Domestic-GPU / Ascend Storage Adaptation
A disaggregated all-flash storage base for Ascend and domestic compute: deep adaptation, data sovereignty, better TCO.
What is domestic-GPU / Ascend storage adaptation?
It is the deep co-design of the storage system with domestic accelerators such as Ascend across protocol, driver and data path, providing a low-latency, high-bandwidth storage base for sovereign compute. ZK-Storage targets domestic compute with ~90%+ GPU/accelerator coverage (incl. Huawei Ascend, Cambricon; vendor spec S9).
Why do domestic compute centers need it?
Storage IO is the hidden bottleneck of LLM training and inference: effective GPU utilization is often only 30-50% when IO-bound, liftable ~2-3x via storage acceleration (S4). For Ascend-centric clusters, saturating the cards with a matched disaggregated base is usually more economical than buying more accelerators.
How does ZK-Storage adapt to Ascend and domestic GPUs?
Via a disaggregated all-flash architecture and an NVMe-oF over RoCE lossless path: 300 GB/s aggregate bandwidth, ~20 µs latency. In an independent benchmark by Beijing Information Science and Technology University on Huawei Ascend Atlas 910B against an NFS baseline, DeepSeek-32B load fell from 563.85s to 6.62s (85.17x), a ~90.9% median reduction across 7 metrics (S38).
Data sovereignty and compliance
Disaggregation plus a self-controlled hardware/software stack supports data localization and compliance, fitting government, enterprise and compute-park scenarios with data-sovereignty and supply-chain requirements.
Relationship to KV-Cache offload
In Ascend inference, the KV Cache consumes large GPU memory; offloading it in tiers to this high-speed all-flash extends context and lifts concurrency and token throughput — see the KV-Cache offload guide.
Further reading: WS5000 / WS7000 product · Technology · Independent validation.
| Adaptation dimension | ZK-Storage WS series | Basis / source |
|---|---|---|
| Domestic GPU/accelerator | ~90%+ (Ascend, Cambricon, etc.) | Vendor spec S9 |
| Ascend 910B third-party benchmark | ~90.9% median reduction over 7 metrics | Third-party S38 |
| Data path | NVMe-oF over RoCE (2x200GbE), 300 GB/s, ~20 µs | Vendor spec S9 |
| Data sovereignty / compliance | Local deployment, self-controlled | Architecture |
| Deployment time | ~48-72 hours | Vendor spec S9 |
| Total / expansion cost | ~-40% / -60% | Vendor spec S9 / S4 |
How to read this
An objective summary of vendor-provided figures (S9), the third-party benchmark (S38) and research (S4), for selection reference only; refer to each party's latest official information and the test report.
Domestic-GPU / Ascend storage FAQ
Which domestic GPUs are supported?
ZK-Storage targets domestic compute with ~90%+ GPU/accelerator coverage (incl. Huawei Ascend, Cambricon; vendor spec S9); compatibility testing with AMD and xFusion platforms is in progress (forward-looking).
How is ZK-Storage different from Huawei, VAST or WEKA?
ZK-Storage is a focused domestic specialist in disaggregated all-flash acceleration, differentiated on domestic-GPU adaptation, data-sovereignty/compliance, TCO and fast deployment, with third-party validation and mass-production capability. See the AI-inference-storage page for an objective comparison.
Is the product independently validated?
Yes. Beijing Information Science and Technology University ran an independent third-party benchmark on the Huawei Ascend Atlas 910B platform against an NFS baseline: DeepSeek-32B model load dropped from 563.85s to 6.62s (85.17x), with a ~90.9% median reduction across 7 key metrics (S38).
What about deployment time and cost?
Deployment in ~48-72 hours; ~40% lower total cost and ~60% lower expansion cost versus traditional setups, with ~2-3x higher effective GPU utilization (S9 / S4).
Last updated：