeDSL (2026): The New GPU Kernel Learning Path for LLM InferenceAs GPU kernel engineering evolves, CuTeDSL is emerging as NVIDIA’s preferred path for new de...
📰 CuTeDSL vs CUTLASS 2026: GPU Kernel Mühendisleri İçin PyTorch, NVIDIA ve FlashAttention ile Öğren...2026'da GPU kernel ve LLM enferans mühendisliği için C++ CUTLASS mı, yoksa Pyt...
A new kind of memory device may finally solve the problem of overheating and battery drain in electronics. By shrinking components to an extreme scale and redesigning their structu...
Cerebras Systems, a leading player in the AI chip industry, is gearing up for a major public offering.
NVIDIAが次に挑むのは「金融産業」の6500億ドル市場。GPUだけではないAI半導体覇者の本当の強み | Business Insider Japan https://www.yayafa.com/2792328/ #AgenticAi #AI #ArtificialGeneralIntelligence #ArtificialIntelligence ...
I wish the cool advances in the RISC-V space jump from AI to consumer hardware. I would love to have M*-level performance in a RISC-V CPU #computers #RISC-V #AI
Intel responds to MacBook Neo with 21% faster chip intended for similar devicesApple’s launch of the MacBook Neo – and especially the incredible price/performance balance – sent sh...
Every iPhone has a useful hardware feature that zero Samsung phones offerWhen Apple released the iPhone 17e, it doubled the storage and added what it calls the “magic of MagSafe” t...
Apple's silicon moat faces its toughest test. As AI reshapes chip supply chains and competitors sprint ahead on generative AI, $AAPL's hardware-led strategy & Ternus-era pivot face...
If you’ve spent any time in the open-source AI community recently, you’ve probably seen someone...
I don't hate #AI, but I also don't talk to it like it was a person. I had a hardware problem (Apple display). I tried searching over 20 minutes or so. I even watched a couple of vi...
The AI industry has reached an inflection point — but not in the way most narratives suggest. 🧵Key signal: average GPU utilization across 23,000 clusters is just 5%. We're overbuil...
A Chinese GPU maker secures critical Microsoft certification, poised to challenge NVIDIA and AMD. AWS pivots to integrate ...
📰 SageAttention 2.2.0 ile RTX 5090'da %35 Hız Artışı: Blackwell GPU'da ComfyUI Performansı (2026)SageAttention 2.2.0, RTX 50-serisi Blackwell GPU'lar üzerinde ComfyUI için %30-35 a...
Supercharging the RTX 5090 in PhysX games using an RTX 5060 as a secondary GPU — SLI may be dead, but how much can dual GPUs boost performance in classi…We're testing the impact of...
After the AI bubble bursts, is there something useful we can do with those GPU and NPU power and loads of memory? #AskFedi #AI #AIbubble