NVIDIA B200 SXM6 Tensor Core GPU 180GB
$21,000.00
NVIDIA B200 SXM6 180GB is a verified exact-model listing researched against current manufacturer documentation. The NVIDIA B200 SXM6 180GB is a Blackwell data-center accelerator for HGX B200 systems, built for large-scale AI training, inference and high-performance computing where memory capacity, memory bandwidth and multi-GPU scale matter more than desktop graphics features. Buyers should treat this as platform infrastructure, not as a conventional PCIe graphics card: server qualification, power delivery, cooling, HGX baseboard compatibility and software stack are part of the purchasing decision. Use this page to compare specifications, compatibility, power and deployment requirements before requesting a B2B quotation.
Explore Related
Related products, categories and research.
NVIDIA B200 SXM6 180GB: technical buying overview
The NVIDIA B200 SXM6 180GB is a Blackwell data-center accelerator for HGX B200 systems, built for large-scale AI training, inference and high-performance computing where memory capacity, memory bandwidth and multi-GPU scale matter more than desktop graphics features. Buyers should treat this as platform infrastructure, not as a conventional PCIe graphics card: server qualification, power delivery, cooling, HGX baseboard compatibility and software stack are part of the purchasing decision.
Key NVIDIA B200 SXM6 180GB specification highlights
- 180GB HBM3e per GPU gives substantially more accelerator memory than H100 NVL and H200 NVL PCIe-class products.
- Up to 8 TB/s of memory bandwidth per GPU is aimed at memory-intensive AI and HPC workloads.
- HGX B200 scales to eight GPUs on a validated baseboard rather than relying on ad-hoc workstation integration.
- Blackwell supports modern CUDA software and current NVIDIA AI infrastructure tooling.
- MIG support allows administrators to partition accelerator resources for selected multi-tenant compute workflows.
NVIDIA B200 SXM6 180GB product overview
NVIDIA B200 is designed for tightly coupled accelerator systems. In an HGX B200 8-GPU baseboard, eight Blackwell GPUs are connected as a scale-up platform with 1.44TB of HBM3e memory in aggregate. The practical value is the ability to keep very large models and working sets closer to GPU compute while reducing pressure on host memory and inter-node traffic. That makes the B200 relevant to enterprise generative AI, large-language-model training, high-throughput inference, recommender systems, scientific simulation and other workloads that can exploit tensor compute and high memory bandwidth.
Core features and capabilities
The most important performance characteristic for procurement is not a single synthetic benchmark. B200 is an HGX-scale accelerator. NVIDIA’s current enterprise reference architecture lists 180GB HBM3e and up to 8 TB/s bandwidth per GPU, with 1.44TB memory and up to 64 TB/s aggregate bandwidth across an eight-GPU node. Those figures explain why the platform is positioned above Hopper-generation HGX H100/H200 for large AI models and bandwidth-sensitive workloads.
Detailed technical specifications
| Architecture | NVIDIA Blackwell |
|---|---|
| Deployment class | SXM GPU used in validated HGX B200 server platforms |
| GPU memory | 180GB HBM3e per GPU |
| Memory bandwidth | Up to 8 TB/s per GPU |
| 8-GPU HGX memory | 1.44TB HBM3e aggregate |
| 8-GPU HGX memory bandwidth | Up to 64 TB/s aggregate |
| Power | Configurable up to 1 kW per GPU in HGX B200 |
| Compute capability | Blackwell compute capability 10.0 |
| Virtualization | MIG profiles supported on HGX B200 |
| Cooling / mechanics | Platform-integrated SXM/HGX thermal design; not a desktop add-in card |
Performance and use-case information
The most important performance characteristic for procurement is not a single synthetic benchmark. B200 is an HGX-scale accelerator. NVIDIA’s current enterprise reference architecture lists 180GB HBM3e and up to 8 TB/s bandwidth per GPU, with 1.44TB memory and up to 64 TB/s aggregate bandwidth across an eight-GPU node. Those figures explain why the platform is positioned above Hopper-generation HGX H100/H200 for large AI models and bandwidth-sensitive workloads.
Compatibility, software and connectivity
Do not buy the NVIDIA B200 SXM6 180GB as if it were a standalone workstation GPU. It requires a server platform designed for B200/HGX integration, correct baseboard, firmware, CPU and I/O topology, platform power delivery and validated cooling. CUDA and NVIDIA’s data-center software ecosystem provide the software layer, but actual driver, firmware and system qualification should follow the chosen OEM or server platform bill of materials.
Applications
- LLM training and fine-tuning
- High-throughput generative AI inference
- Large recommender and ranking systems
- Scientific computing and simulation
- Enterprise AI factories and private AI infrastructure
- Multi-tenant GPU compute where supported by the platform
NVIDIA B200 SXM6 180GB comparison with related models
Compared with H200, B200 moves to Blackwell and increases per-GPU memory from 141GB to 180GB while NVIDIA’s HGX reference data shows memory bandwidth rising from 4.8 TB/s to as much as 8 TB/s. Compared with H100 NVL, the deployment model is also fundamentally different: H100 NVL is a PCIe card, while B200 SXM6 is intended for an HGX-scale platform. A buyer choosing between them should compare server architecture, model size, node scale, power budget and software validation rather than price alone.
What is included
NVIDIA does not define a universal retail accessory bundle for an individual B200 SXM module. Delivery contents depend on the server/OEM configuration. The quotation and packing list should state whether the purchase is a GPU module, HGX baseboard, complete server, related cooling hardware, cables, rails or software entitlement. Do not assume these items are included unless they are listed.
Important limitations and buyer considerations
- Not a normal desktop or workstation add-in card.
- Requires validated server/HGX platform integration.
- Per-GPU power can reach the 1 kW class, making facility power and cooling planning essential.
- Exact server dimensions and system weight depend on the OEM chassis, not the GPU alone.
- Software licensing and support entitlements can be separate from hardware.
Who should buy the NVIDIA B200 SXM6 180GB?
This product is appropriate for enterprises, cloud or hosting operators, research institutions and AI infrastructure teams that already understand HGX-class deployment or are procuring a complete validated B200 system. It is not a sensible purchase for a buyer looking for a single desktop GPU.
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 B200 SXM6 180GB technical quick answers
How much memory does the NVIDIA B200 SXM6 180GB have?
180GB HBM3e per GPU.
Is the NVIDIA B200 SXM6 180GB a PCIe graphics card?
No. It is an SXM/HGX data-center accelerator that belongs in a validated B200 server platform.
What is the memory bandwidth?
NVIDIA lists up to 8 TB/s per B200 GPU in current HGX reference documentation.
How much power can it require?
HGX B200 documentation states each GPU is configurable up to 1 kW.
NVIDIA B200 SXM6 180GB FAQ
Can I install it in a normal workstation?
No. B200 SXM is not designed as a normal PCIe workstation add-in card.
Does one purchase automatically include an HGX baseboard?
Not necessarily. The exact scope must be confirmed in the quotation and packing list.
What should I confirm before ordering?
Server model, HGX/B200 compatibility, CPU and network topology, power, cooling, firmware, software support, lead time and exact supplied hardware.
Is B200 mainly for gaming?
No. Its target is enterprise AI and accelerated computing infrastructure.
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 B200 SXM6 180GB
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.
