Ascend 910B vs H20

Ascend 910B vs H20

The two China-market answers head to head: the H20 has more memory (96 GB) and bandwidth (4.0 TB/s) with the CUDA ecosystem; the 910B has roughly double the FP16 compute (320 vs 148 TFLOPS class) with a domestic stack. Compute-bound buyers lean Ascend; ecosystem-bound buyers lean H20.

Workload fit

Ascend 910B: sovereign china ai training and inference, inference at a100-class scale, cuda-native workloads. H20: large-model inference, memory-bound hpc and analytics, llm training.

Which is cheaper to rent, Ascend 910B or H20?

By today's published on-demand rates: Ascend 910B starts at $0.594/GPU/hr (median $0.95, 3 providers) and H20 starts at $1.010/GPU/hr (median $1.29, 3 providers). The cheapest live listing is the Ascend 910B at Gitee AI. Figures recompute from the panel on every refresh.

Ascend 910B or H20 - which should I rent?

Ascend 910B: The Ascend 910B is Huawei's domestic answer to NVIDIA's China-market cards: 64 GB of HBM2e and roughly A100-class FP16 throughput on SMIC 7nm. H20: The H20 is the China-market Hopper: full 96 GB of HBM3 and 4.0 TB/s of memory bandwidth, but tensor compute capped at 148 TFLOPS of FP16 - a fraction of the H100's. The side-by-side spec table and live price rows above give the full picture.

Where can I rent Ascend 910B and H20?

3 providers publish Ascend 910B rates and 3 publish H20 rates on Compute Cafe. Every row links to the provider's own published price - committed-use and reserved rates are excluded.