# JarvisLabs > JarvisLabs is a GPU cloud for AI and ML teams: on-demand and reserved NVIDIA GPUs (H200, H100, B200, A100 80GB/40GB, RTX 6000 Pro Blackwell, L4) with per-minute billing, 1-click Jupyter and VS Code, SSH access, pre-installed PyTorch/CUDA frameworks, and serverless inference endpoints. Infrastructure is hosted in India with global customer access. JarvisLabs was founded in 2021 to make GPU compute cheap, fast to start, and simple for researchers and engineers. Use the canonical pricing and GPU pages below for live rates and specifications — blog posts and FAQs give context but may lag the pricing pages by a few days. ## GPUs - [NVIDIA H200 on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-h200): 141GB HBM3e GPU for large language model training, fine-tuning, and inference; spec sheet, per-hour and reserved pricing, and benchmarks vs H100. - [NVIDIA H100 on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-h100): 80GB SXM and PCIe H100 availability, per-minute pricing, and use cases for LLM training and high-throughput inference. - [NVIDIA B200 on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-b200): Blackwell-generation GPU for frontier LLM training and inference; architecture overview, memory bandwidth, and early pricing. - [NVIDIA A100 80GB on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-a100-80gb): A100 80GB for large-model training and multi-tenant inference; per-hour rates, reserved discounts, and A100 vs H100 guidance. - [NVIDIA A100 40GB on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-a100-40gb): A100 40GB for mid-scale training and cost-sensitive inference; pricing and workload fit. - [NVIDIA RTX 6000 Pro Blackwell on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-rtx-6000-pro-blackwell): Workstation-class Blackwell GPU for visualization, rendering, and mixed AI/graphics workloads. - [NVIDIA L4 on JarvisLabs](https://jarvislabs.ai/gpu/nvidia-l4): Low-power inference GPU for cost-efficient serving of small and medium LLMs, image generation, and video workloads. - [All GPU pricing](https://jarvislabs.ai/pricing): Full per-hour price list for every GPU JarvisLabs offers, including on-demand and reserved rates. ## Compare JarvisLabs vs alternatives - [JarvisLabs vs RunPod](https://jarvislabs.ai/compare/runpod): Side-by-side pricing, GPU availability, billing model, and regional coverage between JarvisLabs and RunPod. - [JarvisLabs vs Vast.ai](https://jarvislabs.ai/compare/vast-ai): Trade-offs between JarvisLabs' curated infrastructure and Vast.ai's marketplace model, with pricing and reliability comparisons. - [JarvisLabs vs Lambda Labs](https://jarvislabs.ai/compare/lambda-labs): Pricing, H100/H200 availability, and workflow comparison between JarvisLabs and Lambda Labs. - [JarvisLabs vs CoreWeave](https://jarvislabs.ai/compare/coreweave): Enterprise-scale CoreWeave vs developer-friendly JarvisLabs — pricing bands, minimum commitments, and target users. - [JarvisLabs vs Paperspace](https://jarvislabs.ai/compare/paperspace): Per-minute pricing, framework presets, and notebook workflow comparison. ## Guides and benchmarks (blog) - [H100 price and availability](https://jarvislabs.ai/blog/h100-price): Current H100 cloud pricing across major providers, JarvisLabs' H100 rates, and when reserved pricing beats on-demand. - [H200 price and availability](https://jarvislabs.ai/blog/h200-price): Market pricing for H200 instances in 2026, JarvisLabs' H200 rate, and H200 vs H100 cost-per-token analysis. - [H100 vs A100 for LLM training and inference](https://jarvislabs.ai/blog/h100-vs-a100): Architecture, memory, throughput, and price-per-TFLOP comparison for choosing between H100 and A100. - [L4 vs A100 for inference](https://jarvislabs.ai/blog/l4-vs-a100): When L4 is cheaper per token than A100 for serving small-to-medium LLMs and diffusion models. - [vLLM optimization techniques](https://jarvislabs.ai/blog/vllm-optimization-techniques): PagedAttention, continuous batching, chunked prefill, and production tuning knobs for vLLM inference. - [vLLM quantization guide and benchmarks](https://jarvislabs.ai/blog/vllm-quantization-complete-guide-benchmarks): AWQ, GPTQ, FP8, INT4 quantization compared on H100 and A100 with throughput and quality numbers. - [Scaling LLM inference: DP, PP, TP](https://jarvislabs.ai/blog/scaling-llm-inference-dp-pp-tp): Data, pipeline, and tensor parallelism explained with vLLM and TensorRT-LLM deployment examples. - [Speculative decoding in vLLM](https://jarvislabs.ai/blog/speculative-decoding-vllm-faster-llm-inference): How draft-model speculative decoding reduces latency for LLM serving, with benchmarks on H100. ## AI FAQs - [Best cloud GPU providers in 2026](https://jarvislabs.ai/ai-faqs/best-cloud-gpu-providers-2026): Ranked comparison of GPU cloud providers across price, availability, and developer experience for AI teams in 2026. - [NVIDIA A100 vs H100 vs H200](https://jarvislabs.ai/ai-faqs/nvidia-a100-vs-h100-vs-h200-gpu-comparison): Spec-by-spec comparison of A100, H100, and H200 — memory, bandwidth, FP16/BF16 performance, and workload fit. - [NVIDIA H100 vs H200](https://jarvislabs.ai/ai-faqs/nvidia-h100-vs-h200-gpu-comparison): Memory-bandwidth upgrade from H100 to H200 and when the H200 premium pays for itself. - [Best GPU for fine-tuning LLMs](https://jarvislabs.ai/ai-faqs/best-gpu-for-fine-tuning-llms): Which GPU to choose for LoRA, QLoRA, and full fine-tuning of 7B through 70B parameter models. - [Best GPU for Llama 70B](https://jarvislabs.ai/ai-faqs/best-gpu-for-llama-70b): Recommended GPUs and configurations for running and fine-tuning Llama 3 70B, including memory and quantization notes. - [NVLink vs PCIe vs InfiniBand](https://jarvislabs.ai/ai-faqs/what-are-the-key-differences-between-nvlink-and-pcie): Bandwidth, latency, and topology differences and how interconnect choice affects multi-GPU training. - [FP16 vs BF16 memory usage](https://jarvislabs.ai/ai-faqs/what-is-the-memory-usage-difference-between-fp16-and-bf16): Numeric range, precision, and memory footprint trade-offs between FP16 and BF16 for training stability. ## Platform and programs - [JarvisLabs SDK and CLI](https://jarvislabs.ai/sdk): Python SDK and CLI for scripting GPU instance creation, managed runs, and agent workflows from local development environments. - [Universities and research program](https://jarvislabs.ai/universities): Discounted GPU access, credits, and support for academic labs, PhD students, and university research teams. - [Supported countries](https://jarvislabs.ai/supported-countries): Current list of countries JarvisLabs serves and payment methods accepted. ## Optional - [Blog index](https://jarvislabs.ai/blog): Full list of JarvisLabs engineering and ML infrastructure posts. - [AI FAQs index](https://jarvislabs.ai/ai-faqs): Full list of short GPU and AI model Q&A pages. - [Contact sales and support](https://jarvislabs.ai/contactus): Reach the JarvisLabs team for reserved pricing, enterprise, and research inquiries. - [Terms of service](https://jarvislabs.ai/termsandservice): Platform terms and acceptable-use policy.