## Summary * If you want instant access to instances of 1x H100, use Lambda Labs or FluidStack. * If you want instant access to instances of 1x-8x A100s, use Runpod. * If you want access to a large H100 cluster, you'll need to talk to sales. You'll probably want to talk with FluidStack, Lambda, and Oracle. * I'll personally be using a mix of FluidStack, Lambda, and Runpod. ## Best GPU Clouds for H100 and A100 instant access This table shows instant/on-demand availability and prices of H100s and A100s at the 3 clouds I believe are currently the best options for most people looking for instant access to reliable instances of fast GPUs: FluidStack, Runpod, and Lambda. 👉 **We made a tool for live price and availability updates of H100s and A100s: see [here](https://gpus.llm-utils.org/tracking-h100-and-a100-gpu-cloud-availability/).** June 29th updates (prior update was June 24th): * Lambda and FluidStack both out of capacity on A100 80GB SXM * Lambda now out of capacity on A100 40GB SXM * FluidStack now out of capacity on A100 40GB PCIe | GPU | Runpod June 29th | Lambda June 29th | FluidStack June 29th | |----------------|------------------|------------------|----------------------| | H100 80GB PCIe | No | ✅ $1.99 | ✅ $1.99 | | A100 80GB SXM | ✅ $1.84 | Out of capacity | Out of capacity | | A100 80GB PCIe | ✅ $1.79 | No | $2.91 | | A100 40GB PCIe | No | Out of capacity | Out of capacity | | A100 40GB SXM | No | Out of capacity | No | [See more details, links, and add comments on the google sheet here](https://docs.google.com/spreadsheets/d/1Wh2jLmfVMdDonCi1eFsL5dnLEQOBFKdPdlhYl4c5hPk/edit#gid=83934879). Note that Runpod has instances of up to 6x H100s available instantly, though at higher prices and in the community cloud rather than secure cloud. ## Moved posts Also covers: which GPUs to use for which use cases, best GPU clouds by use case (Falcon-40B, MPT-30B, Stable Diffusion), H100/A100 alternatives and a GPU ranking, options for accessing large H100 clusters, and a long list of alternative GPU clouds. [Summary of best GPUs and clouds by use case](https://gpus.llm-utils.org/recommended-gpus-and-gpu-clouds-for-ai/) [Best GPUs and clouds for running Falcon-40B](https://gpus.llm-utils.org/gpus-for-running-falcon-40b/) [Best GPUs for running MPT-30B](https://gpus.llm-utils.org/gpus-for-running-mpt-30b/) [Best GPUs for running Stable Diffusion](https://gpus.llm-utils.org/gpus-for-running-stable-diffusion/) [Accessing very large numbers of H100s](https://gpus.llm-utils.org/accessing-large-h100-clusters/) [Alternative high end GPUs and their recommended cloud providers](https://gpus.llm-utils.org/gpu-clouds-for-each-gpu/) [Even more options of GPU clouds for running Falcon-40B](https://gpus.llm-utils.org/gpu-clouds-for-falcon-40b/) [DGX GH200 vs GH200 vs H100](https://gpus.llm-utils.org/dgx-gh200-vs-gh200-vs-h100/) [RTX 6000 vs A6000 vs 6000 Ada](https://gpus.llm-utils.org/rtx-6000-vs-a6000-vs-6000-ada/) [VRAM and vCPUs](https://gpus.llm-utils.org/vram-and-vcpus/) [NVLink, SXM, PCIe](https://gpus.llm-utils.org/nvlink-sxm-and-pcie/) [H100 GPU cloud availability and pricing](https://gpus.llm-utils.org/h100-gpu-cloud-availability-and-pricing/) [A100 GPU cloud availability and pricing](https://gpus.llm-utils.org/a100-gpu-cloud-availability-and-pricing/) [Comparison of the user experience at each of the GPU cloud providers](https://gpus.llm-utils.org/gpu-cloud-user-experience-comparison/) [Giant list of additional GPU cloud options](https://gpus.llm-utils.org/alternative-gpu-clouds/) [DGX Cloud](https://gpus.llm-utils.org/dgx-cloud/) [Upcoming and recent updates](https://gpus.llm-utils.org/recent-update-log/) ### Other relevant resources * Read my post on [[Building your own GPU cluster]] from my conversation with César, the CEO of FluidStack. * Listen to my [[Conversation with Zhen Lu, Founder of Runpod.io]] * Listen to my [[Conversation with Doruk, CEO at Oblivus]] * If you're running batch processing jobs where you don't need the GPUs interlinked and you don't need high RAM cards (e.g. stable diffusion, batch whisper processing jobs), read our overview of [[Salad.com GPU cloud - company overview and conversation with Head of Product Daniel Sarfati]]. Meta: * New posts: [Get notified about new posts via email](https://airtable.com/shr411VWRbl9og1xb). * HN: [See the discussion on Hacker News](https://news.ycombinator.com/item?id=36333321).