Server
8 × H100 server: 482.6 × 355.6 mm. Model-specific illustration, drawn to scale.
Sold by Exxact
8 × H100 server
8 GPUs · 640 GBPrice checked 25 Oct 2026
Exxact's TensorEX 8U server with an HGX H100 board: eight H100 SXM GPUs with 80 GB each, two 48-core EPYC CPUs and 1.5 TB of memory.
$319,924
Key features
- Eight H100 SXM with 80 GB each: 640 GB of HBM3
- 7.9 PFLOPS of dense FP16, 989.4 TFLOPS per GPU by NVIDIA's whitepaper
- Room for a ~256-billion-parameter model at 16-bit, ~1 trillion at 4-bit
- Up to 700 W per GPU, 5.6 kW for the GPUs alone
- Two 48-core EPYC 9454, 1.5 TB of DDR5, 8U, three-year parts and labour
Parts and prices
| Part | Qty | Price |
|---|---|---|
| Exxact TensorEX TS4-193475697, 8 × H100 SXM TS4-193475697 | 1 | $319,924 Exxact, 25 Oct 2026 |
Specifications
Exxact TensorEX TS4-193475697, 8 × H100 SXM
- GPUs
- 8 × H100 SXM (80 GB each)
- CPU
- 2 × AMD EPYC 9454, 48 cores each
- System memory
- 1.5 TB DDR5-4800 (24 × 64 GB)
- Storage
- 960 GB U.2 NVMe
- Networking
- 2 × 10 GbE
- Note
- Each GPU: 80 GB HBM3, up to 700 W (NVIDIA); six 3,000 W power supplies (4 + 2 redundant); 3-year parts and labour
Sources: Exxact configurator ↗ · NVIDIA H100 specifications ↗ · NVIDIA H100 architecture (FP16) ↗
Power adds each part's maximum rated power, before cooling and power losses. Model sizes are memory arithmetic: 2 bytes per parameter at 16-bit, half a byte at 4-bit, keeping a fifth of the memory for the context cache. They say what fits, not how fast it runs.
Can't afford it? Rent it for real.
Cloud providers rent the NVIDIA H100 from $1.99 per GPU an hour, or $0.95 on spot, compared on GlobeGPU (rates of 25 Oct 2026).
Compare NVIDIA H100 prices on GlobeGPU ▸