Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Modal engineer who teaches Full-Stack Deep Learning explained everything an AI engineer needs to know about GPUs in 20 minutes - better than $800 CUDA workshops. memory bandwidth -> SM occupancy -> kernel fusion -> arithmetic intensity -> know when you're compute-bound vs memory-bound. That loop is how you... show more

h100envy

2,239 subscribers

17,766 views • 3 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

PyTorch core engineer at Meta turned CUDA kernel writing into a sport in 13 minutes - better than $1500 GPU programming bootcamps. profile the kernel -> find the bottleneck -> rewrite -> benchmark -> merge the winning code into PyTorch. That loop is how the open community now beats hand-tuned vendor kernels. GPU MODE community + KernelBot competition + winning kernel merged into the framework - that's the stack. Watch it, then steal the loop below.

h100envy

35,390 views • 13 days ago

Ex-Berkeley PhD who leads SGLang at xAI explained how they serve Grok on 100K GPUs in 23 minutes - better than $2000 inference-at-scale courses. split prefill and decode -> shard experts across GPUs -> route tokens per expert -> overlap comm and compute -> serve at DeepSeek-API-killing prices. That loop is why xAI runs Grok on SGLang and third parties beat DeepSeek's own API by 5x on cost. SGLang + prefill-decode disaggregation + expert parallelism + AMD MI300 - that's the stack. Watch and save it, then read the article below.

Ex-Berkeley PhD who leads SGLang at xAI explained how they serve Grok on 100K GPUs in 23 minutes - better than $2000 inference-at-scale courses. split prefill and decode -> shard experts across GPUs -> route tokens per expert -> overlap comm and compute -> serve at DeepSeek-API-killing prices. That loop is why xAI runs Grok on SGLang and third parties beat DeepSeek's own API by 5x on cost. SGLang + prefill-decode disaggregation + expert parallelism + AMD MI300 - that's the stack. Watch and save it, then read the article below.

h100envy

314,507 views • 7 days ago

Alibaba engineer who leads Qwen explained the future of open agent models in 25 minutes - better than $2000 LLM training courses. pre-train the base ->SFT -> RLHF -> tool use -> multi-modal -> ship a whole family (chat / VL / coder / math / QwQ). That loop is why Qwen quietly became the most downloaded open model family on Hugging Face. Qwen base + Qwen-VL + Qwen-Coder + QwQ reasoning - that's the stack. Watch and save it, then read the article below.

Alibaba engineer who leads Qwen explained the future of open agent models in 25 minutes - better than $2000 LLM training courses. pre-train the base ->SFT -> RLHF -> tool use -> multi-modal -> ship a whole family (chat / VL / coder / math / QwQ). That loop is why Qwen quietly became the most downloaded open model family on Hugging Face. Qwen base + Qwen-VL + Qwen-Coder + QwQ reasoning - that's the stack. Watch and save it, then read the article below.

h100envy

112,785 views • 9 days ago

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

481,078 views • 4 days ago

Ex-Google engineer explained AI agent loops, harness, evals in 20 minutes - better than 500$ courses. trace every run → judge it with an LLM → diagnose → fix → ship. That loop is how agents self-improve over time. Agent loops + memory + harness + evals - thats the stack. Watch it, then save the framework below.

Ex-Google engineer explained AI agent loops, harness, evals in 20 minutes - better than 500$ courses. trace every run → judge it with an LLM → diagnose → fix → ship. That loop is how agents self-improve over time. Agent loops + memory + harness + evals - thats the stack. Watch it, then save the framework below.

Movez

553,889 views • 15 days ago

Ex-Google engineer reveals how to build AI agent loops, harnesses, LLM ops, and evals in 19 minutes. Trace → evaluate → diagnose → fix → ship → repeat. That loop is how agents self-improve over time. Agentic loops + harness + memory + evals - that’s the senior engineer stack. This is better than $500 paid courses on the same topic, explained in under 20 minutes. Watch it, then save the framework below.

Ex-Google engineer reveals how to build AI agent loops, harnesses, LLM ops, and evals in 19 minutes. Trace → evaluate → diagnose → fix → ship → repeat. That loop is how agents self-improve over time. Agentic loops + harness + memory + evals - that’s the senior engineer stack. This is better than $500 paid courses on the same topic, explained in under 20 minutes. Watch it, then save the framework below.

Codez

167,880 views • 14 days ago

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 views • 11 days ago

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

Fireside Alpha

38,370 views • 14 days ago

Ex-vLLM core contributor explained how to make LLM inference 10x cheaper in 34 minutes - better than $3000 inference optimization bootcamps. request comes in -> check LMCache -> hit? load KV cache from CPU/SSD/remote -> skip prefill -> serve. That loop is why Bloomberg and other production stacks now push 300 terabytes of KV cache per week. LMCache + vLLM + CPU/SSD/remote storage + zero-copy CUDA kernels - that's the stack. Watch and save it, then wire the KV-offload into your inference stack.

Ex-vLLM core contributor explained how to make LLM inference 10x cheaper in 34 minutes - better than $3000 inference optimization bootcamps. request comes in -> check LMCache -> hit? load KV cache from CPU/SSD/remote -> skip prefill -> serve. That loop is why Bloomberg and other production stacks now push 300 terabytes of KV cache per week. LMCache + vLLM + CPU/SSD/remote storage + zero-copy CUDA kernels - that's the stack. Watch and save it, then wire the KV-offload into your inference stack.

h100envy

58,919 views • 2 days ago

.Beff (e/acc) explains how thermodynamic computing fits into the current computing stack: "We don't do GPUs. They're called TSUs, thermodynamic sampling units. We think people are gonna put them next to GPUs or whatever their favorite accelerator is." "You don't need to run the whole workload on the TSU, but it could run certain parts of the workload, just like the GPU is complementary to the CPU initially." "The TSU is gonna help give you more performance per watt in your data center when combined with existing build-outs." "Let's say you're serving a large diffusion model. What we showed is that by using a TSU, you can use something called an energy-based model instead of a simple probabilistic kernel." "Overall you can have basically 100 times the performance per watt of just running on pure GPUs because our chips are much lower wattage."

.Beff (e/acc) explains how thermodynamic computing fits into the current computing stack: "We don't do GPUs. They're called TSUs, thermodynamic sampling units. We think people are gonna put them next to GPUs or whatever their favorite accelerator is." "You don't need to run the whole workload on the TSU, but it could run certain parts of the workload, just like the GPU is complementary to the CPU initially." "The TSU is gonna help give you more performance per watt in your data center when combined with existing build-outs." "Let's say you're serving a large diffusion model. What we showed is that by using a TSU, you can use something called an energy-based model instead of a simple probabilistic kernel." "Overall you can have basically 100 times the performance per watt of just running on pure GPUs because our chips are much lower wattage."

MTS

42,570 views • 1 month ago

Micron is going to $4,000 and once you understand what inference actually is, the number stops sounding crazy (Save this). Dylan Patel just said that by 2030, OpenAI and Anthropic alone will need over 100 gigawatts of compute combined and by 2040, we may not even be measuring AI infrastructure in gigawatts anymore. We may be talking about terawatts. Every single one of those gigawatts needs memory to function. Without it, the compute is worthless. Most people heard that and thought about Nvidia but they should be thinking about Micron. Every AI model generating a response has two phases. The first is prefill, processing your prompt which is compute-heavy and the second is decode generating each word one token at a time and that phase is almost entirely memory-bound, not compute-bound. During decode, the GPU's processing units sit idle more than 95% of the time, waiting for data to arrive from memory. Google confirmed it in a research paper that decode-phase bottlenecks are dominated by memory bandwidth and capacity not raw compute. The GPU is not the bottleneck but the memory feeding the GPU is. This matters because inference is now where all the money lives. Training a model happens once, Inference happens billions of times a day every ChatGPT response, every Claude output, every agentic workflow running in the background and every one of those token streams is a billing event tied directly to memory performance. Adding more GPUs does not fix this because GPUs are already underutilized in inference because they are sitting idle waiting on memory. Adding more memory bandwidth and capacity is what directly reduces token cost, reduces latency, and allows the same cluster to serve dramatically more users simultaneously. Longer context windows compound the problem further, a model running a 1 million token context window requires dramatically more memory per session than a 10,000 token window, and every new model generation pushes context longer. The market treats memory as a downstream beneficiary of Nvidia orders. The correct framework is the opposite, Micron is the upstream constraint on how much value every Nvidia GPU can actually generate at inference scale. Micron guided Q4 to $50 billion in revenue, has HBM4 ramping at twice the pace of the prior generation, and CEO Sanjay Mehrotra has said supply will not catch demand before the end of 2027. At 8x forward earnings on $112 projected FY2027 EPS, Micron is the most undervalued infrastructure company in the entire AI stack. Inference is memory. Memory is Micron and the inference ramp has barely started. Milk Road Pro members are already up massively on this position and we're just getting started. If you want the full breakdown of what we're buying and why, come join us for just a dollar using the link below!

Micron is going to $4,000 and once you understand what inference actually is, the number stops sounding crazy (Save this). Dylan Patel just said that by 2030, OpenAI and Anthropic alone will need over 100 gigawatts of compute combined and by 2040, we may not even be measuring AI infrastructure in gigawatts anymore. We may be talking about terawatts. Every single one of those gigawatts needs memory to function. Without it, the compute is worthless. Most people heard that and thought about Nvidia but they should be thinking about Micron. Every AI model generating a response has two phases. The first is prefill, processing your prompt which is compute-heavy and the second is decode generating each word one token at a time and that phase is almost entirely memory-bound, not compute-bound. During decode, the GPU's processing units sit idle more than 95% of the time, waiting for data to arrive from memory. Google confirmed it in a research paper that decode-phase bottlenecks are dominated by memory bandwidth and capacity not raw compute. The GPU is not the bottleneck but the memory feeding the GPU is. This matters because inference is now where all the money lives. Training a model happens once, Inference happens billions of times a day every ChatGPT response, every Claude output, every agentic workflow running in the background and every one of those token streams is a billing event tied directly to memory performance. Adding more GPUs does not fix this because GPUs are already underutilized in inference because they are sitting idle waiting on memory. Adding more memory bandwidth and capacity is what directly reduces token cost, reduces latency, and allows the same cluster to serve dramatically more users simultaneously. Longer context windows compound the problem further, a model running a 1 million token context window requires dramatically more memory per session than a 10,000 token window, and every new model generation pushes context longer. The market treats memory as a downstream beneficiary of Nvidia orders. The correct framework is the opposite, Micron is the upstream constraint on how much value every Nvidia GPU can actually generate at inference scale. Micron guided Q4 to $50 billion in revenue, has HBM4 ramping at twice the pace of the prior generation, and CEO Sanjay Mehrotra has said supply will not catch demand before the end of 2027. At 8x forward earnings on $112 projected FY2027 EPS, Micron is the most undervalued infrastructure company in the entire AI stack. Inference is memory. Memory is Micron and the inference ramp has barely started. Milk Road Pro members are already up massively on this position and we're just getting started. If you want the full breakdown of what we're buying and why, come join us for just a dollar using the link below!

Milk Road AI

128,079 views • 13 days ago

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."

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 views • 13 days ago

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.”

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

38,220 views • 3 days ago

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.

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

13,148 views • 1 month ago

INTEL JUST SHIPPED A WORKSTATION CARD WITH TWO GPUs ON ONE PCB AND 48GB OF VRAM. FOUR CARDS GIVE A SINGLE MOTHERBOARD 192GB OF POOLED INFERENCE MEMORY FOR THE PRICE OF TWO RTX 5090s 00:14 he holds one card up to the camera, two GPU dies side by side under the cooler, each one running its own x8 PCIe lane back to the chipset the cluster sees eight discrete accelerators in software. intel's Battle Matrix stack shards a model across all eight, so a 235B parameter network loads in slices and answers requests in parallel what 192GB of VRAM unlocks: an entire 200B class model in memory without quantization. a vision agent reading 100 invoices at once. a research box that holds three frontier models loaded simultaneously, switching between them in under a second intel is the slow side of inference. nvidia is faster per token, that is the honest tradeoff. but the only other path to this much VRAM is a $40,000 nvidia rack or three networked Mac Studios four B60 cards plus the chassis lands under $5,000. power draw averages 800 watts, $55 a month in electricity. one engineer paying $400 a month for combined ChatGPT Pro and Claude Code Max pays the hardware off in less than a year

INTEL JUST SHIPPED A WORKSTATION CARD WITH TWO GPUs ON ONE PCB AND 48GB OF VRAM. FOUR CARDS GIVE A SINGLE MOTHERBOARD 192GB OF POOLED INFERENCE MEMORY FOR THE PRICE OF TWO RTX 5090s 00:14 he holds one card up to the camera, two GPU dies side by side under the cooler, each one running its own x8 PCIe lane back to the chipset the cluster sees eight discrete accelerators in software. intel's Battle Matrix stack shards a model across all eight, so a 235B parameter network loads in slices and answers requests in parallel what 192GB of VRAM unlocks: an entire 200B class model in memory without quantization. a vision agent reading 100 invoices at once. a research box that holds three frontier models loaded simultaneously, switching between them in under a second intel is the slow side of inference. nvidia is faster per token, that is the honest tradeoff. but the only other path to this much VRAM is a $40,000 nvidia rack or three networked Mac Studios four B60 cards plus the chassis lands under $5,000. power draw averages 800 watts, $55 a month in electricity. one engineer paying $400 a month for combined ChatGPT Pro and Claude Code Max pays the hardware off in less than a year

NO1ennn

48,054 views • 14 days ago

The AI boom just hit a wall nobody saw coming. And it's not software. It's not regulation. It's not even energy... It's memory chips. Right now, Dell is raising PC prices by 30%. Intel can't ship chips. Nvidia is slashing GPU production by 40%. And almost nobody understands why. Here's the "hidden" crisis the AI industry is trying to hide: AI data centers are hoarding memory. Not GPUs. Not processors. MEMORY. Every AI server needs massive amounts of high-bandwidth memory (HBM) to run those models everyone's hyping. One problem: There are only 3 companies in the world that can make it. Samsung. SK Hynix. Micron. That's it. And all 3 just diverted their entire production capacity away from normal RAM to feed AI data centers. The math that breaks everything: 1 gigabyte of HBM takes 4X the manufacturing capacity of regular DRAM. AI will consume 20% of global DRAM production in 2026. But the thing is, consumer demand for RAM didn't disappear. PCs still need memory. Phones still need memory. Cars still need memory. But there's no capacity left to make it. The price explosion: RAM prices are up 246% in the last 6 months. DDR5 contract prices jumped 100% month-over-month in some cases. Dell's CFO said he's "never witnessed costs escalating at this pace." SK Hynix and Micron? Sold out through all of 2026. Micron straight up EXITED the consumer memory market entirely to focus on AI customers. If you're not building an AI data center, you're not getting memory chips. AI data centers pay 3-5X margins compared to consumer products. So memory manufacturers are rationally choosing: Serve Microsoft and Google's AI buildout, or serve Dell's laptop business? Easy choice. Every wafer allocated to an Nvidia H100 GPU is a wafer DENIED to your next laptop. It's a zero-sum game. And consumers are losing. The dangerous cascade effect: Nvidia is cutting RTX 50-series GPU production by 30-40% because they can't get GDDR7 memory. Dell, Lenovo, HP are all raising PC prices 15-30% in early 2026. Xiaomi and other smartphone makers are cutting shipment targets. Even Intel's crash last week? Partially driven by memory shortages limiting chip production. This is a PERMANENT reallocation of the world's silicon capacity. Not a temporary supply hiccup. For decades, consumer electronics (phones, PCs, laptops) drove memory production. Now? AI data centers are the priority customer. And that priority shift is reshaping the entire tech economy. The timeline Is worse than you think: Industry analysts project shortages lasting through 2027, maybe 2028. Why? Because building new memory fabs takes 3-5 YEARS. Micron's new Idaho fab won't meaningfully impact supply until 2028. Samsung and SK Hynix are too busy ramping up HBM4 production to expand consumer DRAM. So we're stuck. AI companies need memory to scale. But producing that memory DESTROYS the supply chain for everything else. My question here: Everyone's betting on AI scaling infinitely. But what if the AI boom STALLS because there's not enough memory to support it? What if we're not in an "AI supercycle" but a "memory shortage that kills the AI buildout"? Intel crashed 17% because they can't manufacture enough chips. The root cause though? Memory shortages limiting what they can even produce. Nvidia is cutting GPU production by 40%. AMD is struggling to get GDDR6 for Radeon cards. This isn't just a consumer problem. It's an AI infrastructure problem. And if memory doesn't scale, AI doesn't scale. The AI industry sold you on infinite scaling. But they forgot to mention the part where there's only 3 companies making the memory chips that power everything. And all 3 just chose AI data centers over you. Even Nvidia can't make enough GPUs to meet demand. Not because of energy. Not because of regulation... But because the memory supply chain is BROKEN. And it won't be fixed until 2028.

The AI boom just hit a wall nobody saw coming. And it's not software. It's not regulation. It's not even energy... It's memory chips. Right now, Dell is raising PC prices by 30%. Intel can't ship chips. Nvidia is slashing GPU production by 40%. And almost nobody understands why. Here's the "hidden" crisis the AI industry is trying to hide: AI data centers are hoarding memory. Not GPUs. Not processors. MEMORY. Every AI server needs massive amounts of high-bandwidth memory (HBM) to run those models everyone's hyping. One problem: There are only 3 companies in the world that can make it. Samsung. SK Hynix. Micron. That's it. And all 3 just diverted their entire production capacity away from normal RAM to feed AI data centers. The math that breaks everything: 1 gigabyte of HBM takes 4X the manufacturing capacity of regular DRAM. AI will consume 20% of global DRAM production in 2026. But the thing is, consumer demand for RAM didn't disappear. PCs still need memory. Phones still need memory. Cars still need memory. But there's no capacity left to make it. The price explosion: RAM prices are up 246% in the last 6 months. DDR5 contract prices jumped 100% month-over-month in some cases. Dell's CFO said he's "never witnessed costs escalating at this pace." SK Hynix and Micron? Sold out through all of 2026. Micron straight up EXITED the consumer memory market entirely to focus on AI customers. If you're not building an AI data center, you're not getting memory chips. AI data centers pay 3-5X margins compared to consumer products. So memory manufacturers are rationally choosing: Serve Microsoft and Google's AI buildout, or serve Dell's laptop business? Easy choice. Every wafer allocated to an Nvidia H100 GPU is a wafer DENIED to your next laptop. It's a zero-sum game. And consumers are losing. The dangerous cascade effect: Nvidia is cutting RTX 50-series GPU production by 30-40% because they can't get GDDR7 memory. Dell, Lenovo, HP are all raising PC prices 15-30% in early 2026. Xiaomi and other smartphone makers are cutting shipment targets. Even Intel's crash last week? Partially driven by memory shortages limiting chip production. This is a PERMANENT reallocation of the world's silicon capacity. Not a temporary supply hiccup. For decades, consumer electronics (phones, PCs, laptops) drove memory production. Now? AI data centers are the priority customer. And that priority shift is reshaping the entire tech economy. The timeline Is worse than you think: Industry analysts project shortages lasting through 2027, maybe 2028. Why? Because building new memory fabs takes 3-5 YEARS. Micron's new Idaho fab won't meaningfully impact supply until 2028. Samsung and SK Hynix are too busy ramping up HBM4 production to expand consumer DRAM. So we're stuck. AI companies need memory to scale. But producing that memory DESTROYS the supply chain for everything else. My question here: Everyone's betting on AI scaling infinitely. But what if the AI boom STALLS because there's not enough memory to support it? What if we're not in an "AI supercycle" but a "memory shortage that kills the AI buildout"? Intel crashed 17% because they can't manufacture enough chips. The root cause though? Memory shortages limiting what they can even produce. Nvidia is cutting GPU production by 40%. AMD is struggling to get GDDR6 for Radeon cards. This isn't just a consumer problem. It's an AI infrastructure problem. And if memory doesn't scale, AI doesn't scale. The AI industry sold you on infinite scaling. But they forgot to mention the part where there's only 3 companies making the memory chips that power everything. And all 3 just chose AI data centers over you. Even Nvidia can't make enough GPUs to meet demand. Not because of energy. Not because of regulation... But because the memory supply chain is BROKEN. And it won't be fixed until 2028.

Ricardo

594,453 views • 5 months ago

David Sacks Explains How AI Will Go 1,000,000x in Four Years "I would say the rate of progress is exponential right now on at least three key dimensions." 1) The models "So number one is the algorithms themselves. The models are improving at a rate of, I don't know, 3-4x a year." "They're not just getting faster and better, but qualitatively they're different." "Remember, we started with pure LLM chatbots." "Then we went to reasoning models." "We didn't even get to the agents part of it yet, but that's the next big leap after reasoning models." "We're just starting to scratch the surface there." 2) The chips "Then you've got the chips." "Depending on how you measure it, each generation of chips is probably 3-4x better than the last." "It's not just the individual chips that are getting better, they're figuring out how to network them together." "Like with NVL72, it's like a rack system to create much better performance at the datacenter level." 3) The compute "And that would be the third area where you're seeing basically exponential progress." "Just look at the number of GPUs that are being deployed in datacenters." "So when Elon first started training Grok, I think they had maybe 100K GPUs. Now they're up to 300K. They're on their way to a million. Same thing with OpenAI's data center, Stargate." "And within a couple years they'll be at, I don't know, 5M GPUs, 10M GPUs? How Sacks gets to 1,000,000x: "The algorithms, the chips, and the datacenters are all improving or scaling at a rate of 3-4x a year." "That's 10x every two years." "Where people don't understand exponential progress is that if you're getting better at 10x every two years, that doesn't mean you'll be at 20x in four years." "It means you'll be at a 100x." "So you multiply those things together: the algorithms, the chips, and the raw compute that's available." 100x models 🧠 x 100x chips 💾 x 100x compute ⚡️ = 1,000,000x AI 🤖 "You're talking about 1,000,000x increase." "Some of which will be captured in price reductions, some of it will be in the performance ceiling, and then some of it will just be in the overall amount of AI compute that's available to the economy." "But the impact of this thing is gonna be absolutely massive." "And I think people still don't even appreciate that fact because they don't understand exponential progress."

David Sacks Explains How AI Will Go 1,000,000x in Four Years "I would say the rate of progress is exponential right now on at least three key dimensions." 1) The models "So number one is the algorithms themselves. The models are improving at a rate of, I don't know, 3-4x a year." "They're not just getting faster and better, but qualitatively they're different." "Remember, we started with pure LLM chatbots." "Then we went to reasoning models." "We didn't even get to the agents part of it yet, but that's the next big leap after reasoning models." "We're just starting to scratch the surface there." 2) The chips "Then you've got the chips." "Depending on how you measure it, each generation of chips is probably 3-4x better than the last." "It's not just the individual chips that are getting better, they're figuring out how to network them together." "Like with NVL72, it's like a rack system to create much better performance at the datacenter level." 3) The compute "And that would be the third area where you're seeing basically exponential progress." "Just look at the number of GPUs that are being deployed in datacenters." "So when Elon first started training Grok, I think they had maybe 100K GPUs. Now they're up to 300K. They're on their way to a million. Same thing with OpenAI's data center, Stargate." "And within a couple years they'll be at, I don't know, 5M GPUs, 10M GPUs? How Sacks gets to 1,000,000x: "The algorithms, the chips, and the datacenters are all improving or scaling at a rate of 3-4x a year." "That's 10x every two years." "Where people don't understand exponential progress is that if you're getting better at 10x every two years, that doesn't mean you'll be at 20x in four years." "It means you'll be at a 100x." "So you multiply those things together: the algorithms, the chips, and the raw compute that's available." 100x models 🧠 x 100x chips 💾 x 100x compute ⚡️ = 1,000,000x AI 🤖 "You're talking about 1,000,000x increase." "Some of which will be captured in price reductions, some of it will be in the performance ceiling, and then some of it will just be in the overall amount of AI compute that's available to the economy." "But the impact of this thing is gonna be absolutely massive." "And I think people still don't even appreciate that fact because they don't understand exponential progress."

The All-In Podcast

486,198 views • 1 year ago

Windsurf CEO explains why traditional startup moats are mostly silly. Varun said even with 50-100 of the best MIT engineers, that's still just "hundreds of engineering years" in your product. Someone else can build something similar in the same space. He used Nvidia as an example. People think CUDA is their real moat, but that's not accurate. Large companies like Google, OpenAI, and Anthropic spend tens of billions on chips. If CUDA didn't exist, they'd find a way to write assembly code and make GPUs work. People use CUDA because Nvidia made it really capable at low floating-point math, great interconnect, and fast computers. Not because they're locked in. Every year Nvidia has pressure - if they don't make hardware, interconnect, and memory bandwidth much faster, their profit margins shrink and AMD will compete. For startups, the only moat is speed. Learning where the dead bodies are. Understanding what doesn't work and building compounding advantages through that knowledge. Speed lets you learn from the market faster, which means you're first to the next idea too.

Windsurf CEO explains why traditional startup moats are mostly silly. Varun said even with 50-100 of the best MIT engineers, that's still just "hundreds of engineering years" in your product. Someone else can build something similar in the same space. He used Nvidia as an example. People think CUDA is their real moat, but that's not accurate. Large companies like Google, OpenAI, and Anthropic spend tens of billions on chips. If CUDA didn't exist, they'd find a way to write assembly code and make GPUs work. People use CUDA because Nvidia made it really capable at low floating-point math, great interconnect, and fast computers. Not because they're locked in. Every year Nvidia has pressure - if they don't make hardware, interconnect, and memory bandwidth much faster, their profit margins shrink and AMD will compete. For startups, the only moat is speed. Learning where the dead bodies are. Understanding what doesn't work and building compounding advantages through that knowledge. Speed lets you learn from the market faster, which means you're first to the next idea too.

Aish

334,732 views • 1 year ago

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below

John Doe

40,605 views • 1 month ago

The bottleneck in AI has quietly shifted. - It's not the models. They are capable. - It's not the frameworks. They are mature. - It's not even the data, in many cases. When you want to train a model today, the first question isn't "what architecture should I use?" Instead, it's: "Where am I going to get infrastructure that actually works?" Not just GPUs but the entire stack: compute, deployment, scaling, storage. The traditional path is major cloud providers or specialized GPU clouds. Both have the same problem: they're built for enterprises with committed workloads, minimum spend requirements, contract negotiations, and involve quota approvals that take days. Even the "on-demand" options require you to piece together training, deployment, and scaling across different services. By the time you're actually training, hours, if not days, have passed. And there's a subtler cost: part of your brain is always managing infrastructure instead of thinking about the actual problem. I've been using Runpod for a while now, and it's the closest I've found to infrastructure that just disappears. I pay for the serverless solution by the second, and stop when I'm done. This sounds like it should be the default across all providers, but it isn't. For instance, when I'm prototyping, I don't need an H100. Instead, I need the flexibility to use cheaper GPUs that are actually available, where I can iterate fast and not worry about cost. An A40 at a few cents per hour is perfect for this. Then, when the approach is validated, I scale up. This matches how good engineering actually works. Running distributed training across multiple nodes for multi-GPU training usually requires significant infra work. RunPod abstracts most of this away. A lot of the advantage in AI comes from iteration speed. Infra that adds days of latency to that loop is a real cost, even if it's hard to measure. But good infra gets out of your way. It's available when you need it, invisible when you don't. In the video below, I have shown a simple model training workflow trained using PyTorch in Jupyter Lab. It runs in a dedicated PyTorch Pod hosted on Runpod, and I worked with the team to put this together for you. Find a link to start using Runpod in the replies!

The bottleneck in AI has quietly shifted. - It's not the models. They are capable. - It's not the frameworks. They are mature. - It's not even the data, in many cases. When you want to train a model today, the first question isn't "what architecture should I use?" Instead, it's: "Where am I going to get infrastructure that actually works?" Not just GPUs but the entire stack: compute, deployment, scaling, storage. The traditional path is major cloud providers or specialized GPU clouds. Both have the same problem: they're built for enterprises with committed workloads, minimum spend requirements, contract negotiations, and involve quota approvals that take days. Even the "on-demand" options require you to piece together training, deployment, and scaling across different services. By the time you're actually training, hours, if not days, have passed. And there's a subtler cost: part of your brain is always managing infrastructure instead of thinking about the actual problem. I've been using Runpod for a while now, and it's the closest I've found to infrastructure that just disappears. I pay for the serverless solution by the second, and stop when I'm done. This sounds like it should be the default across all providers, but it isn't. For instance, when I'm prototyping, I don't need an H100. Instead, I need the flexibility to use cheaper GPUs that are actually available, where I can iterate fast and not worry about cost. An A40 at a few cents per hour is perfect for this. Then, when the approach is validated, I scale up. This matches how good engineering actually works. Running distributed training across multiple nodes for multi-GPU training usually requires significant infra work. RunPod abstracts most of this away. A lot of the advantage in AI comes from iteration speed. Infra that adds days of latency to that loop is a real cost, even if it's hard to measure. But good infra gets out of your way. It's available when you need it, invisible when you don't. In the video below, I have shown a simple model training workflow trained using PyTorch in Jupyter Lab. It runs in a dedicated PyTorch Pod hosted on Runpod, and I worked with the team to put this together for you. Find a link to start using Runpod in the replies!

Avi Chawla

13,696 views • 6 months ago