H100 PCIe Price & Used Prices: Which Local LLMs Can It Run?
The H100 PCIe has 80GB of HBM2e memory with 2,000 GB/s of bandwidth. No price data yet. At 4-bit quantization and 8K context it fits a dense model of up to about 136B parameters, at roughly 18 tokens/s.
Xianyu
H100 PCIe price history
Xianyu
Buy links may carry affiliate tags; they do not change the price you pay.
Full H100 PCIe specifications
- Memory80 GB · HBM2e
- Bandwidth2,000 GB/s
- FP16 / BF16756 TFLOPS
- FP8 / INT81,513 / 1,513 TOPS
- Bus width
- 5,120-bit
- Architecture
- Hopper (GH100)
- Cores
- 14,592
- Tensor cores
- 456
- Ecosystem
- CUDA
- Interface
- PCIe 5.0 x16
- NVLink
- 600 GB/s
- Power
- 350 W
- Power connector
- 16-pin
- Slots
- 2
- Length
- 268 mm
- Type
- Data center
- Launch date
- Mar 22, 2022
What LLMs can the H100 PCIe run?
Models up to 50B parameters on a single H100 PCIe, versions released since 2026 only
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 291 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 271.9 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 240.8 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 192.2 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 111 | 256K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 130.8 | 256K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 201.7 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 107.2 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 210.3 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 235.1 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 209.8 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 172.8 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 170.7 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 80.6 | 256K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 81.9 | 256K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 185.4 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 72.7 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 188.3 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 155.5 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 136 | 256K |
| Gemma 4 12B12B | 12.7 GB | Runs | 104.2 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Runs | 145.2 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Runs | 48.9 | 256K |
| Qwen3.8 27B27.8B | 29 GB | Runs | 48.3 | 256K |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Runs | 158.5 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Runs | 42.6 | 256K |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Runs | 162.6 | 256K |
RunsFits fully in VRAM; speed is estimated from VRAM bandwidth.
Needs RAM offloadMoE only: expert weights sit in system RAM, so speed is estimated from 70 GB/s RAM bandwidth.
Too bigWon’t fit on one card at this quantization, even with RAM offload.
Dual, 4x and 8x H100 PCIe for LLMs: which models fit?
Cards needed for 50B+ models released since 2026, with vLLM tensor parallelism at 32K context
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 2 cards | No price | 160 GB | 164 | 700 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 2 cards | No price | 160 GB | 141.4 | 700 W |
 Hy3295B (21B active) | 107.2 GB | 2 cards | No price | 160 GB | 78.4 | 700 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 2 cards | No price | 160 GB | 120.7 | 700 W |
| GLM 5.3744B (40B active) | 253.9 GB | 4 cards | No price | 320 GB | 70.1 | 1,400 W |
| Hy4 preview770B (49B active) | 286.1 GBest. | 8 cards | No price | 640 GB | 54.7 | 2,800 W |
 Kimi K2.7 Code1,000B (32B active) | 339.5 GB | 8 cards | No price | 640 GB | 82.3 | 2,800 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 2 cards | No price | 160 GB | 146 | 700 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 4 cards | No price | 320 GB | 112.5 | 1,400 W |
| Hy3295B (21B active) | 182.2 GB | 4 cards | No price | 320 GB | 61 | 1,400 W |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | 4 cards | No price | 320 GB | 85 | 1,400 W |
| GLM 5.3744B (40B active) | 467.3 GB | 8 cards | No price | 640 GB | 45.3 | 2,800 W |
| Hy4 preview770B (49B active) | 467.3 GB | 8 cards | No price | 640 GB | 39.7 | 2,800 W |
| DeepSeek V4.1 Flash763B (8 / 16B active) | 510.3 GB | 8 cardsEngram in RAM | No price | 640 GB | 75.1 | 2,800 W |
| Kimi K2.7 Code1,000B (32B active) | 583.7 GB | 8 cards | No price | 640 GB | 57.2 | 2,800 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 4 cards | No price | 320 GB | 109.9 | 1,400 W |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | 4 cards | No price | 320 GB | 115.8 | 1,400 W |
| Hy3295B (21B active) | 317.7 GB | 8 cards | No price | 640 GB | 43.6 | 2,800 W |
| GLM 5.3 Flash320B (18B active) | 341 GB | 8 cards | No price | 640 GB | 58.2 | 2,800 W |
| Kimi K2.7 Code1,000B (32B active) | 594.5 GB | 8 cards | No price | 640 GB | 56.5 | 2,800 W |
Card counts are 1 / 2 / 4 / 8, each using 90% of its VRAM. Speed is based on one card’s bandwidth; price covers GPUs only. Rows marked “Engram in RAM” need about 203 GB of extra system RAM. Supports NVLink (600 GB/s). Whether it runs also depends on vLLM support.
H100 PCIe rental price per hour
On-demand rental of the H100 PCIe starts at $1.99 per hour (RunPod), about $484/month at 8 hours a day.
Other markets
eBay
Amazon.jpBuy links may carry affiliate tags; they do not change the price you pay.
FAQ
How large a model can the H100 PCIe run?
A single H100 PCIe has 80GB of HBM2e VRAM. At 4-bit quantization and 8K context it fits a dense model of up to about 136B parameters, or about 73B at 8-bit.
Can the H100 PCIe run a 70B model?
Yes. At 4-bit with an 8K context a 70B dense model needs about 43GB, which fits in the H100 PCIe's 80GB.
What are the best LLMs to run on the H100 PCIe?
Among models released in 2026, the largest dense model that fits entirely in VRAM at 4-bit is Gemma 4 31B, at an estimated 72.7 tokens/s. Among MoE models, Qwen3.6 35B A3B is the fastest at about 188.3 tokens/s. The full list is in the table above.
How many tokens per second does the H100 PCIe get on LLMs?
Estimated from memory bandwidth at 4-bit and 8K context: 8B dense: about 226.5 tokens/s, 14B dense: about 142.8 tokens/s, 32B dense: about 68.8 tokens/s, and 70B dense: about 33.1 tokens/s. MoE models read only the active parameters for each token, so they run much faster. Real-world results are usually 70%–100% of the estimate.
What LLMs can dual H100 PCIe cards run?
Two H100 PCIe cards (160GB total) with vLLM tensor parallelism at 4-bit and 32K context can fit Qwen3.8 Flash Next. To run Kimi K2.7 Code at 4-bit with 32K context, you need 8 cards.
How much does a H100 PCIe cost?
No price data yet; it will appear here once the feed is live.
How much does it cost to rent a H100 PCIe per hour?
As of Sep 29, 2026: RunPod at $1.99 per hour and Vast.ai at $2.59 per hour. All are single-GPU on-demand rates, excluding storage and data transfer.
H100 PCIe vs A100 80GB PCIe: which is better for LLMs?
The A100 80GB PCIe has 80GB of VRAM and 1,935 GB/s of bandwidth; the H100 PCIe has 80GB and 2,000 GB/s. For 70B dense models at 4-bit, the H100 PCIe is estimated at about 33.1 tokens/s and the A100 80GB PCIe at about 32 tokens/s.