- 1 Top cover + fan wall
- 2 8x GPU tray (NVLink)
- 3 Chassis + power backplane
Server
Built to order
8x H100 SXM server
A full server with 8 H100 GPUs wired together with NVLink, acting as one large pool of memory and compute.
SKU
SRV-H100X8-640G
PriceiChecked against NVIDIA's own product listings and major retailers (B&H, Micro Center, NVIDIA's marketplace) as of July 2026 — real published prices, not estimates.
$250,000
Specifications
GPU memory
640 GB HBM3i"High Bandwidth Memory" — memory stacked directly on the GPU chip itself for very fast, very close access. Used in datacenter GPUs like the H100, not desktop cards.
Power draw
5.6 kW
Tensor / AI compute
~31.7 petaflops FP8 (8-GPU aggregate, sparse)
Rack units
8U typical (air-cooled 8-GPU HGX chassis; varies by system integrator)
Per-GPU memory80 GB HBM3
Per-GPU bandwidth3.35 TB/s
GPU-to-GPU linkNVLink, 900 GB/s
Form factor8x SXM in one server chassis
Which models this can run
Real memory math against the 3 models in the model advisor — parameters × 1 GB, plus 20% working room.
Qwen3.5-397B
476.4 GB required · this unit has 640 GB
Runs on a single unit.
DeepSeek-V3.2
805.2 GB required · this unit has 640 GB
Needs 2x units clustered together (not a single unit).
Kimi K2.6
1,200.0 GB required · this unit has 640 GB
Needs 2x units clustered together (not a single unit).
Power and cooling
Electrical requirement: Draws 5,600 W — 3x what a single standard 120V/20A circuit can safely carry continuously (1,920 W). Requires 208V/240V power distribution and, at this scale, dedicated cooling — standard in datacenters, not offices or homes.
About 4.7x the continuous draw of an average home
(1200 W). Run flat-out for a month, that's roughly
4,032 kWh — about $524/month
at a $0.13/kWh commercial rate (US average, EIA — your
actual rate depends on region and utility contract).
Power cost, relative to the rest of the catalog
$524/mo
Cooling: Air-cooled chassis fans; liquid-cooled variants available from some integrators
What this is good for
- Training and fine-tuning models up to several hundred billion parameters
- High-throughput, multi-user inference
- Workloads that benefit from NVLink GPU-to-GPU memory sharing
What this is NOT good for
- Small teams or single-user workloads — significant overkill on price and power
- Budget-constrained prototyping
- Sites without 3-phase power and datacenter-grade cooling
Included components
Complete server: 8x H100 GPUs, dual host CPUs, redundant power supplies, and NVLink baseboard. Does not include networking switches, cabling, or rack enclosure.
Lead time and warranty
Delivery window estimateBuilt-to-order enterprise hardware — typically an 8–16 week lead time from order confirmation, depending on allocation.
WarrantyManufacturer warranty applies (enterprise service terms vary by reseller).