Atlas 300i Duo 96GB — Single Card
Dual Ascend 310B NPU · 96GB LPDDR4X ECC · 280 TOPS INT8 · 150W TDP. Run DeepSeek V4-Flash, Qwen 72B and Llama 3 70B fully locally, privately, without rate limits or subscriptions.
We are unable to ship to your location.
Technical Specifications
| Total memory | 96GB LPDDR4X ECC (2× 48GB) |
| Architecture | Dual Ascend 310B NPU |
| Compute | 280 TOPS INT8 · 140 TFLOPS FP16 |
| Interface | PCIe Gen4 x16 |
| Power | 150W TDP |
| Virtualisation | Up to 7 virtual NPUs per processor |
| Models supported | 70B–100B+ parameter at reasonable quantisation |
| Condition | Brand new and unused |
The Huawei Atlas 300I Duo 96GB is a dual-chip AI inference accelerator featuring 96GB of LPDDR4X VRAM (2x48GB) and 408 GB/s memory bandwidth, designed primarily for running large language models (LLMs) and video analytics. It offers a significantly lower cost-per-GB compared to competitors like the NVIDIA RTX 6000 Pro.
Key Specifications
Architecture: Equipped with 2x Ascend 310B processors (16 Da Vinci AI cores each), delivering up to 280 TOPS (INT8) and ~140 TFLOPS (FP16).
Efficiency: It operates with a low 150W TDP and uses passive cooling, making it suitable for energy-constrained edge or data center deployments.
Compatibility: The card uses a PCIe Gen4.0 x16 interface and supports virtualization, splitting each processor into up to 7 virtual NPUs.
Software ecosystem
The Atlas 300i Duo runs on Huawei’s CANN/Ascend stack—now a genuinely mature platform, backed by serious commitments from Alibaba and ByteDance to the upcoming Atlas 350. Two strong, actively maintained paths keep things moving today:
-
CANN backend (llama.cpp) — widely validated for GGUF-format LLM inference
-
PyTorch via torch-npu — a complete Python ML stack covering both training and fine-tuning
Huawei Official Documentation
Atlas 300i Duo 96GB — Single Card — product demonstration
Compatible Platforms
- ✓ Taishan 200 2280 V2 (ARM) — recommended
- ✓ Atlas 800 3000 — supported
- ✓ Huawei RH2288H V5 (x86) — requires driver recompile