正在加载视频...

视频加载失败

返回首页

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto...

omarsar0's profile picture

elvis

309,597 subscribers

261,622 次观看 • 2 年前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

10 条评论

elvis 的头像
elvis2 年前

source: github:

Adam Pippert 的头像
Adam Pippert2 年前

This is HUGE. I always suspected that there would be a way to break up the architecture of a model to alleviate the GPU cost and availability bottleneck. Will play with this on LLaMa2, but highly anticipating Mistral-7B.

Hamid R. Darabi 的头像
Hamid R. Darabi2 年前

It's very interesting! 11.69x improvement over llama.cpp sounds a lot, given that it's already super efficient. Are you sure it's not 11.7%? Even that could be a good result.

Loki (cute/acc) 的头像
Loki (cute/acc)2 年前

I almost tried this before I realized it's one more model exchange format 🥲 ".powerinfer.gguf" Can't all of you just use onnx and move on in life? 😭

catid (e/acc) 的头像
catid (e/acc)2 年前

Nice I was working in this direction recently noting the same things. 12x speedup is very respectable compared to 8x being the best so far achieved by any one approach (e.g. pruning/quantization)!

Filippo Pedrazzini 的头像
Filippo Pedrazzini2 年前

15 GiBs of weights? And this is supposed to run on Consumer Devices?! 🧐

Medium Boss - 70b_Float16.Q8.gguf 的头像
Medium Boss - 70b_Float16.Q8.gguf2 年前

The week I got a new PC with a beefy CPU. Fucking nice.

Sahar Mor 的头像
Sahar Mor2 年前

Paper tl;dr

s3nh 的头像
s3nh2 年前

Fastest bookmark 🤫🤫🤫

Ruairi 的头像
Ruairi2 年前

Noob question, is this just for the activation function or is it also cutting down the number of entries for GPU matrix multiplication as well?

相关视频

🚨 YOUR GPU IS PROBABLY WASTING MORE THAN YOU THINK. vLLM is built to squeeze far more useful work out of your GPU when serving LLMs. Running an LLM at scale isn’t just about having a powerful GPU. The real problem is how efficiently you use its memory and compute. That’s where vLLM comes in. → High-throughput LLM inference and serving → PagedAttention for smarter KV-cache memory management → Continuous batching to keep GPUs busy → Prefix caching + chunked prefill → OpenAI-compatible API out of the box → Supports a huge range of modern LLM architectures → Quantization support for running models more efficiently And the crazy part? You can start an OpenAI-compatible inference server with: `vllm serve ` So your application can talk to your own model almost like it’s talking to OpenAI. The bigger idea: Don’t just buy more GPUs. Make the GPUs you already have work harder. That’s why vLLM has become such a major project in LLM inference. 🔥 #vLLM #AI #LLM #Inference #GPU #MachineLearning #AIInfrastructure #OpenSource #AIAgents #Developers
0:20

🚨 YOUR GPU IS PROBABLY WASTING MORE THAN YOU THINK. vLLM is built to squeeze far more useful work out of your GPU when serving LLMs. Running an LLM at scale isn’t just about having a powerful GPU. The real problem is how efficiently you use its memory and compute. That’s where vLLM comes in. → High-throughput LLM inference and serving → PagedAttention for smarter KV-cache memory management → Continuous batching to keep GPUs busy → Prefix caching + chunked prefill → OpenAI-compatible API out of the box → Supports a huge range of modern LLM architectures → Quantization support for running models more efficiently And the crazy part? You can start an OpenAI-compatible inference server with: `vllm serve ` So your application can talk to your own model almost like it’s talking to OpenAI. The bigger idea: Don’t just buy more GPUs. Make the GPUs you already have work harder. That’s why vLLM has become such a major project in LLM inference. 🔥 #vLLM #AI #LLM #Inference #GPU #MachineLearning #AIInfrastructure #OpenSource #AIAgents #Developers

Vikas gupta

14,138 次观看 • 13 天前

NVIDIA'S $30K H100 JUST GOT OUT-INFERENCED BY A $250 RYZEN. 18 TOKENS/SEC ON A 35B MODEL, $0.03 IN ELECTRICITY, $47B GONE FROM NVDA BY FRIDAY download qwen3 35b-a3b -> load in lm studio -> disable gpu entirely -> run on cpu + ram -> hit 18 tokens/sec on a 6-core ryzen -> pay nothing per token. that loop is why moe models just made every consumer cpu a viable ai host and nvidia's inference monopoly is quietly on borrowed time. qwen3 35b-a3b + lm studio + ryzen 5 9600x + 32gb ddr5 + zero gpu - that's the stack. watch and save it, then run a 35b model on your laptop this weekend.
6:33

NVIDIA'S $30K H100 JUST GOT OUT-INFERENCED BY A $250 RYZEN. 18 TOKENS/SEC ON A 35B MODEL, $0.03 IN ELECTRICITY, $47B GONE FROM NVDA BY FRIDAY download qwen3 35b-a3b -> load in lm studio -> disable gpu entirely -> run on cpu + ram -> hit 18 tokens/sec on a 6-core ryzen -> pay nothing per token. that loop is why moe models just made every consumer cpu a viable ai host and nvidia's inference monopoly is quietly on borrowed time. qwen3 35b-a3b + lm studio + ryzen 5 9600x + 32gb ddr5 + zero gpu - that's the stack. watch and save it, then run a 35b model on your laptop this weekend.

Antid

23,745 次观看 • 1 个月前

GPU tradeoff series: A100 is not much more powerful than 4090 🫠 GPU Perf and Price: - 4090: 330 fp16 TFLOPs, $1,749 - A100 (80GB): 312 fp16 TFLOPs, $20,000 > A100 is 11.4X more pricy Training speed for GPT-2(124M) with llm.c: - 4090: 153K tokens/s - A100 (80GB): 195K tokens/s > A100 is only 1.3X faster (both trained using a single card, A100 llm.c training is shown in the video, 4090 video is in the quoted tweet) Conclusion: 4090 has a much better cost vs performance ratio Why: As in the H100 vs. 4090 comparison, the biggest difference between A100 and 4090 is their GPU memory size/bandwidth and cross-GPU communication bandwidth, which does not matter too much if your model can fit into a single 4090. Specs: 4090: - GPU memory size: 24GB - memory bandwidth: 1 TB/s - communication bandwidth: 64 GB/s A100: - GPU memory size: 80GB - memory bandwidth: 2 TB/s - communication bandwidth: 900 GB/s Nvidia killed off NVLink (a high-speed communication link that connects GPUs) on 4090. (Jensen Huang smiling face) If multiple 4090s could be interconnected via NVLink, their performance would be closer to datacenter-grade A100 GPUs, even for training larger models. Additionally, 4090 isn't allowed in datacenters, that's how Nvidia makes 💰💰💰
0:32

GPU tradeoff series: A100 is not much more powerful than 4090 🫠 GPU Perf and Price: - 4090: 330 fp16 TFLOPs, $1,749 - A100 (80GB): 312 fp16 TFLOPs, $20,000 > A100 is 11.4X more pricy Training speed for GPT-2(124M) with llm.c: - 4090: 153K tokens/s - A100 (80GB): 195K tokens/s > A100 is only 1.3X faster (both trained using a single card, A100 llm.c training is shown in the video, 4090 video is in the quoted tweet) Conclusion: 4090 has a much better cost vs performance ratio Why: As in the H100 vs. 4090 comparison, the biggest difference between A100 and 4090 is their GPU memory size/bandwidth and cross-GPU communication bandwidth, which does not matter too much if your model can fit into a single 4090. Specs: 4090: - GPU memory size: 24GB - memory bandwidth: 1 TB/s - communication bandwidth: 64 GB/s A100: - GPU memory size: 80GB - memory bandwidth: 2 TB/s - communication bandwidth: 900 GB/s Nvidia killed off NVLink (a high-speed communication link that connects GPUs) on 4090. (Jensen Huang smiling face) If multiple 4090s could be interconnected via NVLink, their performance would be closer to datacenter-grade A100 GPUs, even for training larger models. Additionally, 4090 isn't allowed in datacenters, that's how Nvidia makes 💰💰💰

Yuchen Jin

234,951 次观看 • 2 年前

I had to test it myself to believe this unreal inference speed. 3,000 tokens/s for 1 user on standard datacenter GPUs. They leveraged a hidden efficiency gap in how GPUs generate tokens. Kog just achieved 3,000 tokens/s on 8× AMD MI300X GPUs and 2,100 on 8× NVIDIA H200 (FP16, no speculative decoding). Their tech preview is on a 2B model, and they show how their techniques will scale to large frontier MoE models at similar speeds. That's a huge number because normal low-batch GPU decoding for 2B to 8B models is usually closer to 100 to 300 tokens/s per request, so Kog is claiming something like a 10X to 30X jump in the speed one user actually feels. Their trick: they are getting the speed by treating LLM decoding as a memory streaming problem, not mainly a math problem. For 1 user at batch size 1, the GPU is not doing big, efficient matrix-matrix work like in training or large-batch serving; it is repeatedly pulling the model’s active weights from high-bandwidth memory for each new token, so speed depends on how smoothly those weights keep flowing. Normal inference stacks keep breaking that flow. They run many separate GPU programs for different parts of the model, move intermediate results through memory, wait at synchronization points, talk back to the CPU for scheduling or sampling, and then repeat this token after token. Kog’s answer is to co-design 3 things that are usually tuned separately: the runtime, the low-level GPU code, and the model architecture. The biggest engineering move is the monokernel, where the whole decode pass runs as 1 persistent GPU-resident program, including sampling, so the system does not keep stopping for kernel launches, CPU scheduling, and intermediate memory round trips. They also rebuilt synchronization, because their own measurements say grid sync was eating around 35% of token-generation time; instead of making every compute unit wait at a broad barrier, each unit waits only for the exact data it needs. On AMD MI300X, they also map memory access around the chiplet layout, because memory latency changes depending on which die makes the request. Then their Laneformer model uses Delayed Tensor Parallelism, which lets cross-GPU communication happen in the background instead of blocking every layer.
1:08

I had to test it myself to believe this unreal inference speed. 3,000 tokens/s for 1 user on standard datacenter GPUs. They leveraged a hidden efficiency gap in how GPUs generate tokens. Kog just achieved 3,000 tokens/s on 8× AMD MI300X GPUs and 2,100 on 8× NVIDIA H200 (FP16, no speculative decoding). Their tech preview is on a 2B model, and they show how their techniques will scale to large frontier MoE models at similar speeds. That's a huge number because normal low-batch GPU decoding for 2B to 8B models is usually closer to 100 to 300 tokens/s per request, so Kog is claiming something like a 10X to 30X jump in the speed one user actually feels. Their trick: they are getting the speed by treating LLM decoding as a memory streaming problem, not mainly a math problem. For 1 user at batch size 1, the GPU is not doing big, efficient matrix-matrix work like in training or large-batch serving; it is repeatedly pulling the model’s active weights from high-bandwidth memory for each new token, so speed depends on how smoothly those weights keep flowing. Normal inference stacks keep breaking that flow. They run many separate GPU programs for different parts of the model, move intermediate results through memory, wait at synchronization points, talk back to the CPU for scheduling or sampling, and then repeat this token after token. Kog’s answer is to co-design 3 things that are usually tuned separately: the runtime, the low-level GPU code, and the model architecture. The biggest engineering move is the monokernel, where the whole decode pass runs as 1 persistent GPU-resident program, including sampling, so the system does not keep stopping for kernel launches, CPU scheduling, and intermediate memory round trips. They also rebuilt synchronization, because their own measurements say grid sync was eating around 35% of token-generation time; instead of making every compute unit wait at a broad barrier, each unit waits only for the exact data it needs. On AMD MI300X, they also map memory access around the chiplet layout, because memory latency changes depending on which die makes the request. Then their Laneformer model uses Delayed Tensor Parallelism, which lets cross-GPU communication happen in the background instead of blocking every layer.

Rohan Paul

13,282 次观看 • 3 个月前

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.
0:42

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

Akshay 🚀

343,270 次观看 • 1 个月前

Qwen3.8-Flash-Next just got a massive speed boost on my old hardware. This guy started with roughly 11 tok/s. After tuning the inference stack, I’m now getting around 24–28 tok/s. That’s roughly a 2× speedup without changing the GPU. The setup: → RTX 3090 24GB → 64GB DDR4 system RAM → Qwen3.8-Flash-Next → Cafe-llama.cpp The interesting part is that Cafe-llama.cpp isn’t just another llama.cpp fork. The optimizations are making a real difference for this particular architecture, especially when you’re dealing with a model that relies heavily on system memory and offloading. Going from 11 tok/s to nearly 30 tok/s completely changes how practical the model feels for local use. And now I’m curious what happens with DDR5. If memory bandwidth becomes another bottleneck, a faster RAM subsystem could squeeze even more performance out of the same 3090. This is why local AI optimization is getting so interesting. Sometimes you don’t need a new GPU. You just need a better inference stack. 11 tok/s → 24–28 tok/s.
0:50

Qwen3.8-Flash-Next just got a massive speed boost on my old hardware. This guy started with roughly 11 tok/s. After tuning the inference stack, I’m now getting around 24–28 tok/s. That’s roughly a 2× speedup without changing the GPU. The setup: → RTX 3090 24GB → 64GB DDR4 system RAM → Qwen3.8-Flash-Next → Cafe-llama.cpp The interesting part is that Cafe-llama.cpp isn’t just another llama.cpp fork. The optimizations are making a real difference for this particular architecture, especially when you’re dealing with a model that relies heavily on system memory and offloading. Going from 11 tok/s to nearly 30 tok/s completely changes how practical the model feels for local use. And now I’m curious what happens with DDR5. If memory bandwidth becomes another bottleneck, a faster RAM subsystem could squeeze even more performance out of the same 3090. This is why local AI optimization is getting so interesting. Sometimes you don’t need a new GPU. You just need a better inference stack. 11 tok/s → 24–28 tok/s.

FHILY👑

23,243 次观看 • 24 天前

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”
2:27

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Daniel Romero

48,894 次观看 • 2 个月前

GLM-5.2 running on Mac Studio Clusters or RTX GPU rigs isn't how Local AI will get mass adopted. For mass adoption we need to focus on two axes: 1. Intelligence/Speed Frontier. This is being tracked on local dot ai. Aggressive co-design is needed across the entire stack to push up the frontier specifically for local hardware. Better kernels, inference tricks like spec decoding, quantizing models, model routing. 2. Usability / UX. To use Local AI today, you need to install CLIs, select a model, a quantization, a harness that works best with that model, and you need to configure your inference engine. It's not very accessible. More work needs to be done on making this seamless, so the user doesn't even know the AI is running locally.
1:52

GLM-5.2 running on Mac Studio Clusters or RTX GPU rigs isn't how Local AI will get mass adopted. For mass adoption we need to focus on two axes: 1. Intelligence/Speed Frontier. This is being tracked on local dot ai. Aggressive co-design is needed across the entire stack to push up the frontier specifically for local hardware. Better kernels, inference tricks like spec decoding, quantizing models, model routing. 2. Usability / UX. To use Local AI today, you need to install CLIs, select a model, a quantization, a harness that works best with that model, and you need to configure your inference engine. It's not very accessible. More work needs to be done on making this seamless, so the user doesn't even know the AI is running locally.

Alex Cheema

30,854 次观看 • 2 个月前

Gavin Baker (Gavin Baker) says the disaggregation of inference can extend GPU useful lives from 3-4 years to 10-15. That may single-handedly save private credit and reduce the financing rates for GPUs, which will drive demand and help finance the build-out. "The disaggregation of prefill and inference is going to be amazing for the useful lives of GPU and may single-handedly save private credit. Private credit is in pain from these SaaS loans. But there's a lot of private credit in GPUs too. They were underwriting that to 3-4. The disaggregation of inference means that these GPUs are going to have 10 or 15-year lives. The AI skeptics are like, "Oh, these companies are all cooking their books. The useful life of a GPU is only a year or two. The useful life of a CPU is only four years because the rapid technological change." No. What rapid technological change has done with the disaggregation of prefill and inference is you can put a Cerebras system or Groq LPUs effectively in front of a Hopper or even an Ampere, use that Hopper and Ampere for prefill, and extend the useful life of that GPU until it melts. This is going to be really good for the whole private credit industry. It's gonna help finance the AI build-out. Because if you can start to finance GPUs at 5% or 6% instead of – I think CoreWeave's lowest financing was low sevens – that actually mathematically changes the cost to finance this build-out."
1:51

Gavin Baker (Gavin Baker) says the disaggregation of inference can extend GPU useful lives from 3-4 years to 10-15. That may single-handedly save private credit and reduce the financing rates for GPUs, which will drive demand and help finance the build-out. "The disaggregation of prefill and inference is going to be amazing for the useful lives of GPU and may single-handedly save private credit. Private credit is in pain from these SaaS loans. But there's a lot of private credit in GPUs too. They were underwriting that to 3-4. The disaggregation of inference means that these GPUs are going to have 10 or 15-year lives. The AI skeptics are like, "Oh, these companies are all cooking their books. The useful life of a GPU is only a year or two. The useful life of a CPU is only four years because the rapid technological change." No. What rapid technological change has done with the disaggregation of prefill and inference is you can put a Cerebras system or Groq LPUs effectively in front of a Hopper or even an Ampere, use that Hopper and Ampere for prefill, and extend the useful life of that GPU until it melts. This is going to be really good for the whole private credit industry. It's gonna help finance the AI build-out. Because if you can start to finance GPUs at 5% or 6% instead of – I think CoreWeave's lowest financing was low sevens – that actually mathematically changes the cost to finance this build-out."

Invest Like the Best

211,336 次观看 • 4 个月前

Jensen Huang just identified the next $200 billion market (Save this). The shift starts with a observation about agentic AI that changes everything about infrastructure. In the era of training and inference, the GPU was everything while CPU was a traffic cop, scheduling work, managing memory, dispatching tasks while the GPU did the heavy lifting. Agentic AI breaks that model entirely. An AI agent does not just run a single inference pass but rather it plans, calls tools, executes code in sandboxes, retrieves data from multiple sources and loops through complex multi-step reasoning sequences often thousands of times per second at scale. Every one of those operations runs through the CPU and the GPU sits idle waiting for the CPU to prepare the next task, supply the right context and execute the retrieval and tool calling logic fast enough to keep the accelerators fed. The CPU is now the conductor and the GPU is the orchestra and the bottleneck is the conductor falling behind. This is showing up in production AI factory utilization right now, which is exactly why Jensen built Vera from scratch rather than licensing x86. Vera achieves 40% lower peak memory latency than x86, 50% faster core to core communication, and 1.8 times the agentic sandbox performance of current x86 processors on a purpose-built architecture designed around the agentic loop. Now here is where the investment thesis gets interesting. The obvious beneficiary is Nvidia itself, and that thesis is real. Nvidia's CFO has guided for nearly $20 billion in Vera CPU revenue this fiscal year alone, a market Nvidia had zero presence in just three years ago. Intel held 60% of server CPU market share as recently as Q4 2025 and that transition is now happening at a pace Intel structurally cannot respond to. But the deeper question is, what architecture is Vera actually built on? Vera's Olympus cores are ARM compatible and every single Vera CPU deployed in every Vera Rubin rack in every data center in the world runs on ARM architecture. And ARM Holdings collects a royalty on every one of them. ARM does not make chips but rather licenses the instruction set architecture and CPU core designs that others build on top of. Every time Nvidia ships a Vera CPU, every time a hyperscaler deploys a Vera Rubin rack, every time an enterprise qualifies Vera for their AI factory, ARM earns a royalty. The secular tailwind here is almost perfectly constructed for ARM's business model. Amazon's Graviton, Microsoft's Cobalt, Google's Axion, Apple's silicon stack, and Qualcomm's data center push all run on ARM. And now Nvidia's Vera, which is projected to displace Intel as the largest server CPU supplier by revenue in a single fiscal year, is ARM. ARM's royalty rate on high end server chips is estimated at roughly 1 to 2% of chip selling price. At $5,000 per Vera CPU and 4 million units projected for FY2027, that is a royalty line growing from near zero to potentially $400 million to $800 million annually from Nvidia's data center CPU business alone before counting Amazon, Microsoft, Google, Apple, and Qualcomm. The total ARM addressable royalty base across all the silicon it already licenses is compounding at a rate that the current $130 billion market cap does not fully reflect. Jensen's CPU thesis is the most underappreciated catalyst in ARM's fundamental story, and the royalty compounding has barely started. Come join Milk Road Pro and get our full ARM royalty model and our entire AI trade thesis. Link below!
3:41

Jensen Huang just identified the next $200 billion market (Save this). The shift starts with a observation about agentic AI that changes everything about infrastructure. In the era of training and inference, the GPU was everything while CPU was a traffic cop, scheduling work, managing memory, dispatching tasks while the GPU did the heavy lifting. Agentic AI breaks that model entirely. An AI agent does not just run a single inference pass but rather it plans, calls tools, executes code in sandboxes, retrieves data from multiple sources and loops through complex multi-step reasoning sequences often thousands of times per second at scale. Every one of those operations runs through the CPU and the GPU sits idle waiting for the CPU to prepare the next task, supply the right context and execute the retrieval and tool calling logic fast enough to keep the accelerators fed. The CPU is now the conductor and the GPU is the orchestra and the bottleneck is the conductor falling behind. This is showing up in production AI factory utilization right now, which is exactly why Jensen built Vera from scratch rather than licensing x86. Vera achieves 40% lower peak memory latency than x86, 50% faster core to core communication, and 1.8 times the agentic sandbox performance of current x86 processors on a purpose-built architecture designed around the agentic loop. Now here is where the investment thesis gets interesting. The obvious beneficiary is Nvidia itself, and that thesis is real. Nvidia's CFO has guided for nearly $20 billion in Vera CPU revenue this fiscal year alone, a market Nvidia had zero presence in just three years ago. Intel held 60% of server CPU market share as recently as Q4 2025 and that transition is now happening at a pace Intel structurally cannot respond to. But the deeper question is, what architecture is Vera actually built on? Vera's Olympus cores are ARM compatible and every single Vera CPU deployed in every Vera Rubin rack in every data center in the world runs on ARM architecture. And ARM Holdings collects a royalty on every one of them. ARM does not make chips but rather licenses the instruction set architecture and CPU core designs that others build on top of. Every time Nvidia ships a Vera CPU, every time a hyperscaler deploys a Vera Rubin rack, every time an enterprise qualifies Vera for their AI factory, ARM earns a royalty. The secular tailwind here is almost perfectly constructed for ARM's business model. Amazon's Graviton, Microsoft's Cobalt, Google's Axion, Apple's silicon stack, and Qualcomm's data center push all run on ARM. And now Nvidia's Vera, which is projected to displace Intel as the largest server CPU supplier by revenue in a single fiscal year, is ARM. ARM's royalty rate on high end server chips is estimated at roughly 1 to 2% of chip selling price. At $5,000 per Vera CPU and 4 million units projected for FY2027, that is a royalty line growing from near zero to potentially $400 million to $800 million annually from Nvidia's data center CPU business alone before counting Amazon, Microsoft, Google, Apple, and Qualcomm. The total ARM addressable royalty base across all the silicon it already licenses is compounding at a rate that the current $130 billion market cap does not fully reflect. Jensen's CPU thesis is the most underappreciated catalyst in ARM's fundamental story, and the royalty compounding has barely started. Come join Milk Road Pro and get our full ARM royalty model and our entire AI trade thesis. Link below!

Milk Road AI

11,819 次观看 • 3 个月前

Andrew Feldman, co-founder and CEO of Cerebras gives the best explanation of why Cerebras' wafer-scale architecture is 2,500X faster than a GPU during LLM inference. During inference, there are 2 stages: - pre-fill, where the model first processes the user's prompt, and - decode, where it generates the answer 1 token at a time in sequence. During that sequencial Decode phase, before each token is calculated the model weights have to be moved from memory into compute. On a GPU those weights are moved from HBM, while Cerebras keeps them in much faster SRAM spread across its very large wafer-scale processor, so the memory-to-compute movement that must happen for every token is about 2,500× faster ---- From The MAD Podcast with Matt Turck and Cerebras YouTube channel, (full video link in comment)
2:48

Andrew Feldman, co-founder and CEO of Cerebras gives the best explanation of why Cerebras' wafer-scale architecture is 2,500X faster than a GPU during LLM inference. During inference, there are 2 stages: - pre-fill, where the model first processes the user's prompt, and - decode, where it generates the answer 1 token at a time in sequence. During that sequencial Decode phase, before each token is calculated the model weights have to be moved from memory into compute. On a GPU those weights are moved from HBM, while Cerebras keeps them in much faster SRAM spread across its very large wafer-scale processor, so the memory-to-compute movement that must happen for every token is about 2,500× faster ---- From The MAD Podcast with Matt Turck and Cerebras YouTube channel, (full video link in comment)

Rohan Paul

39,999 次观看 • 26 天前

New course: Efficient Inference with SGLang: Text and Image Generation, built in partnership with LMSys LMSYS Org and RadixArk RadixArk, and taught by Richard Chen Richard Chen, a Member of Technical Staff at RadixArk. Running LLMs in production is expensive, and much of that cost comes from redundant computation. This short course teaches you to eliminate that waste using SGLang, an open-source inference framework that caches computation already done and reuses it across future requests. When ten users share the same system prompt, SGLang processes it once, not ten times. The speedups compound quickly, especially when there's a lot of shared context across requests. Skills you'll gain: - Implement a KV cache from scratch to eliminate redundant computation within a single request - Scale caching across users and requests with RadixAttention, so shared context is only processed once - Accelerate image generation with diffusion models using SGLang's caching and multi-GPU parallelism Join and learn to make LLM inference faster and more cost-efficient at scale!
1:51

New course: Efficient Inference with SGLang: Text and Image Generation, built in partnership with LMSys LMSYS Org and RadixArk RadixArk, and taught by Richard Chen Richard Chen, a Member of Technical Staff at RadixArk. Running LLMs in production is expensive, and much of that cost comes from redundant computation. This short course teaches you to eliminate that waste using SGLang, an open-source inference framework that caches computation already done and reuses it across future requests. When ten users share the same system prompt, SGLang processes it once, not ten times. The speedups compound quickly, especially when there's a lot of shared context across requests. Skills you'll gain: - Implement a KV cache from scratch to eliminate redundant computation within a single request - Scale caching across users and requests with RadixAttention, so shared context is only processed once - Accelerate image generation with diffusion models using SGLang's caching and multi-GPU parallelism Join and learn to make LLM inference faster and more cost-efficient at scale!

Andrew Ng

102,747 次观看 • 5 个月前

We are excited to share a deeper look into the GPU Sharing Platform that will power AI applications on the InfraX V4 DApp. In this demo we showcase a GPU running and completing an image generation task. Our backend successfully leverages this GPU (and CPU) in real time to complete the job, earning the user money, and powering the infraX platform. Here’s how it works 👉 when you connect to the InfraX DApp, our platform automatically detects the GPU on your machine and immediately checks it eligibility and where it can be used. From there, each user dictates exactly how and when they want to contribute their GPU power by setting the days their PC is most available (throughout the night, for example). Additional, each user will be able to pause their GPU's connection to infraX whenever they need to use it locally, resuming with a just a single click. Once activated, your GPU seamlessly plugs into the InfraX network and begins fulfilling compute jobs, earning YOU money. Go on, take the $INFRA pill..
1:14

We are excited to share a deeper look into the GPU Sharing Platform that will power AI applications on the InfraX V4 DApp. In this demo we showcase a GPU running and completing an image generation task. Our backend successfully leverages this GPU (and CPU) in real time to complete the job, earning the user money, and powering the infraX platform. Here’s how it works 👉 when you connect to the InfraX DApp, our platform automatically detects the GPU on your machine and immediately checks it eligibility and where it can be used. From there, each user dictates exactly how and when they want to contribute their GPU power by setting the days their PC is most available (throughout the night, for example). Additional, each user will be able to pause their GPU's connection to infraX whenever they need to use it locally, resuming with a just a single click. Once activated, your GPU seamlessly plugs into the InfraX network and begins fulfilling compute jobs, earning YOU money. Go on, take the $INFRA pill..

infraX | $INFRA

30,615 次观看 • 1 年前

Dylan Patel of SemiAnalysis says a worse GPU with better storage and memory now beats the best chip without them, so buying the newest GPU alone no longer wins inference. So, an AMD GPU with more memory can outperform Nvidia in some cases. "So what we have is we have over $80 million of compute, GPUs from Nvidia, AMD, TPUs from Google, Trainium from Amazon, and we run this benchmark constantly on the newest inference engine, newest drivers, newest PyTorch version, whatever it is." "Every day it runs on an automated CI, and we run it on all the latest Chinese models, from GLM, Zhipu, Moonshot, Kimi, Alibaba, all these models we run." "Initially, when we were benchmarking the difference between these chips and different engines, different schemes for parallelism, we were just running it fixed context length." "But now with Agent X, we've analyzed over $5 million worth of Claude Code traces. This is real production traffic that people have donated to us as well as internally generated. Now we know what the actual agent workload looks like." "And then as we implement that and run those benchmarks, it turns out yes, the chip you're using is very important, but now even more important is how are you handling this memory offload?" "And so while an Nvidia GPU is faster than an AMD GPU in most cases, because AMD GPUs have more memory, they actually end up outperforming in some cases." "Or you can have a worse GPU, but a much better storage solution, and now you can outperform what the best GPU can do without those solutions. So just buying the newest and latest GPU alone doesn't get you the best inference economics." "Actually, you need to layer in all these other innovations including storage and memory." [ Who's the top player on your chart? ] "That really is a difficult multivariable problem. And generally that means you need to have, yes, you need to have the best GPU, a GB300, but you also need to have the best storage solutions. And so I won't spoil who's the best right here, but I will say that storage solutions matter a lot and memory solutions matter a lot, as does your front-end networking. That matters a lot."
2:00

Dylan Patel of SemiAnalysis says a worse GPU with better storage and memory now beats the best chip without them, so buying the newest GPU alone no longer wins inference. So, an AMD GPU with more memory can outperform Nvidia in some cases. "So what we have is we have over $80 million of compute, GPUs from Nvidia, AMD, TPUs from Google, Trainium from Amazon, and we run this benchmark constantly on the newest inference engine, newest drivers, newest PyTorch version, whatever it is." "Every day it runs on an automated CI, and we run it on all the latest Chinese models, from GLM, Zhipu, Moonshot, Kimi, Alibaba, all these models we run." "Initially, when we were benchmarking the difference between these chips and different engines, different schemes for parallelism, we were just running it fixed context length." "But now with Agent X, we've analyzed over $5 million worth of Claude Code traces. This is real production traffic that people have donated to us as well as internally generated. Now we know what the actual agent workload looks like." "And then as we implement that and run those benchmarks, it turns out yes, the chip you're using is very important, but now even more important is how are you handling this memory offload?" "And so while an Nvidia GPU is faster than an AMD GPU in most cases, because AMD GPUs have more memory, they actually end up outperforming in some cases." "Or you can have a worse GPU, but a much better storage solution, and now you can outperform what the best GPU can do without those solutions. So just buying the newest and latest GPU alone doesn't get you the best inference economics." "Actually, you need to layer in all these other innovations including storage and memory." [ Who's the top player on your chart? ] "That really is a difficult multivariable problem. And generally that means you need to have, yes, you need to have the best GPU, a GB300, but you also need to have the best storage solutions. And so I won't spoil who's the best right here, but I will say that storage solutions matter a lot and memory solutions matter a lot, as does your front-end networking. That matters a lot."

Fireside Alpha

178,673 次观看 • 2 个月前

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:
0:20

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:

Awni Hannun

42,810 次观看 • 1 年前

125B on a single RTX 4090 at 25 tok/s is actually insane Qwen3.8-Flash-Next + MTP is hitting: 80K context 25.35 tok/s decode 471 tok/s prefill On one consumer GPU. Yes, it’s an MoE, so you’re not running all 125B parameters for every token. But still, getting a model this large to run this fast on a 4090 is wild. MTP is doing a lot of the heavy lifting here too. Instead of generating one token at a time, it can speculate on multiple tokens and verify them together. And 25 tok/s is the important number.
0:20

125B on a single RTX 4090 at 25 tok/s is actually insane Qwen3.8-Flash-Next + MTP is hitting: 80K context 25.35 tok/s decode 471 tok/s prefill On one consumer GPU. Yes, it’s an MoE, so you’re not running all 125B parameters for every token. But still, getting a model this large to run this fast on a 4090 is wild. MTP is doing a lot of the heavy lifting here too. Instead of generating one token at a time, it can speculate on multiple tokens and verify them together. And 25 tok/s is the important number.

FHILY👑

39,811 次观看 • 21 天前

Etched is deploying two new technologies in chip design: low-voltage inference and cluster-scale memory. CEO Gavin Uberti says they'll make their chips much more power-efficient and way, way faster than today's leading GPUs. He breaks it down: "We looked at a lot of early research directions, and we realized the key things that models need are way more compute and way faster memory." "If you think about inference, there are two key parts: prefill and decode. For prefill, it's a compute-bound problem. You need to have more FLOPS, more operations per second on each of your chips." "On our GPU, the bottleneck's actually thermals. You can't really run a GPU at more than around 50% of what it could theoretically do, or it'll melt." "So we're using a new technology today called low-voltage inference to try to solve this problem. You bring the voltage of the chip down dramatically, which allows us to have way, way better efficiency in terms of how much power is drawn per unit of math, and thus fit way way more flops onto the chip..." "For decode, it's all about bandwidth. Not just bandwidth on a chip, but bandwidth across your cluster. That's why we have this technology we call cluster-scale memory. It reduces the amount of time it takes to communicate from one chip to another dramatically." "As a result we can go use all of our HBM, HBM bandwidth, SRAM, SRAM bandwidth, and our scale-up domain as a single coherent pool. And that means if you're a user, you can go get much faster tokens per second, while still keeping your costs low."
3:19

Etched is deploying two new technologies in chip design: low-voltage inference and cluster-scale memory. CEO Gavin Uberti says they'll make their chips much more power-efficient and way, way faster than today's leading GPUs. He breaks it down: "We looked at a lot of early research directions, and we realized the key things that models need are way more compute and way faster memory." "If you think about inference, there are two key parts: prefill and decode. For prefill, it's a compute-bound problem. You need to have more FLOPS, more operations per second on each of your chips." "On our GPU, the bottleneck's actually thermals. You can't really run a GPU at more than around 50% of what it could theoretically do, or it'll melt." "So we're using a new technology today called low-voltage inference to try to solve this problem. You bring the voltage of the chip down dramatically, which allows us to have way, way better efficiency in terms of how much power is drawn per unit of math, and thus fit way way more flops onto the chip..." "For decode, it's all about bandwidth. Not just bandwidth on a chip, but bandwidth across your cluster. That's why we have this technology we call cluster-scale memory. It reduces the amount of time it takes to communicate from one chip to another dramatically." "As a result we can go use all of our HBM, HBM bandwidth, SRAM, SRAM bandwidth, and our scale-up domain as a single coherent pool. And that means if you're a user, you can go get much faster tokens per second, while still keeping your costs low."

TBPN

20,404 次观看 • 2 个月前

AMD JUST KILLED THE $200 - A - MONTH AI SUBSCRIPTION - A 65-WATT BOX WITH 128GB OF UNIFIED MEMORY THAT RUNS 70B MODELS LOCALLY. that clip is amd's ryzen ai max+ 395 mini box, held up on stage. the whole pitch is one spec: 128gb of high-speed unified memory. why it matters: cpu, gpu and npu share the same 128gb pool, so a model that used to need a rack of vram just fits. what it actually runs: llama 70b in full precision on the built-in gpu open models up to ~100b parameters, locally, no cloud all of it in a mini pc pulling around 65 watts that's the part that ends the subscription: your prompts, your data, your model, on a box on your desk, not metered by the token. honest line: this runs open models, not gpt-5 or claude opus. for chat, coding help and agents on your own data it's more than enough. for the absolute frontier, the cloud still wins. and it isn't free: the box runs ~$1,500 and up. but that's a few months of a $200 sub, then it's yours forever. no monthly ai bill, no tokens metered, no prompt leaving the room. the uncomfortable part isn't that amd made a fast box. it's that "you need the cloud for ai" was a pricing choice, and unified memory just made it optional. follow RetroChainer for the hardware that turns rent into ownership, and bookmark this before your next ai renewal.
0:41

AMD JUST KILLED THE $200 - A - MONTH AI SUBSCRIPTION - A 65-WATT BOX WITH 128GB OF UNIFIED MEMORY THAT RUNS 70B MODELS LOCALLY. that clip is amd's ryzen ai max+ 395 mini box, held up on stage. the whole pitch is one spec: 128gb of high-speed unified memory. why it matters: cpu, gpu and npu share the same 128gb pool, so a model that used to need a rack of vram just fits. what it actually runs: llama 70b in full precision on the built-in gpu open models up to ~100b parameters, locally, no cloud all of it in a mini pc pulling around 65 watts that's the part that ends the subscription: your prompts, your data, your model, on a box on your desk, not metered by the token. honest line: this runs open models, not gpt-5 or claude opus. for chat, coding help and agents on your own data it's more than enough. for the absolute frontier, the cloud still wins. and it isn't free: the box runs ~$1,500 and up. but that's a few months of a $200 sub, then it's yours forever. no monthly ai bill, no tokens metered, no prompt leaving the room. the uncomfortable part isn't that amd made a fast box. it's that "you need the cloud for ai" was a pricing choice, and unified memory just made it optional. follow RetroChainer for the hardware that turns rent into ownership, and bookmark this before your next ai renewal.

RetroChainer

16,995 次观看 • 1 个月前

Nvidia announces the new RTX Spark, a new platform powered by the NX1 CPU, and shows off Spark laptops running 007 First Light and Forza 6. The CPU has 20 ARM based cores and a Blackwell RTX GPU with 6144 CUDA Cores. This is the same core count as a 5070, but with 128GB of unified LPDDR5X RAM memory sitting in the same package as the CPU and GPU. The entire Nvidia software stack is available, particularly CUDA, vital for AI. Nvidia's new laptops will likely be ideal for running local LLMs be cause the unified memory means you can load models up to 120-180B parameters (quantized). These laptops are expected to ship later this year and could become strong competitors to high-end MacBooks and even Mac Studios for local AI workloads, thanks to CUDA support and unified memory. Price is unannounced.
2:24

Nvidia announces the new RTX Spark, a new platform powered by the NX1 CPU, and shows off Spark laptops running 007 First Light and Forza 6. The CPU has 20 ARM based cores and a Blackwell RTX GPU with 6144 CUDA Cores. This is the same core count as a 5070, but with 128GB of unified LPDDR5X RAM memory sitting in the same package as the CPU and GPU. The entire Nvidia software stack is available, particularly CUDA, vital for AI. Nvidia's new laptops will likely be ideal for running local LLMs be cause the unified memory means you can load models up to 120-180B parameters (quantized). These laptops are expected to ship later this year and could become strong competitors to high-end MacBooks and even Mac Studios for local AI workloads, thanks to CUDA support and unified memory. Price is unannounced.

Grummz

30,573 次观看 • 3 个月前