正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

gpu kernels on on my computer writing other gpu kernels for the same chip to simulate 100ks of particles, each an agent + neural network sensing K bg fields and allocating energy to A actions, at 1000s steps/s, for thermodynamic artificial life i think maybe not a bubble lads

Fraser Price

1,191 subscribers

32,653 次观看 • 3 个月前 •via X (Twitter)

健康养生教育科学技术

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Today I release new GPU Particle Volumes system and the GPU Snow Particle shaders for it! This snow shader can render hundreds of thousands of particles at the same time even on a weak GPU. It also supports VRC Light Volumes. Get it on my Booth or Patreon!

Today I release new GPU Particle Volumes system and the GPU Snow Particle shaders for it! This snow shader can render hundreds of thousands of particles at the same time even on a weak GPU. It also supports VRC Light Volumes. Get it on my Booth or Patreon!

RED_SIM

43,816 次观看 • 10 个月前

It's here: We just hit superhuman performance on AI kernel optimization! Real customer models & production settings. Not toy problems (what I typically see). This is the year that Claude writes its own kernels, Codex its own kernels, for every new GPU that it wants to run on -- something that takes months to port between GPU generations today. This has a massive impact to scaling intelligence. More compute means getting the next frontier model sooner.

It's here: We just hit superhuman performance on AI kernel optimization! Real customer models & production settings. Not toy problems (what I typically see). This is the year that Claude writes its own kernels, Codex its own kernels, for every new GPU that it wants to run on -- something that takes months to port between GPU generations today. This has a massive impact to scaling intelligence. More compute means getting the next frontier model sooner.

Sharon Zhou

247,624 次观看 • 7 个月前

I wrote GPU kernels for GPT2 124M inference engine from scratch in ~1500 lines of CUDA. Here I explain it simply.

I wrote GPU kernels for GPT2 124M inference engine from scratch in ~1500 lines of CUDA. Here I explain it simply.

ninjaHattori1

160,092 次观看 • 1 个月前

timelapse #164 (15.5 hrs) - Burning midnight oil with the Kimi team benchmarking on kernels - Getting a Persistent Fused Mega Kernel for a Puffer Lib Environment (the full Craftax game) - Dipped for the night to have models run my overnight training runs on autopilot, including some kernel-based GRPO on qwen3.6 - private contract work - downloaded 1.1 TB of minecraft videos to train a foundation model on - optimized my minecraft foundation model training pipeline and kernels to train on 500k hrs of 128px at 10 fps in ~28 hrs on single gpu (5 hrs of learning per second on rtx pro 6000 blackwell) - doing it for the love of the game

timelapse #164 (15.5 hrs) - Burning midnight oil with the Kimi team benchmarking on kernels - Getting a Persistent Fused Mega Kernel for a Puffer Lib Environment (the full Craftax game) - Dipped for the night to have models run my overnight training runs on autopilot, including some kernel-based GRPO on qwen3.6 - private contract work - downloaded 1.1 TB of minecraft videos to train a foundation model on - optimized my minecraft foundation model training pipeline and kernels to train on 500k hrs of 128px at 10 fps in ~28 hrs on single gpu (5 hrs of learning per second on rtx pro 6000 blackwell) - doing it for the love of the game

Elliot Arledge

28,210 次观看 • 3 个月前

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps. pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU. That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own. Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack. Watch and save it, then fine-tune your first model tonight.

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps. pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU. That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own. Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack. Watch and save it, then fine-tune your first model tonight.

h100envy

497,788 次观看 • 3 个月前

China’s frontier models already run on China’s own chips. “Remember, the GLM-5 paper is almost a year old at this point. They say: ‘From the first day, GLM-5 is full-stack adapted to Chinese GPU ecosystems. We have successfully completed deep optimization, spanning from underlying kernels to upper-level inference frameworks, across seven mainstream domestic chip platforms, including Huawei Ascend, Moore Threads, Hygon, Cambricon, Kunlunxin, MetaX and Enflame.’” “It’s not necessarily NVIDIA GPUs and kernels we’re talking about. They’re serving this model, at least they claim, for many months on seven different domestic Chinese accelerators.” “People thought it was impossible that it’s served on GPUs by a Chinese lab, because Chinese labs aren’t that GPU rich. But turns out they’re using Chinese hardware. They plumb through everything, the kernels, the systems, while serving those models efficiently.”

China’s frontier models already run on China’s own chips. “Remember, the GLM-5 paper is almost a year old at this point. They say: ‘From the first day, GLM-5 is full-stack adapted to Chinese GPU ecosystems. We have successfully completed deep optimization, spanning from underlying kernels to upper-level inference frameworks, across seven mainstream domestic chip platforms, including Huawei Ascend, Moore Threads, Hygon, Cambricon, Kunlunxin, MetaX and Enflame.’” “It’s not necessarily NVIDIA GPUs and kernels we’re talking about. They’re serving this model, at least they claim, for many months on seven different domestic Chinese accelerators.” “People thought it was impossible that it’s served on GPUs by a Chinese lab, because Chinese labs aren’t that GPU rich. But turns out they’re using Chinese hardware. They plumb through everything, the kernels, the systems, while serving those models efficiently.”

SemiAnalysis

44,762 次观看 • 14 天前

Ray tracing sound on the GPU (Work in progress). Since the whole game is running on the GPU, I found it convenient to do the sound management and the ray tracing on the GPU as well. Each frame the CPU receives a sound queue from the GPU to process, and applies effects like reverb

Ray tracing sound on the GPU (Work in progress). Since the whole game is running on the GPU, I found it convenient to do the sound management and the ray tracing on the GPU as well. Each frame the CPU receives a sound queue from the GPU to process, and applies effects like reverb

xima

19,839 次观看 • 2 年前

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

Alex Ker 🔭

311,780 次观看 • 3 个月前

Nova uses an energy to enhance her cybernetics and weapons. Working closely with Quantum we were able to make a custom system using his gpu particles to bring this effect to life in #VRChat

Nova uses an energy to enhance her cybernetics and weapons. Working closely with Quantum we were able to make a custom system using his gpu particles to bring this effect to life in #VRChat

Meechu

15,753 次观看 • 1 年前

Big win for local LLM inference! This open-source engine generates tokens 2.7x faster than Ollama on a MacBook: I tested it on a base M5 MacBook Pro with Qwen3.5 4B at 4-bit and the same 9.2k-token prompt. Ollama ran its MTP build, so both had speculative decoding. - Prefill: Ollama 464 tok/s, Magnitude 755 tok/s (1.6x) - Decode: Ollama 22.6 tok/s, Magnitude 60.0 tok/s (2.7x) - End to end, Magnitude finished in half the time The video below shows both runs side by side. The Ollama numbers don’t mean it’s badly built. Its kernels were simply tuned in advance. A kernel is the small GPU program behind every matrix multiply and attention step. Its speed depends on choices like how work is split across GPU cores and how data moves through memory. The best choices change with every chip. Apple Silicon alone spans 8 to 80 GPU cores and under 70 to over 800 GB/s of bandwidth, and every model has its own layer sizes. A prebuilt engine has to pick one set of settings for all of them, so most local setups run on a compromise. Magnitude replaces that compromise with tuning on your machine: ↳ When you download a model, it compiles that model’s kernels locally. ↳ It tries different kernel configurations and keeps whichever runs fastest on your chip. ↳ It pairs each model with the fastest speculative decoding drafter it has found, so each pass of the model produces more than one token. GitHub repo: (don’t forget to star it 🌟) This matters most for agents. Every file and tool output has to be processed before the model writes anything, and that prefill step is where kernel efficiency shows up. Concurrent sessions also share prefix caches, so parallel agents don’t slow each other down. Magnitude is an open-source (Apache 2.0) inference engine written in Rust. It plugs into Claude Code, Codex, OpenCode, Pi and other agents with one click. I shared the GitHub repo above.

Big win for local LLM inference! This open-source engine generates tokens 2.7x faster than Ollama on a MacBook: I tested it on a base M5 MacBook Pro with Qwen3.5 4B at 4-bit and the same 9.2k-token prompt. Ollama ran its MTP build, so both had speculative decoding. - Prefill: Ollama 464 tok/s, Magnitude 755 tok/s (1.6x) - Decode: Ollama 22.6 tok/s, Magnitude 60.0 tok/s (2.7x) - End to end, Magnitude finished in half the time The video below shows both runs side by side. The Ollama numbers don’t mean it’s badly built. Its kernels were simply tuned in advance. A kernel is the small GPU program behind every matrix multiply and attention step. Its speed depends on choices like how work is split across GPU cores and how data moves through memory. The best choices change with every chip. Apple Silicon alone spans 8 to 80 GPU cores and under 70 to over 800 GB/s of bandwidth, and every model has its own layer sizes. A prebuilt engine has to pick one set of settings for all of them, so most local setups run on a compromise. Magnitude replaces that compromise with tuning on your machine: ↳ When you download a model, it compiles that model’s kernels locally. ↳ It tries different kernel configurations and keeps whichever runs fastest on your chip. ↳ It pairs each model with the fastest speculative decoding drafter it has found, so each pass of the model produces more than one token. GitHub repo: (don’t forget to star it 🌟) This matters most for agents. Every file and tool output has to be processed before the model writes anything, and that prefill step is where kernel efficiency shows up. Concurrent sessions also share prefix caches, so parallel agents don’t slow each other down. Magnitude is an open-source (Apache 2.0) inference engine written in Rust. It plugs into Claude Code, Codex, OpenCode, Pi and other agents with one click. I shared the GitHub repo above.

Akshay 🚀

34,322 次观看 • 1 天前

The rise of AI Agents is driving GPU demand to new heights. ⚡ At NodeAI, we’re scaling our decentralized GPU network across Asia, Europe, and the US to meet this growing demand. The AI Agent economy is here, and we’re ready to fuel it. #AI #GPU #NodeAI

The rise of AI Agents is driving GPU demand to new heights. ⚡ At NodeAI, we’re scaling our decentralized GPU network across Asia, Europe, and the US to meet this growing demand. The AI Agent economy is here, and we’re ready to fuel it. #AI #GPU #NodeAI

Node AI | $GPU

99,329 次观看 • 1 年前

I implemented Google Research's TurboQuant as a CUDA-native compression engine on Blackwell B200. 5x KV cache compression on Qwen 2.5-1.5B, near-loseless attention scores, generating live from compressed memory. 5 custom cuTile CUDA kernels ft: - fused attention (with QJL corrections) - online softmax -on-chip cache decompression - pipelined TMA loads Try it out: s/o Bryce, the CUDA Colonel and the cuTile team at NVIDIA for lending me Blackwell GPU access :) cc sunny madra Gavin

I implemented Google Research's TurboQuant as a CUDA-native compression engine on Blackwell B200. 5x KV cache compression on Qwen 2.5-1.5B, near-loseless attention scores, generating live from compressed memory. 5 custom cuTile CUDA kernels ft: - fused attention (with QJL corrections) - online softmax -on-chip cache decompression - pipelined TMA loads Try it out: s/o Bryce, the CUDA Colonel and the cuTile team at NVIDIA for lending me Blackwell GPU access :) cc sunny madra Gavin

ani

819,861 次观看 • 7 个月前

CMU PhD who built the kernels NVIDIA now ships in TensorRT-LLM explained fast attention in 68 minutes - better than $1200 GPU programming courses. pick the attention pattern -> generate a fused CUDA kernel -> drop it into vLLM/SGLang -> same GPU, way more tokens per second. That loop is why FlashInfer now powers inference at NVIDIA, vLLM, SGLang, and half the serving stacks you use. FlashInfer + Triton + JIT-compiled kernels + paged-KV attention - that's the stack.

CMU PhD who built the kernels NVIDIA now ships in TensorRT-LLM explained fast attention in 68 minutes - better than $1200 GPU programming courses. pick the attention pattern -> generate a fused CUDA kernel -> drop it into vLLM/SGLang -> same GPU, way more tokens per second. That loop is why FlashInfer now powers inference at NVIDIA, vLLM, SGLang, and half the serving stacks you use. FlashInfer + Triton + JIT-compiled kernels + paged-KV attention - that's the stack.

h100envy

32,605 次观看 • 3 个月前

We just open-sourced the world's fastest WebGPU kernels for 200+ machine learning operations. Specialized for your hardware. They run entirely locally in your browser. Now available on Hugging Face 🤗 Tell your agent to use @huggingface/kernels in your next project.

We just open-sourced the world's fastest WebGPU kernels for 200+ machine learning operations. Specialized for your hardware. They run entirely locally in your browser. Now available on Hugging Face 🤗 Tell your agent to use @huggingface/kernels in your next project.

Xenova

21,707 次观看 • 29 天前

every other day i yearn for more research fields to adopt the culture from the ML and ML adjacent fields of building a demo website for each paper and put code on github

every other day i yearn for more research fields to adopt the culture from the ML and ML adjacent fields of building a demo website for each paper and put code on github

outside five sigma

27,415 次观看 • 1 年前

Whenever I need a GPU, I go to HPC-AI and rent it for the lowest price you'll find. An H200 costs $1.89/hour. Compare that to AWS, where the same GPU costs $10.60/hour! No commitments. Pay as you go.

Whenever I need a GPU, I go to HPC-AI and rent it for the lowest price you'll find. An H200 costs $1.89/hour. Compare that to AWS, where the same GPU costs $10.60/hour! No commitments. Pay as you go.

Santiago

261,311 次观看 • 1 年前

I ported Vercel's fx coding agent to Windows 98 it talks to an LLM over a 56k modem and supports computer use (via MCP.exe) I now proudly run my own gamedev studio from 1999 my agents email each other in Outlook and store code on a network drive (git was not invented yet, sorry) good times

I ported Vercel's fx coding agent to Windows 98 it talks to an LLM over a 56k modem and supports computer use (via MCP.exe) I now proudly run my own gamedev studio from 1999 my agents email each other in Outlook and store code on a network drive (git was not invented yet, sorry) good times

alexey taktarov

31,848 次观看 • 17 天前

Lyapunov proved in 1892 that you never need to solve a differential equation to know if a system will stabilize. Find one function that behaves like energy — positive everywhere, zero at equilibrium — and if it's always decreasing, the system is guaranteed to converge. No solution required. The trick is that real energy often fails the test. A pendulum's kinetic energy hits zero at every moment of stillness, not just at rest — so it can't prove asymptotic stability alone. You need a generalized artificial energy with cross terms mixing position and velocity. The same Lyapunov stability analysis is what ML engineers run on neural network training dynamics to prove gradient descent actually converges instead of oscillating forever on GPU clusters. 16 minutes. Bookmark & watch today.

Lyapunov proved in 1892 that you never need to solve a differential equation to know if a system will stabilize. Find one function that behaves like energy — positive everywhere, zero at equilibrium — and if it's always decreasing, the system is guaranteed to converge. No solution required. The trick is that real energy often fails the test. A pendulum's kinetic energy hits zero at every moment of stillness, not just at rest — so it can't prove asymptotic stability alone. You need a generalized artificial energy with cross terms mixing position and velocity. The same Lyapunov stability analysis is what ML engineers run on neural network training dynamics to prove gradient descent actually converges instead of oscillating forever on GPU clusters. 16 minutes. Bookmark & watch today.

Deliriumm

46,282 次观看 • 1 个月前

Millions of GGUF downloads later, those same llama.cpp checkpoints can now run in 🤗 transformers. Same models, more ways to use them, and fast local inference on Mac powered by ggml kernels! Blog: ggml kernels:

Millions of GGUF downloads later, those same llama.cpp checkpoints can now run in 🤗 transformers. Same models, more ways to use them, and fast local inference on Mac powered by ggml kernels! Blog: ggml kernels:

Marc Sun

66,530 次观看 • 1 个月前

$Announcing the limited alpha release of our GPU marketplace, with $50 in free credits for 100 selected testers—apply now for a chance to participate! 🤘🏼🌪️ Hyperbolic’s GPU marketplace optimizes global GPU infrastructure, aggregating resources from data centers, mining farms, personal machines, and on-prem setups to make GPUs accessible at a fraction of the cost. Apply to gain access:$

Announcing the limited alpha release of our GPU marketplace, with $50 in free credits for 100 selected testers—apply now for a chance to participate! 🤘🏼🌪️ Hyperbolic’s GPU marketplace optimizes global GPU infrastructure, aggregating resources from data centers, mining farms, personal machines, and on-prem setups to make GPUs accessible at a fraction of the cost. Apply to gain access:

Hyperbolic

26,234 次观看 • 2 年前