| Product SKU | H100-NVL-94GB |
|---|---|
| FP64 | 30 teraFLOPs |
| FP64 Tensor Core | 60 teraFLOPs |
| FP32 | 60 teraFLOPs |
| TF32 Tensor Core * | 835 teraFLOPs |
| BFLOAT16 Tensor Core * | 1,671 teraFLOPS |
| FP16 Tensor Core * | 1,671 teraFLOPS |
| FP8 Tensor Core * | 3,341 teraFLOPS |
| INT8 Tensor Core * | 3,341 TOPS |
| GPU Memory | 94GB |
| GPU Memory Bandwidth | 3.9TB/s |
| Decoders | 7 NVDEC 7 JPEG |
| Max Thermal Design Power (TDP) | 350-400W (configurable) |
| Multi-Instance GPUs | Up to 7 MIGS @ 12GB each |
| Form Factor | PCIe dual-slot air-cooled |
| Interconnect | NVIDIA NVLink: 600GB/s PCIe Gen5: 128GB/s |
| Server Options | Partner and NVIDIA-Certified Systems with 1–8 GPUs |
| NVIDIA AI Enterprise | Included |
Data Center GPUs, GPU
NVIDIA H100 NVL 94GB
The NVIDIA H100 NVL 94GB is a data-center GPU, with 94GB of memory and 3.9TB/s of memory bandwidth. It combines 3,341 teraFLOPS FP8 and 1,671 teraFLOPS FP16 Tensor Core performance with a board power of 350-400 W (configurable) in a PCIe dual-slot air-cooled form factor.
Overview
The NVIDIA H100 NVL 94GB is the PCIe-class member of the H100 family, built for servers that need Hopper-class AI compute without an SXM baseboard. It carries 94 GB of memory at 3.9 TB/s and delivers 3,341 TFLOPS FP8 and 1,671 TFLOPS FP16/BF16, in a 350–400 W configurable, dual-slot air-cooled PCIe card with a 600 GB/s NVLink bridge.
At a glance
- Memory: 94 GB at 3.9 TB/s
- Tensor Core peaks (with sparsity): 3,341 TFLOPS FP8 · 1,671 TFLOPS FP16/BF16 · 835 TFLOPS TF32
- FP64/FP32: 30 TFLOPS FP64 · 60 TFLOPS FP64 Tensor Core · 60 TFLOPS FP32
- Interconnect: NVLink 600 GB/s; PCIe Gen5 128 GB/s
- Form factor: PCIe dual-slot air-cooled; TDP 350–400 W (configurable)
- Multi-Instance GPU: up to 7 MIG instances of 12 GB
- Platforms: partner and NVIDIA-Certified Systems with 1–8 GPUs; AI Enterprise included
When to choose the H100 NVL
Choose the H100 NVL when the servers are standard PCIe platforms rather than HGX/DGX baseboards, and you want the largest H100 memory footprint available on PCIe (94 GB) with an NVLink bridge for pairing two cards. It suits inference and training in 1–8-GPU PCIe servers; for maximum NVLink scale, the SXM5 version is the better fit.
Platform & compatibility
The H100 NVL installs in PCIe Gen5 x16 slots in partner and NVIDIA-Certified systems with 1–8 GPUs, with a 350–400 W configurable TDP and dual-slot air cooling. NVIDIA AI Enterprise is included with the card.
FAQ
Is this the SXM module? No — it is a PCIe dual-slot card with an NVLink bridge, distinct from the SXM5 module.
How much memory does it have? 94 GB at 3.9 TB/s.
Is AI Enterprise included? Yes, it is listed as included with this card.
Related
- All NVIDIA data-center GPUs
- Data-center GPU buying guide: A100 to B300
- Browse all GPUs and accelerators
Configuration, availability and lead time depend on the required quantity, server pairing and delivery destination. Send your requirement to receive a quotation.
