Huawei Atlas 950 SuperPoD
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.
We will point you at suppliers who have it. Free, and no signup.
System Details
GPU Configuration
System Specifications
Precision Performance Breakdown
| Precision | System Performance | Per GPU | Efficiency |
|---|---|---|---|
| 1000.000 PFLOPS | 976.6 TFLOPS | — | |
| 2000.000 PFLOPS | 1953.1 TFLOPS | — |
All figures are dense. Vendors commonly headline the number, which is twice the dense one.
Powered by Huawei Ascend 950DT
This system utilizes 1024 × Huawei Ascend 950DT GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
—
Per GPU Memory
144 GB
Process Node
—
Architecture
Da Vinci v3
Documentation & Resources
Official Datasheet
Huawei Atlas 950 SuperPoD technical specifications
Typical Use Cases
The Huawei Atlas 950 SuperPoD runs 1024× Ascend 950DT GPUs over UnifiedBus 2.0 (Lingqu) all-optical, 1.72 PB/s aggregate scale-up, 64 x 1.68 TB/s bidirectional per cabinet, delivering 1,000 PFLOPS FP8 dense.