# FAQ: FX Storage Acceleration, Benchmarks, Pricing, Engagement

> Common questions about the Xinyuanyi FX series: how much KV cache tiering helps, how the data is verified, what FX100 costs, which accelerators are supported and how to start.

URL: https://vuke.top/en/faq · Updated November 12, 2026

Figures in the answers carry report IDs; follow an ID to see what that report tested.

## What does Xinyuanyi do?

Xinyuanyi (Tianjin) Semiconductor Equipment Co., Ltd. builds storage acceleration for LLM inference. The product is the FX series of all-flash NVMe-oF arrays, which holds the KV cache that overflows GPU memory so that evicted sessions are read back instead of recomputed. Around it we offer enablement for domestic and non-NVIDIA accelerators, datacenter efficiency optimization, datacenter construction and inference-stack software development.

## What is the FX series storage acceleration platform?

The FX series is an integrated hardware and software storage acceleration system developed on Guangdong Luisuan GP-series hardware. It ships with a vLLM + LMCache KV cache tiering stack that includes Xinyuanyi's own parallel-read patch. FX100, FX200 and FX300 are in production; FX400 is planned for volume production in late 2026. FX100 offers 100 Gb per port and 16M IOPS; FX300 is a PCIe 5.0 platform with 400 Gb per port and 60M IOPS (vendor specifications).

## How much does KV cache tiering improve inference performance?

With Qwen3-Coder-480B-FP8 on 8 × AMD MI308X in TP=8 production topology, the measured results under a long-context cold-recovery load are: 29–40% more inference throughput and 26–32% lower TTFT p50 than a single local NVMe drive [R2](https://vuke.top/en/evidence/R2) [R3](https://vuke.top/en/evidence/R3), and 8.6–20× faster than full recompute without external storage [R2](https://vuke.top/en/evidence/R2). These are application-level throughput and time to first token, not microsecond latencies of the storage device.

## Where does it help most, and where is the benefit limited?

The gain comes from cache hits. Long contexts, multi-turn conversations and high-concurrency session recovery (agents, coding assistants, customer service) hit the cache often and benefit the most. Short, low-concurrency conversations hit it rarely and benefit little. That is why we recommend testing before buying: measure with your own session-length distribution and concurrency, then decide.

## How can these numbers be verified? Are they reproducible?

Every key figure carries a report ID (R1–R9), and the reports are available on request. R4 is an official report issued by a third-party testing organization. R8 is the test code export and raw data: the LMCache parallel-read patch, the load client, orchestration scripts and forensic probe scripts, so a third party can rerun the tests on its own cluster. Two independent runs at the same load level deviate by no more than 5% [R3](https://vuke.top/en/evidence/R3).

## Does adopting FX require changes to the inference engine?

No rewrite of the inference engine is needed. The integration point is the open-source vLLM + LMCache combination: LMCache moves KV cache that does not fit in GPU memory down to host RAM and external storage, and the FX array is that external tier. We adapted the stack at source level on ROCm, and the parallel-read patch is delivered as part of R8 [R1](https://vuke.top/en/evidence/R1) [R8](https://vuke.top/en/evidence/R8).

## How much does a fully populated FX100 cost?

The reference price of a fully populated FX100 is about ¥371,200: ¥160,000 for the platform plus 24 enterprise NVMe SSDs of 7.68 TB each (reference price ¥8,800 per drive), 184.3 TB raw, or about ¥2,014 per TB. SSD prices follow the market, and the purchase contract sets the final price. FX200 and FX300 are quoted by configuration; production pricing for FX400 has not been announced.

## We do not want to add GPUs to our existing cluster. Does storage acceleration still help?

Yes. That is the idea behind improving efficiency before adding capacity. Measured against NFS on Huawei Ascend 910B, DeepSeek-32B serving load time falls from 691 s to 112 s (6.2×) and DeepSeek-70B from 1399 s to 150 s (9.3×) [R9](https://vuke.top/en/evidence/R9). On MI308X, training checkpoint saves are 1.9× faster, and at 20 model switches per hour effective compute utilization rises from 46.7% to 62.8% [R1](https://vuke.top/en/evidence/R1).

## Which domestic or non-NVIDIA accelerators are supported?

Platforms with measured reports are AMD Instinct MI308X (ROCm 7.2, R1–R4), Huawei Ascend 910B [R9](https://vuke.top/en/evidence/R9) and MetaX N260 [R5](https://vuke.top/en/evidence/R5). Each platform has its own report and its own labels; the numbers are never mixed. For multimodal work, all 7 ComfyUI + LTX-Video 2.3 models run end to end on MI308X [R6](https://vuke.top/en/evidence/R6) [R7](https://vuke.top/en/evidence/R7).

## How many inference nodes can one FX100 serve?

The production sizing basis is one FX100 per 8 inference nodes, anchored on the measurement that 100 GbE sustains KV read-back for 8 nodes. At that ratio, storage spend stays within roughly 10% of the GPU budget on the joint-test basis. Your actual ratio depends on session length and concurrency; gate G3 of the joint test verifies the case of 8 nodes sharing one array.

## How do we start working together?

With a gated joint test of about 10 weeks and four gates: G1 delivery acceptance, G2 single-node baseline, G3 main joint-test gate (TTFT reduction of ≥25% and a throughput gain inside the measured +29–40% band [R2](https://vuke.top/en/evidence/R2) [R3](https://vuke.top/en/evidence/R3)) and G4 72-hour stability sign-off. Missing any gate stops the project; passing all of them opens the discussion about scale. The full costing model is provided as Python source after an NDA. Call or email us via the contact page to book a slot.

## How do I get the original test reports?

Email or call us via the contact page and tell us which report IDs you need. Each report's page on this site already publishes its scope, platform and key findings, so you can judge which ones match your scenario first. R8 (code export and raw data) and R9 (the Ascend 910B report) are provided on request in the same way.

## How do AISSD5000, WS5000 and FX100 relate?

They are names for the same product at different times. Earlier test-report filenames say AISSD5000, and before that it was called WS5000 / GP5000; some reports are signed “Mingxin”. This site uses the FX naming throughout (FX100 / FX200 / FX300 / FX400), and report entries keep their original titles for cross-checking.