H200 NVL 141GB
141 GB of HBM3e on a PCIe card. Drops into a 4U chassis, bridges to 4-way NVLink. The practical path to 564 GB per box.
In stock. Lead time 72 hours.
Volume pricing
| Units | Discount | Unit price |
|---|---|---|
| 1 unit | List | $31,000 |
| 2 to 7 units | 3% off | $30,070 |
| 8 to 31 units | 6% off | $29,140 |
| 32+ units | 10% off | $27,900 |
Unit price at 1: $31,000
Request a bulk quote
The full sheet.
Figures from the manufacturer datasheet. Price is an indicative street price, shown as from.
- VRAM
- 141 GB
- Memory
- HBM3e
- Bandwidth
- 4.8 TB/s
- FP8
- 3.34 PFLOPS
- FP4
- Not supported
- Board power
- 600 W
- Form factor
- PCIe dual-slot
- NVLink
- Yes
- Architecture
- Hopper
- GPUs
- 1
- Per GB of VRAM
- $219.86
- Availability
- In stock, ships 72h
- Sale
- Buy online
- SKU
- H200-NVL
What it is for.
Workloads, sizing and what ships in the box.
141 GB of HBM3e in a PCIe slot.
The H200 memory on a dual-slot PCIe card at 600 W. It fits a 4U air-cooled chassis and bridges to 4-way NVLink, so four cards give 564 GB in one NVLink group. 4.8 TB/s per card, 3.34 PFLOPS FP8. In stock, ships in 72 hours. $31,000, about $220 per GB. NVLink bridges are sold separately.
Good for
- Llama 3.1 70B at FP8 (84 GB with headroom) on one card.
- Llama 3.1 405B at FP8 (486 GB) on four bridged cards.
- Qwen2.5 72B at FP16 (173 GB) on two cards.
Fit rule: FP16 weights take 2 bytes per parameter, FP8 1 byte, 4-bit 0.5 bytes, plus 20% headroom for KV cache at short context.
Pairs with
- H100 NVL 94GB, cheaper Hopper PCIe for smaller models.
- RTX PRO 6000 Blackwell Server Edition 96GB, for $/GB-led inference racks.
NVLink bridges sold separately, 2-way and 4-way.
How much VRAM per box.
Card count is capped by slots and by the chassis power budget.
Parts bought alongside.
Same workload, a different trade between VRAM, bandwidth and price.