NVIDIA H100 PCIe 96 GB
Live eBay prices, cheapest first. The best one right now is $29,802 (new).
Other sizes: 80 GB
Specs
- VRAM
- 96 GB HBM2e
- Bandwidth
- 2,000 GB/s
- FP32
- 51 TFLOPS
- FP16
- 756 TFLOPS
- Power
- 350 W
- Connector
- PCIe 5.0
- Released
- 2022
- Launch price
- $25,000 (est.)
- CUDA compute
- 9.0
- Llama 3.1 8B at Q4 (est.)
- 206–309 tokens/s
Running it at home
- Cooling
- Passive: it has no fan. In a desktop it needs air pushed through it, front to back: a fan shroud on its end (3D-printed ones with a 40 mm blower or a 92 mm fan are sold for most of these cards) or a server chassis. Without one it overheats within minutes under load.
- Power
- One 16-pin (12VHPWR / 12V-2x6) power input. Rated 350 W: allow that for each card on top of the rest of the PC.
- Size
- Full height, two slots.
- Motherboard
- Turn on "Above 4G decoding" in the BIOS: without it many boards won't start with a card with this much memory.
- Screen
- No display outputs: use the processor's graphics or a small second card for a monitor.
A server card: made for a rack server's airflow and power. With these sorted it works in a normal PC.
Price right now
Live on eBayFrom the live listings below; shipping included.
Listings
Live on eBay| Listing | Condition | Price | Checked |
|---|---|---|---|
| New | $29,802or best offerfree shipping |
Auctions ending soon
No auctions ending soon right now.
Prices include shipping and were read from eBay at the time shown. They can change: check the listing before you buy.
Owner reviews
No owner reviews yet. Own one? Tell others how it runs.
From signed-in owners, checked before they're published. PerGB doesn't pay for or edit reviews.
How it compares
Live on eBay- Software: CUDA 11.8 and newer.
Faster, for the least
- NVIDIA H200 NVL: $32,300, 777% of the GeForce RTX 3090, 141 GB
- NVIDIA H100 NVL: $32,600, 700% of the GeForce RTX 3090, 94 GB
More VRAM, for not much more
- NVIDIA H200 NVL: $32,300, 777% of the GeForce RTX 3090, 141 GB
From the live listings of every GPU, shipping included; performance against the GeForce RTX 3090 (FP16 or FP32 compute and memory bandwidth). Updated with every build.
Local AI: what it runs
| Model (Q4, 8K context) | H100 PCIe 96 GB needed | Costs now | Speed (est.) |
|---|---|---|---|
| Llama 3.1 8B | 1 card | $29,802 | 206–309 tokens/s |
| Qwen3 14B | 1 card | $29,802 | 111–167 tokens/s |
| gpt-oss 20B | 1 card | $29,802 | 275–412 tokens/s |
| Gemma 3 27B | 1 card | $29,802 | 60–90 tokens/s |
| Qwen3 32B | 1 card | $29,802 | 50–75 tokens/s |
| Llama 3.3 70B | 1 card | $29,802 | 23–35 tokens/s |
| gpt-oss 120B | 1 card | $29,802 | 194–291 tokens/s |
| Qwen3 235B-A22B | 2 cards | not enough listings | 45–67 tokens/s |
| DeepSeek R1 / V3 671B | 5 cards | not enough listings | 27–40 tokens/s |
Cost: that many of the cheapest live listings. Speed: an estimate from the memory bandwidth for one person chatting. Try other models, sizes and contexts.
Decision models
| Model (Q4, 8K context) | H100 PCIe 96 GB needed | Costs now | Per question (est.) |
|---|---|---|---|
| Julia-10.144B | 1 card | $29,802 | 1.6–3.5 ms |
| Laya0.421B | 1 card | $29,802 | 2.7–5.8 ms |
| Kev-4B4B | 1 card | $29,802 | 6.5–14 ms |
| lev4B | 1 card | $29,802 | 19–42 ms |
| OpenJev27B | 1 card | $29,802 | 23–50 ms |
| Clef27B | 1 card | $29,802 | 23–50 ms |
| Perplexity Decider v1.227B | 1 card | $29,802 | 23–50 ms |
| mammoth-127B | 1 card | $29,802 | 23–50 ms |
Decision models score the options you give them instead of writing text, so they're timed per question. The time is scaled from each model's reported speed on an RTX PRO 6000 by this card's compute and memory bandwidth, within about half either way. About the models.
Compare elsewhere
Search other shops for NVIDIA H100 PCIe 96 GB. These open the shop's own search; we don't show their prices here.
