H200 vs H100 — Specifications
The H200 is a drop-in upgrade to the H100 — same GH100 die, same 700W TDP. The difference: 141 GB HBM3e at 4.8 TB/s vs H100's 80 GB at 3.35 TB/s.
| Specification | NVIDIA H100 | NVIDIA H200 |
|---|---|---|
| Architecture | GH100 (Hopper) | GH100 (Hopper) |
| CUDA Cores | 16,896 | 16,896 |
| Tensor Cores | 528 (4th Gen) | 528 (4th Gen) |
| FP8 Tensor | 1,979 TFLOPS | 1,979 TFLOPS |
| Memory | 80 GB HBM3 | 141 GB HBM3e |
| Bandwidth | 3.35 TB/s | 4.8 TB/s |
| PCIe | Gen5 ×16 | Gen5 ×16 |
| TDP | 700W | 700W |
| Form Factor | SXM5 / PCIe | SXM5 / PCIe |
| Price (Q3 2026) | $27,000–$35,000 | $30,000–$40,000 |
LLM Inference Benchmarks
vLLM 0.6.0, FP8 quantization, 2048 input / 512 output tokens, batch size 32:
| Model | H100 (tok/s) | H200 (tok/s) | Speedup |
|---|---|---|---|
| Llama 3.1 8B (FP8) | 4,820 | 5,010 | +3.9% |
| Llama 3.1 70B (FP8) | 2,180 | 3,340 | +53.2% |
| Llama 3.1 70B (FP16) | 880 (offloads) | 1,790 | +103% |
| Mixtral 8×7B (FP8) | 2,640 | 4,120 | +56.1% |
H200 is memory-bandwidth-limited: for models fitting in 80 GB, gains are modest. For 70B+ models at FP16, H200 doubles throughput because H100 can't fit the model in VRAM.
Full Product Line
| GPU | VRAM | Bandwidth | FP8 TFLOPS | PCIe | Price (USD) | Stock |
|---|---|---|---|---|---|---|
| NVIDIA B300 (DGX) | 288 GB HBM3e | 8.0 TB/s | — | SXM6 | $300K–$350K (DGX) | IN STOCK |
| NVIDIA B200 | 180 GB HBM3e | 8.0 TB/s | 4,500 | Gen5 ×16 | $45K–$55K | IN STOCK |
| NVIDIA RTX PRO 6000 Blackwell | 96 GB GDDR7 ECC | 1.8 TB/s | ~125 FP16 | Gen5 ×16 | $8.5K–$12K | IN STOCK |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | 96 GB GDDR7 ECC | 1.8 TB/s | ~125 FP16 | Gen5 ×16 | $8K–$10K | IN STOCK |
| NVIDIA H200 | 141 GB HBM3e | 4.8 TB/s | 1,979 | Gen5 ×16 | $30K–$40K | IN STOCK |
| NVIDIA H100 | 80 GB HBM3 | 3.35 TB/s | 1,979 | Gen5 ×16 | $27K–$35K | IN STOCK |
| NVIDIA L40S | 48 GB GDDR6 | 864 GB/s | 733 | Gen4 ×16 | $9K–$12K | IN STOCK |
| RTX 6000 Pro | 96 GB GDDR7 | 1.8 TB/s | — | Gen5 ×16 | $8K–$9K | IN STOCK |
| NVIDIA RTX 5090 | 32 GB GDDR7 | 1.79 TB/s | — | Gen5 ×16 | $3.5K–$5.0K | IN STOCK |
| NVIDIA A100 SXM4 | 80 GB HBM2e | 2.0 TB/s | — | SXM4 | $13K–$15K | IN STOCK |
| NVIDIA A100 PCIe | 80 GB HBM2e | 2.0 TB/s | — | Gen4 ×16 | $12K–$14K | IN STOCK |
| NVIDIA H100 NVL | 94 GB HBM3 | 3.9 TB/s | — | Gen5 ×16 | $27K–$30K | IN STOCK |
| NVIDIA L20 | 48 GB GDDR6 | 864 GB/s | 59.8 FP8 | Gen4 ×16 | $7.5K–$8.5K | IN STOCK |
| NVIDIA RTX A6000 | 48 GB GDDR6 | 768 GB/s | 38.7 FP32 | Gen4 ×16 | $4.5K–$5.5K | IN STOCK |
| NVIDIA RTX 4000 Ada | 20 GB GDDR6 | 360 GB/s | 26.7 FP32 | Gen4 ×16 | $1.2K–$1.4K | IN STOCK |
Need H100/H200 for your AI cluster? Bulk pricing available for 8+ units.
Request GPU QuoteOfficial Part Numbers (MPN)
| GPU | Form Factor | NVIDIA MPN |
|---|---|---|
| NVIDIA B200 180GB | SXM6 | 900-2G133-4000-000 |
| NVIDIA RTX PRO 6000 Blackwell 96GB | PCIe | 900-2G153-0000-000 |
| NVIDIA H200 141GB | SXM5 | 900-21001-1000-000 |
| NVIDIA H100 80GB | SXM5 | 900-21001-0000-000 |
| NVIDIA H100 80GB | PCIe | 900-21001-0010-000 |
| NVIDIA L40S 48GB | PCIe | 900-2G133-0080-000 / 900-2G133-0380-030 |
| RTX 6000 Pro 96GB | PCIe | VCNRTX6000ADA-PB |
| NVIDIA RTX 5090 32GB | PCIe | PG145 SKU 30 |
| NVIDIA A100 80GB | SXM4 | 900-2G133-0000-000 |
| NVIDIA A100 80GB | PCIe | 900-2G133-0010-000 |
| NVIDIA H100 NVL 94GB | PCIe | 900-21001-0050-000 |
| NVIDIA L20 48GB | PCIe | 900-2G133-0100-000 |
| NVIDIA RTX A6000 48GB | PCIe | VCNRTXA6000-PB |
| NVIDIA RTX 4000 Ada 20GB | PCIe | VCNRTX4000ADA-PB |