llms.txt
+86 −24
https://arrakis.goodailabs.com/llms.txt
− - Live market: 349 GPUs on 120 machines, H100 from $2.28 per GPU-hour (4 Nov 2026, 19:47 UTC).− Updated 4 Nov 2026, 19:47 UTC. Lowest price per GPU-hour, cheapest machine of each GPU, in US dollars. CSV: [/gpus.csv](/gpus.csv).− - B300 SXM6 PC: from $9.19 per GPU-hour, 11 GPUs on 4 machines ([details](/gpus/b300.html.md))− - B200: from $6.77 per GPU-hour, 20 GPUs on 6 machines ([details](/gpus/b200.html.md))− - H200: from $4.43 per GPU-hour, 10 GPUs on 8 machines ([details](/gpus/h200.html.md))− - RTX PRO 6000 Blackwell Server Edition: from $1.73 per GPU-hour, 29 GPUs on 7 machines ([details](/gpus/rtx-pro-6000.html.md))− - RTX 5090: from $0.58 per GPU-hour, 55 GPUs on 14 machines ([details](/gpus/rtx-5090.html.md))− - H100 80GB HBM3: from $2.28 per GPU-hour, 51 GPUs on 9 machines ([details](/gpus/h100.html.md))− - A100 SXM4 80GB: from $1.44 per GPU-hour, 16 GPUs on 2 machines ([details](/gpus/a100.html.md))− - L40S: from $0.44 per GPU-hour, 34 GPUs on 8 machines ([details](/gpus/l40s.html.md))− - RTX PRO 5000 Blackwell: from $0.92 per GPU-hour, 3 GPUs on 2 machines− - L40: from $0.38 per GPU-hour, 2 GPUs on 1 machine− - RTX 4090: from $0.41 per GPU-hour, 27 GPUs on 12 machines ([details](/gpus/rtx-4090.html.md))− - RTX A6000: from $0.46 per GPU-hour, 45 GPUs on 20 machines ([details](/gpus/rtx-a6000.html.md))− - RTX 6000 Ada: from $0.92 per GPU-hour, 15 GPUs on 7 machines ([details](/gpus/rtx-6000-ada.html.md))− - RTX 5080: from $0.33 per GPU-hour, 2 GPUs on 2 machines− - RTX 3090: from $0.22 per GPU-hour, 10 GPUs on 8 machines− - RTX 5070 Ti: from $0.29 per GPU-hour, 4 GPUs on 4 machines− - RTX 5060 Ti: from $0.26 per GPU-hour, 1 GPU on 1 machine− - RTX 4070 Ti SUPER: from $0.26 per GPU-hour, 3 GPUs on 1 machine− - T4 Tensor Core GPU: from $0.26 per GPU-hour, 1 GPU on 1 machine− - CMP 170HX: from $0.35 per GPU-hour, 1 GPU on 1 machine− - RTX PRO 4500 Blackwell Server Edition: from $0.52 per GPU-hour, 1 GPU on 1 machine− - RTX 6000D: from $1.50 per GPU-hour, 8 GPUs on 1 machine+ - Live market: 563 GPUs on 179 machines, H100 from $1.73 per GPU-hour (14 Nov 2026, 07:42 UTC).+ Updated 14 Nov 2026, 07:42 UTC. Lowest price per GPU-hour, cheapest machine of each GPU, in US dollars. CSV: [/gpus.csv](/gpus.csv).+ - B300 SXM6 AC: from $5.41 per GPU-hour, 51 GPUs on 9 machines ([details](/gpus/b300.html.md))+ - B200: from $5.75 per GPU-hour, 14 GPUs on 2 machines ([details](/gpus/b200.html.md))+ - H200: from $4.03 per GPU-hour, 45 GPUs on 12 machines ([details](/gpus/h200.html.md))+ - RTX PRO 6000 Blackwell Server Edition: from $1.61 per GPU-hour, 60 GPUs on 14 machines ([details](/gpus/rtx-pro-6000.html.md))+ - RTX 5090: from $0.57 per GPU-hour, 153 GPUs on 34 machines ([details](/gpus/rtx-5090.html.md))+ - H100 80GB HBM3: from $1.73 per GPU-hour, 50 GPUs on 10 machines ([details](/gpus/h100.html.md))+ - A100 80GB PCIe: from $0.94 per GPU-hour, 1 GPU on 1 machine ([details](/gpus/a100.html.md))+ - L40S: from $0.86 per GPU-hour, 1 GPU on 1 machine ([details](/gpus/l40s.html.md))+ - RTX PRO 5000 Blackwell: from $0.92 per GPU-hour, 1 GPU on 1 machine+ - RTX 4090: from $0.35 per GPU-hour, 75 GPUs on 30 machines ([details](/gpus/rtx-4090.html.md))+ - RTX A6000: from $0.46 per GPU-hour, 18 GPUs on 16 machines ([details](/gpus/rtx-a6000.html.md))+ - RTX 6000 Ada: from $0.92 per GPU-hour, 9 GPUs on 9 machines ([details](/gpus/rtx-6000-ada.html.md))+ - RTX 3090: from $0.19 per GPU-hour, 28 GPUs on 15 machines+ - RTX 5070 Ti: from $0.29 per GPU-hour, 32 GPUs on 11 machines+ - RTX 5060 Ti: from $0.18 per GPU-hour, 12 GPUs on 5 machines+ - GTX 1660 Ti: from $0.12 per GPU-hour, 1 GPU on 1 machine+ - RTX A4500: from $0.20 per GPU-hour, 1 GPU on 1 machine+ - RTX 4060 Ti: from $0.21 per GPU-hour, 3 GPUs on 2 machines+ - RTX 5070: from $0.22 per GPU-hour, 2 GPUs on 1 machine+ - RTX 4070 Ti SUPER: from $0.23 per GPU-hour, 3 GPUs on 2 machines+ - RTX 4080: from $0.25 per GPU-hour, 1 GPU on 1 machine+ - CMP 170HX: from $0.43 per GPU-hour, 2 GPUs on 1 machine+ ## Blog+ Posts for people renting GPUs and for the agents that rent them: what each GPU costs, which one a model needs and how to run it, with real numbers. In English only, grouped by topic, each with a Markdown twin. Every post: [/blog](/blog.html.md). RSS: [/blog/feed.xml](/blog/feed.xml).+ ### Latest models+ - [Qwen-Image 2.1 GPU Guide: Images and Edits From 24 GB](/blog/qwen-image-2-1-gpu.html.md): Qwen-Image 2.1 GPU needs: a 7B image model, 32.4 GB for the whole pipeline in 16-bit. One L40S serves it as an API; a 24 GB card runs it in ComfyUI.+ - [Mistral Large 4 GPU Requirements: 1T Parameters, 52B Active](/blog/mistral-large-4-gpu.html.md): Mistral Large 4 GPU requirements: about 1T parameters in 1,061 GB of weights, so 8x B200 serves it; the NVFP4 copy, 610 GB, fits 4x B300. How to run it now.+ - [MiniMax-H3 GPU Guide: Video With Stereo Audio, Hourly](/blog/minimax-h3-video-gpu.html.md): MiniMax-H3 GPU needs: a 33.1B transformer, 204.8 GB for the whole pipeline in 16-bit. Deploy it as an API on vLLM-Omni or open it in the ComfyUI app.+ - [Qwen3.8-Flash-Next GPU Requirements: One B300 in FP8](/blog/qwen-3-8-flash-next-gpu.html.md): Qwen3.8-Flash-Next GPU requirements: 180B parameters, 360 GB in 16-bit on 4x H200, 185.5 GB in Qwen's FP8 copy on one B300. Deploy it on vLLM in one call.+ - [LTX-2.5 GPU Requirements: Video With Audio, by the Hour](/blog/ltx-2-5-video-gpu.html.md): LTX-2.5 GPU requirements from its files: a 22B transformer at 42 GB in 16-bit, 21.5 GB in int8, 18.7 GB in FP4, plus its text encoder. Which card fits.+ - [Idol Image to 3D: Run It on a Rented GPU by the Hour](/blog/idol-image-to-3d-gpu.html.md): Idol turns one image into a 3D Gaussian splat on 7.2 GB of GPU memory, says its model card. Run the top trending model on a rented RTX GPU by the hour.+ - [Kimi K3.1 Pro GPU Requirements: One 8x B300 Runs It](/blog/run-kimi-k3-1-pro.html.md): Kimi K3.1 Pro GPU requirements: 2.78T parameters published in 4-bit, about 1,669 GB to serve. One 8x B300 holds it. How to run it by the hour.+ - [DeepSeek V4.1 Flash 1017 GPU Guide: 8x H200 Runs It](/blog/deepseek-v4-1-flash-1017-gpu.html.md): DeepSeek V4.1 Flash 1017 GPU needs: 763B parameters in 510 GB of published weights. One 8x H200 serves it with room to spare. Deploy it on vLLM in one call.+ - [FLUX 3 Klein GPU Guide: One RTX 4090 in ComfyUI](/blog/flux-3-klein-gpu.html.md): FLUX 3 Klein GPU needs: a 5.5B image model, an 11.1 GB file in 16-bit. One RTX 4090 runs it in the ComfyUI app. The gated download, the files and the cost.+ ### Start here+ - [GPU Cloud Promo Code: $25 Off Your First $50 Deposit](/blog/first-deposit-25-off.html.md): Our GPU cloud promo code takes $25 off a first deposit of $50 or more until 31 December 2026. What $50 rents at live prices, and how to use it.+ - [Rent a Cloud GPU: Your First Machine in Five Minutes](/blog/rent-first-cloud-gpu.html.md): Rent a cloud GPU in five minutes: sign in, add $20, paste an SSH key, pick a machine and connect as root. What it costs, and how to stop paying on time.+ - [How Much GPU Memory Do I Need? The Plain-English Answer](/blog/how-much-gpu-memory-beginners.html.md): How much GPU memory do I need? Multiply the model's billions of parameters by 2.4, 1.2 or 0.6 GB, then rent the cheapest GPU that holds the result.
Showing the first 40 of each.