Huawei Atlas 800T A3
The training member of Huawei's Atlas 800 A3 family: a 10U air-cooled compute node with eight Ascend 910 NPUs and four Kunpeng 920 CPUs, rated at 6.0 PFLOPS of FP16 with 1,024 GB of on-chip memory at 3.2 TB/s per NPU and 784 GB/s of bidirectional device-to-device bandwidth. Huawei sells three models on this chassis at 6.0, 5.0 and 4.48 PFLOPS FP16, which works out at 750, 625 and 560 TFLOPS per NPU; this is the top one, and the 4.48 model is the separately listed Atlas 800I A3. Multiple nodes combine into an Atlas 900 A3 SuperPoD of up to 384 cards. Huawei publishes only an FP16 figure for this machine, and no INT8: a previously held INT8 value of 12.0 PFLOPS was 8 x 1,500 TFLOPS, a per-NPU rate Huawei states nowhere, and has been removed. Power is also not attributable: Huawei's family page gives 16.2 kW as the maximum input power for the Atlas 800 A3 line, but that line covers three compute variants and the Atlas 800I A3's own page separately says 14.6 kW, so no figure is recorded here rather than guessing which model the 16.2 kW describes. Huawei labels nothing dense or sparse.
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 |
|---|---|---|---|
| 6,000 TFLOPS | 750 TFLOPS | — |
Where a vendor states a basis, the figure shown is the dense one, and any figure it publishes is listed separately. Vendors commonly headline the sparse number instead: for NVIDIA's 2:4 structured sparsity that is exactly twice the dense figure, though other vendors' sparse modes do not all follow that ratio.
Powered by Huawei Ascend 910C
This system utilizes 8 × Huawei Ascend 910C GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
600W
Per GPU Memory
128 GB
Process Node
SMIC 7nm
Architecture
Da Vinci v2 (dual-die)
Documentation & Resources
Flopper Spec Sheet
Huawei Atlas 800T A3 specifications, generated from our database. Printable.
Vendor product page
Huawei Atlas 800T A3 documentation
Huawei HiSilicon Da Vinci architecture, Hot Chips 31 (2019)
Huawei • 2019-08-19
Typical Use Cases
The Huawei Atlas 800T A3 runs 8× Ascend 910C GPUs over UnifiedBus (UB): 784 GB/s bidirectional D2D per NPU; scales to a 384-NPU Atlas 900 A3 SuperPoD, delivering 6,000 TFLOPS FP16 dense.