Key application
Scale-Out All-Flash NVMe Storage
High-bandwidth NVMe building blocks for AI training sets, HPC scratch, and analytics - sized in GB/s and IOPS, not just terabytes.
Key application
Scale-Out All-Flash NVMe Storage
Scale-out NVMe is where AI and HPC data pipelines live. When hundreds of GPUs or thousands of cores read the same dataset, capacity is the easy part - sustained bandwidth, metadata performance, and predictable latency under mixed load are what determine whether expensive compute sits idle.
NTS builds all-flash storage nodes as balanced systems: drive count and PCIe lane budget matched to CPU and memory, network ports sized to the drives behind them, and the parallel filesystem or software-defined stack you standardize on. Nodes are burned in, firmware-leveled, and performance-verified in Fremont, CA before they ship.
Use this application hub to map workloads to NTS platforms—GPU servers, storage tiers, and rack integration—with Fremont staging and contract-ready quoting when required.
Each card links to deeper catalog or solution paths. Request an architecture review if you need CLIN structure, lead times, or a dual-vendor GPU comparison.
Workload path
From application intent to validated BOM
Application hubs connect mission use cases to engineer-to-order systems.
Start with the cards below to explore related platforms and capabilities. NTS sizes CPU, accelerator, memory, storage, and fabric together so power, cooling, and drivers match the workload—not orphan SKUs.
Federal and SLED buyers can carry the same architecture onto approved vehicles. Commercial teams get the same staging discipline without the GWAC paperwork when it is not required.
Workloads on this page
Storage applications and capability tiers
Catalog codes that map to this page, plus the adjacent storage tiers NTS builds alongside them.
- U.2, U.3, and E3.S options
- 200/400GbE or InfiniBand
- Linear scale-out building blocks
- GPU-direct storage ready
- Checkpoint write bursts
- Small-file metadata tuning
- Separate metadata and data tiers
- Validated fabric topology
- Filesystem staged before ship
- 60 to 108 bay enclosures
- Hybrid flash and disk nodes
- Tiering to archive targets
Frequently asked questions
By both, and bandwidth usually sets the node count. Tell NTS the aggregate GB/s your compute needs, the read and write mix, and the average file size, and we calculate how many nodes and network ports it takes to sustain that - then confirm the capacity target fits inside it.
- Aggregate GB/s drives node and port count
- Small-file workloads need metadata headroom
- Capacity target validated against bandwidth design
Because an unbalanced node never reaches its rated throughput. Drives, PCIe lanes, CPU cores, and NICs all have to line up. NTS budgets lanes per drive group, matches NIC bandwidth to the flash behind it, and verifies real throughput during burn-in rather than quoting datasheet numbers.
- Lane budget per drive group
- NIC bandwidth matched to drive bandwidth
- Measured throughput at burn-in
NTS builds hardware for parallel and software-defined storage rather than locking you into one stack. Lustre, BeeGFS, GPFS, Ceph, and vendor software-defined platforms are all common, and we can stage and validate the filesystem before delivery.
- Parallel filesystems: Lustre, BeeGFS, GPFS
- Software-defined: Ceph and vendor stacks
- Filesystem staged and validated pre-ship
Scale-out designs grow by adding validated building blocks. NTS documents the node recipe, fabric topology, and firmware baseline so the next expansion matches the original cluster instead of becoming a one-off, and reorders stay consistent across contract years.
- Documented node recipe and firmware baseline
- Add building blocks, not forklift upgrades
- Repeatable reorders on contract vehicles
