NVIDIA H200 NVL Tensor Core GPU 141GB
$19,740.00
NVIDIA H200 NVL 141GB is a verified exact-model listing researched against current manufacturer documentation. The NVIDIA H200 NVL 141GB is a Hopper PCIe data-center GPU built for memory-intensive generative AI, inference and high-performance computing. Its defining advantage over H100 NVL is the move to 141GB HBM3e with 4.8 TB/s of memory bandwidth, giving buyers more room for large models, larger batch sizes and memory-heavy scientific workloads without moving to an SXM/HGX-only architecture. Use this page to compare specifications, compatibility, power and deployment requirements before requesting a B2B quotation.
Explore Related
Related products, categories and research.
NVIDIA H200 NVL 141GB: technical buying overview
The NVIDIA H200 NVL 141GB is a Hopper PCIe data-center GPU built for memory-intensive generative AI, inference and high-performance computing. Its defining advantage over H100 NVL is the move to 141GB HBM3e with 4.8 TB/s of memory bandwidth, giving buyers more room for large models, larger batch sizes and memory-heavy scientific workloads without moving to an SXM/HGX-only architecture.
Key NVIDIA H200 NVL 141GB specification highlights
- 141GB HBM3e for larger model and dataset residency.
- 4.8 TB/s memory bandwidth for bandwidth-sensitive AI and HPC workloads.
- PCIe form factor for supported server platforms.
- Configurable power up to 600W, so platform power and airflow must be validated.
- MIG support for partitioned enterprise compute use cases.
NVIDIA H200 NVL 141GB product overview
H200 NVL sits in an important middle ground. It is a server-class PCIe accelerator rather than a desktop graphics card, but it can fit server architectures that use PCIe GPUs instead of an HGX SXM baseboard. NVIDIA lists the card with the same core Hopper compute-class figures used for H100 NVL while substantially increasing memory capacity and bandwidth. That makes the H200 NVL particularly attractive where workload size and data movement are the limiting factors.
Core features and capabilities
NVIDIA lists H200 NVL at 3,341 TFLOPS FP8 Tensor Core performance with sparsity, 1,671 TFLOPS FP16/BF16 Tensor Core with sparsity and 835 TFLOPS TF32 Tensor Core with sparsity. Those figures should be interpreted in the context of software, precision and workload. The more consistently useful buying numbers are the 141GB HBM3e capacity and 4.8 TB/s bandwidth, because they directly affect whether large models and datasets fit efficiently.
Detailed technical specifications
| Architecture | NVIDIA Hopper |
|---|---|
| GPU memory | 141GB HBM3e |
| Memory bandwidth | 4.8 TB/s |
| Form factor | PCIe |
| Max TDP | Up to 600W configurable |
| FP8 Tensor Core | 3,341 TFLOPS with sparsity per NVIDIA table |
| FP16/BF16 Tensor Core | 1,671 TFLOPS with sparsity |
| TF32 Tensor Core | 835 TFLOPS with sparsity |
| MIG | Up to 7 MIG instances; NVIDIA lists 16.5GB profiles for H200 NVL |
| Media engines | 7 NVDEC and 7 JPEG decoders |
Performance and use-case information
NVIDIA lists H200 NVL at 3,341 TFLOPS FP8 Tensor Core performance with sparsity, 1,671 TFLOPS FP16/BF16 Tensor Core with sparsity and 835 TFLOPS TF32 Tensor Core with sparsity. Those figures should be interpreted in the context of software, precision and workload. The more consistently useful buying numbers are the 141GB HBM3e capacity and 4.8 TB/s bandwidth, because they directly affect whether large models and datasets fit efficiently.
Compatibility, software and connectivity
H200 NVL is a PCIe server accelerator, but compatibility is not automatic. Confirm chassis clearance, system airflow, supported GPU count, PCIe topology, auxiliary power, BIOS/firmware support and the OEM’s qualified GPU list. NVIDIA’s CUDA, AI Enterprise and data-center software ecosystem are the relevant software environment; exact driver and platform versions should follow the server vendor’s validated stack.
Applications
- Large-language-model inference
- Generative AI serving
- Model fine-tuning
- HPC and scientific computing
- Data analytics
- Multi-tenant accelerated compute
NVIDIA H200 NVL 141GB comparison with related models
The clearest comparison is H200 NVL versus H100 NVL. H200 NVL increases GPU memory from 94GB to 141GB and bandwidth from roughly 3.9 TB/s to 4.8 TB/s. H100 NVL can still make sense when a validated Hopper server is already deployed or the workload fits comfortably inside 94GB, but H200 NVL is the stronger choice when memory capacity and bandwidth are the constraint. B200 sits above both as a newer Blackwell HGX-class platform rather than a like-for-like PCIe replacement.
What is included
The exact H200 NVL package depends on the OEM, distributor and server configuration. Do not assume that a standalone accelerator includes server rails, power cables, NVLink hardware, licenses or a complete system. The quotation should identify the exact card configuration and all supplied accessories.
Important limitations and buyer considerations
- Server-class PCIe accelerator, not a consumer desktop GPU.
- Up to 600W power requires validated power delivery and cooling.
- Physical fit and airflow depend on the server platform.
- Actual AI throughput varies materially by precision, model, software and batching.
- Licensing/support scope must be confirmed separately.
Who should buy the NVIDIA H200 NVL 141GB?
Choose H200 NVL when you need Hopper software maturity with substantially more memory than H100 NVL, but your infrastructure strategy is based on qualified PCIe GPU servers rather than an HGX B200 platform.
Authoritative technical and manufacturer resources
Manufacturer specifications can be revised. Confirm the exact product code, hardware revision, regional documentation and supplied scope before purchase.
NVIDIA H200 NVL 141GB technical quick answers
How much memory does the NVIDIA H200 NVL 141GB have?
141GB of HBM3e.
What is H200 NVL memory bandwidth?
4.8 TB/s.
Is H200 NVL PCIe or SXM?
NVIDIA lists H200 NVL as PCIe.
What is the maximum power?
Up to 600W configurable.
NVIDIA H200 NVL 141GB FAQ
Is H200 NVL better than H100 NVL?
It provides materially more memory and bandwidth; whether it is better overall depends on workload, server qualification, cost and software stack.
Can I use it in a normal gaming PC?
No. It is intended for qualified server infrastructure.
Does it support MIG?
Yes, NVIDIA lists up to seven MIG instances.
What should a B2B buyer verify?
Exact part, platform qualification, power, airflow, software/driver support, warranty path, lead time and included hardware.
Related products and internal resources
- Shop Graphic Cards
- NVIDIA RTX PRO 6000 Workstation Edition
- NVIDIA RTX PRO 5000 Blackwell 48GB
- ASUS ROG Astral RTX 5090 OC
- ASUS ROG Strix RTX 5070 Ti OC
Request a B2B quotation for NVIDIA H200 NVL 141GB
For business, university, research, data-center or professional procurement, contact AI Robot Supplier with the required quantity, destination country, intended platform or system, and any compatibility requirements. The quotation should confirm the exact model, price, lead time, supplied hardware, shipping terms and applicable warranty/support route before payment.
