Gigabyte G593-SD1-LAX3
Overview
Performance Metrics
| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 0.3 | 34.0 | — |
| FP32 | 0.5 | 67.0 | — |
| TF32 | 4.0 | 494.5 | — |
| FP16 | 7.9 | 989.5 | — |
| BF16 | 7.9 | 989.5 | — |
| FP8 | 15.8 | 1979.0 | — |
| INT8 | 15.8 | 1979.0 | — |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
Configuration
- Accelerator
- H200 SXM 141GB
- Count
- 8
- Memory per GPU
- 141 GB
- Bandwidth per GPU
- 4.8 TB/s
Power & Efficiency
- Aggregate Power
- —
- TDP per GPU
- 700 W
- FP8 Efficiency
- —
- Aggregate Bandwidth
- 38.4 TB/s
System Details
Interconnect
NVIDIA NVLink with NVSwitch, 900 GB/s GPU-to-GPU
Notes
A 5U direct-liquid-cooled server carrying the NVIDIA HGX H200 baseboard with eight SXM GPUs, connected at 900 GB/s by NVLink and NVSwitch. Both CPUs and GPUs are liquid cooled, which is what lets Gigabyte fit the platform into 5U rather than the 8U an air-cooled equivalent needs. Two LGA 4677 sockets take 5th or 4th generation Intel Xeon Scalable processors or the Xeon CPU Max series, with 32 DIMM slots across eight DDR5 channels, eight Gen5 NVMe bays, twelve PCIe Gen5 x16 slots and BlueField-3 DPU compatibility. Power comes from four plus two redundant 3000 W Titanium supplies. It weighs 91 kg net and orders as 6NG593SD1DR000LBX3. Note that Gigabyte's page quotes "over 32 petaflops of FP8 deep learning compute" for an eight-way HGX H200; that is NVIDIA's own marketing figure for the baseboard rather than a Gigabyte measurement, and it is the sparse number. The dense figures shown here are NVIDIA's for the same baseboard, and they are what this site publishes throughout.