End AI Storage Bottlenecks with NFS over RDMA | QNAP

If this page does not render correctly, click herefor the online version

End AI Storage Bottlenecks

with NFS over RDMA

Any I/O latency (in milliseconds) wastes the computational power of expensive GPUs. The combination of the TS-h1290FX and NFS over RDMA ensures that storage performance keeps pace with computational speed.

[1] Learn NFS over RDMA

Your GPU might be wasting time waiting for data

While the cost of AI training is determined by GPU runtime, storage I/O bottlenecks could be wasting more than 40% of your computation time.

1. The Hidden Cost of the TCP Stack

Every time data is read, the CPU is occupied with processing TCP packets and context switching. Although this contributes nothing to AI computations, it secretly consumes up to 99% of CPU resources.

2. 4–6 rounds of wasted memory copies

With traditional NFS, data is copied between the kernel and user space 4–6 times before reaching the GPU, and every 1 microsecond of added latency results in lost computational power.

3. Real-world losses from GPU idle time

For example, in an 8×H100 cluster, cloud costs exceed US$24 per hour. If GPU utilization drops to 60% while waiting for data, approximately US$10 per hour is completely wasted.
Why NFS over RDMA Changes Everything

By leveraging Kernel Bypass and Zero-Copy technologies, we eliminate the redundant memory copies and CPU overhead inherent in traditional NFS. All protocol processing is offloaded to the hardware, drastically minimizing the time your GPUs spend waiting for data.

100x Latency Reduction

Real-time response for AI training.
85% CPU Load Reduction

Free up your servers for actual computing.

Near Line-Rate Throughput

Achieve up to 100Gbps speed across the network.

<5% GPU Wait Time

Keep your expensive GPUs working at full capacity.

Stop Settling for Legacy Storage Bottlenecks

Feature
QNAP TS-h1290FX
Standard SATA NAS

Performance
NFS over RDMA (Native)
Unsupported (Legacy TCP)

Latency
Ultra-Low (Microseconds)
High (Milliseconds)

CPU Usage
Minimized via RDMA
Heavy Overhead

Drive Interface
U.2 NVMe PCIe Gen 4
SATA 6Gb/s

Ready to Accelerate Your Enterprise AI?

Upgrade to the TS-h1290FX All-Flash Array and unleash the full potential of your compute nodes.

[2] Get Technical Specs Now
Questions?

For any marketing and public
relation queries

[3] Contact us

Follow us!

Get all the latest news

[4]  
[5]  
[6]  
[7]

Where to buy

Find your nearest shop

[8] Search

Copyright © 2026 QNAP Systems, Inc. All rights reserved.

References:

1. https://qnap.benchurl.com/c/l?u=13E016DD&e=1B5E6C5&c=47C09&t=0&seq=1
2. https://qnap.benchurl.com/c/l?u=13E016DE&e=1B5E6C5&c=47C09&t=0&seq=1
3. mailto:marketing@qnap.com
4. https://qnap.benchurl.com/c/l?u=13E016DF&e=1B5E6C5&c=47C09&t=0&seq=1
5. https://qnap.benchurl.com/c/l?u=13E016E0&e=1B5E6C5&c=47C09&t=0&seq=1
6. https://qnap.benchurl.com/c/l?u=13E016E1&e=1B5E6C5&c=47C09&t=0&seq=1
7. https://qnap.benchurl.com/c/l?u=13E016E2&e=1B5E6C5&c=47C09&t=0&seq=1
8. https://qnap.benchurl.com/c/l?u=13E016E3&e=1B5E6C5&c=47C09&t=0&seq=1