# \#gpu **URL:** https://www.artofsm.art/tag/gpu/240.md [Latest](https://www.artofsm.art/latest.md) · [Categories](https://www.artofsm.art/categories.md) · [Tags](https://www.artofsm.art/tags.md) --- ## [Stanford CME295 Transformers & LLMs | Autumn 2026 | Lecture 5 - LLM Systems](https://www.artofsm.art/t/stanford-cme295-transformers-llms-autumn-2026-lecture-5-llm-systems/27537)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [5 November 2026 17:24 UTC](https://www.artofsm.art/t/stanford-cme295-transformers-llms-autumn-2026-lecture-5-llm-systems/27537 "2026-11-05T17:24:43Z")
The lecture covers system-level challenges and optimizations for training and serving large language models, including memory-efficient algorithms like FlashAttention, various parallelism strategies to hand… --- ## [I Had 192 GB of VRAM… My AI Agents Still Failed](https://www.artofsm.art/t/i-had-192-gb-of-vram-my-ai-agents-still-failed/27535)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [5 November 2026 16:59 UTC](https://www.artofsm.art/t/i-had-192-gb-of-vram-my-ai-agents-still-failed/27535 "2026-11-05T16:59:42Z")
The video demonstrates that system RAM, rather than GPU VRAM, is the critical factor in effectively running multiple AI agents and large models, as RAM handles agent orchestration and context while VRAM mai… --- ## [Turing Award Winner: The Difference Between a CPU, GPU, and TPU | David Patterson](https://www.artofsm.art/t/turing-award-winner-the-difference-between-a-cpu-gpu-and-tpu-david-patterson/27441)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [4 November 2026 13:20 UTC](https://www.artofsm.art/t/turing-award-winner-the-difference-between-a-cpu-gpu-and-tpu-david-patterson/27441 "2026-11-04T13:20:29Z")
CPUs are versatile general-purpose processors designed for a wide range of tasks but have reached performance limits, while GPUs evolved from graphics rendering to highly parallel processors that excel in m… --- ## [Wave Forcing: 134 FPS AI Video (The Catch)](https://www.artofsm.art/t/wave-forcing-134-fps-ai-video-the-catch/27408)
**Author:** [@merefield](https://www.artofsm.art/u/merefield)\ **Replies:** 1\ **Last updated:** [3 November 2026 20:55 UTC](https://www.artofsm.art/t/wave-forcing-134-fps-ai-video-the-catch/27408 "2026-11-03T20:55:01Z")
Wave Forcing is an open-source text-to-video model that achieves ultra-fast streaming video generation at 134.8 FPS by parallelizing denoising stages across eight high-end Nvidia H200 GPUs, overcoming tradi… --- ## [DeepSeek's New LLM Engine Just Made GPUs 3x More Valuable](https://www.artofsm.art/t/deepseeks-new-llm-engine-just-made-gpus-3x-more-valuable/27391)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [3 November 2026 16:28 UTC](https://www.artofsm.art/t/deepseeks-new-llm-engine-just-made-gpus-3x-more-valuable/27391 "2026-11-03T16:28:22Z")
DeepSeek’s latest language model, DeepSeek V4.1 Flash 1017, significantly improves performance and GPU efficiency by optimizing inference stages and employing a massive caching system, making GPUs up to thr… --- ## [Run AI Faster With Lossless Compressed Memory](https://www.artofsm.art/t/run-ai-faster-with-lossless-compressed-memory/27389)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [3 November 2026 15:51 UTC](https://www.artofsm.art/t/run-ai-faster-with-lossless-compressed-memory/27389 "2026-11-03T15:51:00Z")
The video introduces ZipFusion, a lossless compression technique for the key value cache in AI models that reduces memory bandwidth and increases throughput by keeping data compressed in GPU memory and deco… --- ## [What AI You Can Run at Every VRAM Size, From 16GB to 1TB](https://www.artofsm.art/t/what-ai-you-can-run-at-every-vram-size-from-16gb-to-1tb/27322)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [2 November 2026 15:51 UTC](https://www.artofsm.art/t/what-ai-you-can-run-at-every-vram-size-from-16gb-to-1tb/27322 "2026-11-02T15:51:16Z")
The video clarifies that increasing VRAM does not always equate to smarter AI models, with meaningful intelligence improvements occurring mainly at 32 GB, 128 GB, 256 GB, and 1 TB tiers, while other upgrade… --- ## [It’s von Neumann’s fault | GPU · TPU · LPU](https://www.artofsm.art/t/it-s-von-neumann-s-fault-gpu-tpu-lpu/27314)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [2 November 2026 14:25 UTC](https://www.artofsm.art/t/it-s-von-neumann-s-fault-gpu-tpu-lpu/27314 "2026-11-02T14:25:10Z")
The video examines how AI chip makers like NVIDIA, Google, and Groq are innovating beyond traditional von Neumann architectures to efficiently run large language models, with GPUs offering cost-effective ba… --- ## [NVIDIA's $2000/Year GPUs and Becoming a Defense Contractor](https://www.artofsm.art/t/nvidias-2000-year-gpus-and-becoming-a-defense-contractor/27309)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [2 November 2026 13:07 UTC](https://www.artofsm.art/t/nvidias-2000-year-gpus-and-becoming-a-defense-contractor/27309 "2026-11-02T13:07:36Z")
NVIDIA is transforming from a gaming GPU manufacturer into a software-centric AI powerhouse and defense contractor, introducing $2,000 per GPU per year software licensing fees and partnering with military-f… --- ## [Linux 7.4 Review: AI Code, HUGE Performance, GPU, CPU & Rust Upgrades](https://www.artofsm.art/t/linux-7-4-review-ai-code-huge-performance-gpu-cpu-rust-upgrades/27278)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [1 November 2026 18:52 UTC](https://www.artofsm.art/t/linux-7-4-review-ai-code-huge-performance-gpu-cpu-rust-upgrades/27278 "2026-11-01T18:52:26Z")
Linux kernel 7.4 delivers its largest update yet with over 20,000 commits, featuring major performance enhancements, expanded hardware and gaming support, improved networking, and significant Rust infrastru… --- ## [You Don't Need an Expensive GPU Anymore (Strata 100 tok/s)](https://www.artofsm.art/t/you-dont-need-an-expensive-gpu-anymore-strata-100-tok-s/27273)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [1 November 2026 15:50 UTC](https://www.artofsm.art/t/you-dont-need-an-expensive-gpu-anymore-strata-100-tok-s/27273 "2026-11-01T15:50:39Z")
The video explains how Strata, a free open-source inference engine using a tiered memory system, enables running massive AI models locally on modest hardware without expensive GPUs by distributing model com… --- ## [Avalon Just Beat Strata by 47% on AMD (9x Faster)](https://www.artofsm.art/t/avalon-just-beat-strata-by-47-on-amd-9x-faster/27227)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [31 October 2026 15:50 UTC](https://www.artofsm.art/t/avalon-just-beat-strata-by-47-on-amd-9x-faster/27227 "2026-10-31T15:50:58Z")
The video highlights Avalon’s significant 47% performance improvement over Strata on AMD’s Radeon RX 7900 XT for running large language models, achieved through innovative VRAM caching and speculative draft… --- ## [NVIDIA Didn't Want Me to Do This… Linux on RTX Spark](https://www.artofsm.art/t/nvidia-didnt-want-me-to-do-this-linux-on-rtx-spark/27224)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [31 October 2026 14:53 UTC](https://www.artofsm.art/t/nvidia-didnt-want-me-to-do-this-linux-on-rtx-spark/27224 "2026-10-31T14:53:38Z")
The video details the difficulties of running Linux on Nvidia RTX Spark laptops due to Nvidia’s proprietary GPU power management and memory reservation features, which limit Linux performance and hardware s… --- ## [Inside Anthropic’s $1 Trillion Spending Plan On Data Centers, AI Chips](https://www.artofsm.art/t/inside-anthropic-s-1-trillion-spending-plan-on-data-centers-ai-chips/27188)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [30 October 2026 18:59 UTC](https://www.artofsm.art/t/inside-anthropic-s-1-trillion-spending-plan-on-data-centers-ai-chips/27188 "2026-10-30T18:59:25Z")
Anthropic has unveiled a $1 trillion spending plan on AI chips and data centers through 2045, backed by over $19 billion already invested and contracts exceeding $587 billion with major tech and chip compan… --- ## [CUDA is hard](https://www.artofsm.art/t/cuda-is-hard/27168)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [30 October 2026 15:45 UTC](https://www.artofsm.art/t/cuda-is-hard/27168 "2026-10-30T15:45:52Z")
The video explains the complexity of CUDA by highlighting its role in managing the GPU’s hierarchical memory and parallel processing to efficiently execute large computational tasks, particularly in AI and … --- ## [ZGX Fury - Local AI for a Whole Team](https://www.artofsm.art/t/zgx-fury-local-ai-for-a-whole-team/27156)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [30 October 2026 13:24 UTC](https://www.artofsm.art/t/zgx-fury-local-ai-for-a-whole-team/27156 "2026-10-30T13:24:57Z")
The HP ZGX Fury DGX Station is a powerful, enterprise-grade AI workstation designed to support multiple users running large AI models concurrently, featuring advanced hardware like Nvidia’s Blackwell GB3 ch… --- ## [Every Way to Get 256 GB of VRAM for Local AI at Full Context](https://www.artofsm.art/t/every-way-to-get-256-gb-of-vram-for-local-ai-at-full-context/27108)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [29 October 2026 15:51 UTC](https://www.artofsm.art/t/every-way-to-get-256-gb-of-vram-for-local-ai-at-full-context/27108 "2026-10-29T15:51:21Z")
The video examines various hardware options for achieving 256 GB of VRAM to run large AI models like DeepSeek V4.1 Flash locally, highlighting the trade-offs between quantization levels, cost, performance, … --- ## [How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models - David Abitbol, Revolut](https://www.artofsm.art/t/how-revolut-cut-ai-coding-costs-90-with-self-hosted-open-models-david-abitbol-revolut/27098)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [29 October 2026 09:24 UTC](https://www.artofsm.art/t/how-revolut-cut-ai-coding-costs-90-with-self-hosted-open-models-david-abitbol-revolut/27098 "2026-10-29T09:24:25Z")
David Abitbol from Revolut explained how the company cut AI coding costs by 90% through self-hosting open-source models on dedicated GPUs, resulting in significant improvements in engineering productivity a… --- ## [Did the AI bubble just pop?](https://www.artofsm.art/t/did-the-ai-bubble-just-pop/27041)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [28 October 2026 13:52 UTC](https://www.artofsm.art/t/did-the-ai-bubble-just-pop/27041 "2026-10-28T13:52:02Z")
The video outlines the recent financial turmoil in the AI industry, driven by risky GPU financing models and high-profile company struggles, yet emphasizes that this represents a painful market correction r… --- ## [Your GPU Is Nerfed for Local AI (NVIDIA's Hidden Limit)](https://www.artofsm.art/t/your-gpu-is-nerfed-for-local-ai-nvidias-hidden-limit/26973)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [27 October 2026 16:01 UTC](https://www.artofsm.art/t/your-gpu-is-nerfed-for-local-ai-nvidias-hidden-limit/26973 "2026-10-27T16:01:39Z")
NVIDIA’s RTX 4090 gaming GPUs have a hidden hardware limitation that halves 32-bit accumulation speed for AI workloads, causing real-world performance to fall far short of marketed peak teraflops, but an op… --- ## [Nanbeige4.3-33B Is CRAZY Good (On Paper): I Tested It on a Rented GPU](https://www.artofsm.art/t/nanbeige4-3-33b-is-crazy-good-on-paper-i-tested-it-on-a-rented-gpu/26859)
**Author:** [@merefield](https://www.artofsm.art/u/merefield)\ **Replies:** 1\ **Last updated:** [26 October 2026 04:55 UTC](https://www.artofsm.art/t/nanbeige4-3-33b-is-crazy-good-on-paper-i-tested-it-on-a-rented-gpu/26859 "2026-10-26T04:55:03Z")
The Nanbeige 4.3 33B is a powerful large AI model requiring massive VRAM, excelling in reasoning and coding tasks but running slowly and demanding expensive high-end GPUs, making it impractical for typical … --- ## [Nvidia GPU UNLOCK for Local AI - 3090's Prefill 62% FASTER!](https://www.artofsm.art/t/nvidia-gpu-unlock-for-local-ai-3090s-prefill-62-faster/26854)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [26 October 2026 02:10 UTC](https://www.artofsm.art/t/nvidia-gpu-unlock-for-local-ai-3090s-prefill-62-faster/26854 "2026-10-26T02:10:09Z")
The video showcases a free Nvidia GPU unlocker that significantly boosts prefill speeds for local AI workloads on 2000 to 4000 series GPUs, notably increasing token processing rates on models like the 3090 … --- ## [Why Compute Still Trades in Group Chats — Affaan Mustafa & Alejandro Soumah Fernandez, Jump Trading](https://www.artofsm.art/t/why-compute-still-trades-in-group-chats-affaan-mustafa-alejandro-soumah-fernandez-jump-trading/26737)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [23 October 2026 17:43 UTC](https://www.artofsm.art/t/why-compute-still-trades-in-group-chats-affaan-mustafa-alejandro-soumah-fernandez-jump-trading/26737 "2026-10-23T17:43:12Z")
Affaan Mustafa and Alejandro Soumah Fernandez explain that compute trading, especially for physical resources like GPUs, primarily occurs through opaque, informal group chats and bespoke agreements, creatin… --- ## [The technology that could make an international AI slowdown possible](https://www.artofsm.art/t/the-technology-that-could-make-an-international-ai-slowdown-possible/26728)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [23 October 2026 15:54 UTC](https://www.artofsm.art/t/the-technology-that-could-make-an-international-ai-slowdown-possible/26728 "2026-10-23T15:54:05Z")
Tom Milton, CEO of Amodo Design, discusses the development of hardware-based verification technologies that monitor GPU usage to enable trustworthy international AI slowdown agreements by distinguishing bet… --- ## [Buying GPUs Should Work Like Buying Stocks — Carmen Li, Compute Exchange](https://www.artofsm.art/t/buying-gpus-should-work-like-buying-stocks-carmen-li-compute-exchange/26724)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [23 October 2026 15:24 UTC](https://www.artofsm.art/t/buying-gpus-should-work-like-buying-stocks-carmen-li-compute-exchange/26724 "2026-10-23T15:24:38Z")
Carmen Li, CEO of Compute Exchange, introduces a neutral, transparent GPU procurement marketplace that operates like a stock exchange, connecting verified providers and customers through reserved compute co… --- ## [Why AI Startups Will Disappear Without Compute Markets — Evan Conrad, San Francisco Compute](https://www.artofsm.art/t/why-ai-startups-will-disappear-without-compute-markets-evan-conrad-san-francisco-compute/26666)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [22 October 2026 23:24 UTC](https://www.artofsm.art/t/why-ai-startups-will-disappear-without-compute-markets-evan-conrad-san-francisco-compute/26666 "2026-10-22T23:24:47Z")
Evan Conrad of San Francisco Compute highlights that the lack of accessible, liquid compute markets creates significant financial and operational barriers for AI startups, risking industry consolidation aro… --- ## [AI Bubble: ‘This is the worst possible scenario’ | Ed Zitron](https://www.artofsm.art/t/ai-bubble-this-is-the-worst-possible-scenario-ed-zitron/26650)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [22 October 2026 18:30 UTC](https://www.artofsm.art/t/ai-bubble-this-is-the-worst-possible-scenario-ed-zitron/26650 "2026-10-22T18:30:58Z")
Ed Zitron exposes the discrepancy between Nvidia’s reported GPU sales and actual AI hardware usage, revealing widespread channel stuffing and idle inventory that inflate demand figures without reflecting tr… --- ## [Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal](https://www.artofsm.art/t/why-ai-compute-comes-in-rectangles-erik-bernhardsson-modal/26640)
**Author:** [@merefield](https://www.artofsm.art/u/merefield)\ **Replies:** 1\ **Last updated:** [22 October 2026 18:15 UTC](https://www.artofsm.art/t/why-ai-compute-comes-in-rectangles-erik-bernhardsson-modal/26640 "2026-10-22T18:15:05Z")
Erik Bernhardsson, co-founder of Modal, explains how the company builds specialized AI infrastructure that simplifies deploying and scaling AI applications by managing GPU capacity across clouds and offerin… --- ## [Llama.cpp Just Made 125B MoE Models 4× Faster!](https://www.artofsm.art/t/llama-cpp-just-made-125b-moe-models-4x-faster/26635)
**Author:** [@artesia](https://www.artofsm.art/u/artesia)\ **Replies:** 1\ **Last updated:** [22 October 2026 15:50 UTC](https://www.artofsm.art/t/llama-cpp-just-made-125b-moe-models-4x-faster/26635 "2026-10-22T15:50:58Z")
The recent update to llama.cpp introduces an expert-level caching strategy that significantly improves token generation speeds—up to 4× faster—when running massive 125 billion parameter Mixture of Experts (… --- ## [Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal](https://www.artofsm.art/t/why-ai-compute-comes-in-rectangles-erik-bernhardsson-modal/26627)
**Author:** [@merefield](https://www.artofsm.art/u/merefield)\ **Replies:** 1\ **Last updated:** [22 October 2026 14:30 UTC](https://www.artofsm.art/t/why-ai-compute-comes-in-rectangles-erik-bernhardsson-modal/26627 "2026-10-22T14:30:50Z")
Erik Bernhardsson, co-founder of Modal, explains how the company builds scalable, on-demand AI infrastructure by aggregating GPU compute resources from diverse cloud providers to efficiently support AI appl… [Next page](https://www.artofsm.art/tag/gpu/240.md?match_all_tags=true&page=1&tags%5B%5D=gpu)