Huawei Atlas 950 SuperPoD
Overview
Performance Metrics
| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP8 | 1000.0 | 976.6 | — |
| FP4 | 2000.0 | 1953.1 | — |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
Configuration
- Accelerator
- Ascend 950DT
- Count
- 1024
- Memory per GPU
- 144 GB
- Bandwidth per GPU
- 4.0 TB/s
Power & Efficiency
- Aggregate Power
- 100.0 kW
- TDP per GPU
- —
- FP8 Efficiency
- 10.00 TFLOPS/W
- Aggregate Bandwidth
- 4096.0 TB/s
System Details
Interconnect
UnifiedBus 2.0 (Lingqu) all-optical, 1.72 PB/s aggregate scale-up, 64 x 1.68 TB/s bidirectional per cabinet
Notes
Huawei's current product specification: 16 compute cabinets plus 4 UnifiedBus interconnect cabinets (44OU each), up to 1,024 Ascend 950DT, 1,024 x 96 GB on-chip memory at 4.0 TB/s, 1 EFLOPS mxFP8/FP8/HiF8 and 2 EFLOPS mxFP4, 1.72 PB/s total scale-up bandwidth, 256 TB globally addressable memory, 100 kW, fully liquid cooled. Offered in 64-card and 1,024-card configurations. Huawei's HC2025 keynote of 18 September 2025 described an 8,192-card configuration delivering 8 EFLOPS FP8 and 16 EFLOPS FP4; that figure does not appear in the current product table and is recorded here for reference. Huawei states no dense or sparse basis for any of these figures.