NTS GPU Systems Built in the USA
Elite APEX GPU Servers for AI & HPC
NVIDIA HGX Blackwell/Hopper and AMD Instinct platforms—engineer-to-order with fabric, cooling, and contract-ready BOMs from Fremont, CA.
GPU Systems
APEX platforms sized to workload—not brochure TFLOPS
NTS Elite APEX GPU Systems deliver NVIDIA HGX Blackwell/Hopper and AMD Instinct platforms with the same engineer-to-order discipline as the rest of the catalog—host PCIe topology, HBM, NIC count, and cooling documented on every RFQ.
Choose 8-GPU HGX when NVSwitch training density matters, 2U/4U PCIe for inference and HPC attach, Instinct when ROCm or dual-vendor strategy is required, and DLC when rack kW exceeds air envelopes.
Procurement spans SEWP V, ITES-4H, and SLED vehicles; 48-hour burn-in with GEMM and NCCL/RCCL patterns ships from Fremont with STIG imaging and serial manifests as required.
Every configuration is engineer-to-order: NTS validates power, cooling, firmware, and drivers in Fremont before systems ship.
Federal, defense, and SLED buyers can request CLIN-aligned quotes on SEWP V, ITES-4H, GSA, or cooperative vehicles; commercial teams receive the same staging discipline.
NTS documents rack elevation, PDU draw, thermal headroom, and driver baselines on every RFQ so facility teams are not discovering constraints at install.
Related storage, networking, and software lines can ride the same quote when your vehicle and SOW allow—keeping multi-vendor programs under one accountable integrator.
Buying with NTS
Catalog paths with integrator accountability
Browse the platforms below, then configure or request an architecture review.
Category hubs organize NTS Elite and partner platforms by form factor and workload. Cards open deeper catalogs or configuration lists; related GPU, storage, and networking options stay on one accountable BOM.
Use Configure on product pages for CPU, memory, accelerators, and drives. For multi-rack or liquid-cooled designs, contact engineering early so facility constraints inform the quote.
After you shortlist a platform, use product Configure pages for BOM options or request an architecture review for multi-rack, liquid-cooled, or dual-vendor GPU comparisons.
Product categories
Explore our server platforms
Rackmount, GPU, blade, storage, cluster, high-density, and Ampere — engineered for scalability and efficiency.
- B200 / B300 / H100 / H200
- NDR or RoCE east-west
- 48h burn-in
- Quad / dual GPU SKUs
- L40S & H100 NVL
- Edge-ready 2U
- ROCm stack guidance
- RCCL fabric maps
- Export diversification
- Manifold & QD design
- Thermal soak tests
- Link to liquid hubs
NTS catalog
Available GPU Systems configurations
Select a platform below to configure, request a quote, or add to your procurement package.
- Configure & quote
- CPU / memory / storage specs
- Contract-ready procurement
Configure & quote
CPU / memory / storage specs
Contract-ready procurementDisplay per page
NVIDIA GB300 NVL72
4xAMD Instinct™ MI300A APUs
128GB HBM3 per APU
8 x 2.5" Gen5 NVMe bays
3+1 3000W 80 PLUS Titanium redundant power
NTS Elite Studio Dual‑Socket 4U Tower Intel Xeon Server
2S Intel Emerald Rapids / Sapphire Rapids CPU
16 DIMM slots w/ DDR5 4800 memory support
2 FH/10.5"L double-wide PCIe 5.0 x16 + 1 FH/10.5"L single-wide PCIe 5.0 x16 slots
8 3.5" SATA + 2 2.5" SATA hot swap , tool less drive bays
1 NVMe M.2 2280 slot
Single ATX12V PSU
NTS Elite Studio 5U Liquid-Cooled GPU Workstation
5th/4th Gen Intel® Xeon® Scalable processor
6 PCIe 5.0 x16
8x 2.5" peripheral drive bays
Closed-loop liquid cooling for CPUs and GPUs
Supports up to four liquid-cooled GPU
NTS Elite Studio 5U Liquid-Cooled GPU
Dual 4th Gen Intel® Xeon® Scalable Processors, 6444Y (16 cores at 3.6Ghz) installed
4 liquid-cooled NVIDIA A100 GPUs with NVIDIA® NVLink™ (2+2) installed
Dual-port active-cooled NVIDIA Connect-X6 Dx 25G Network Adapter installed
Supermicro Cloud Orchestrator AI software stack
Ubuntu 22.04 LTS installed
NTS Elite Studio 4U Tower Dual‑Socket Intel Xeon Server with 4 GPUs
Dual LGA4677 socket
(8+8) Support up to 2,048GB RDIMM/3DS RDIMM DDR5 5600/4800 memory
(8) 3.5" Hot-swap SATA + (2) 2.5" screwless SATA in (2) swappable carriers at front
(2) FH/FL PCIe Gen.5 x8 + (3) FH/FL PCIe Gen.5 x16 slots
NTS Elite Studio 4U High-Density GPU Server
(1) AMD EPYC™ 9004 Series Processor
(8) DDR5 DIMM slots, Support up to DDR5 RDIMM/3DS RDIMM 4800
(8) 3.5" Hot-swap SATA + (2) 2.5" hot-swap U.2 NVMe drive bays
2 NVMe M.2 22110 / 2280 slots
2 10GbE + 2 GbE + 1 IPMI ports
Single ATX12V PSU
NTS Elite Studio 4U GPU-Optimized Server
1S AMD Turin / Genoa / Bergamo / Genoa X CPU
8 DIMM slots w/ DDR5 6000 memory support
8 3.5" SATA + 2 NVMe U.2 hot swap, tool less drive bays
2 NVMe M.2 22110 / 2280 slots
2 GbE + 1 IPMI ports
Single ATX12V PSU
NTS Elite Studio 4U GPU Workstation & AI Compute Tower
Intel 5th/4th Gen Xeon Scalable
2 TB DDR5 ECC RDIMM
8 3.5" SATA/SAS/NVMe Hot-Swap
7 PCIe 5.0 x16
GPU-Optimized
NTS Elite Studio 4U AI Compute Server
1S Intel Emerald Rapids / Sapphire Rapids CPU
8 DIMM slots with DDR5-4800 memory support
6 3.5" SATA + 2 NVMe U.2 hot-swap, tool-less drive bays
1 NVMe M.2 2280 slot (PCIe 3.0 x4)
1 1000Base-T LAN + 1 dedicate 1000Base-T IPMI ports
Single ATX12V PSU
NTS Elite Studio 4U Accelerated Compute Server
1S AMD Turin / Genoa / Bergamo / Genoa X CPU
8 DIMM slots w/ DDR5 6000 memory support
8 3.5" SATA + 2 NVMe U.2 hot swap, tool less drive bays
2 NVMe M.2 22110 / 2280 slots
2 GbE + 1 IPMI ports
Single ATX12V PSU
NTS Elite APX 8U 8‑GPU AI Server
- Intel 3rd Gen Xeon Scalable
- 2 TB DDR4 ECC RDIMM
- 4 2.5" NVMe Hot-Swap
- GPU-Optimized
- 2 PCIe 4.0 x16
NTS Elite APEX with HGX B200 for Extreme AI Performance
Dual AMD EPYC™ 9005/9004 Series Processors
Liquid-cooled NVIDIA HGX B200
12-Channel DDR5 RDIMM, 24 x DIMMs
8 x 2.5" Gen5 NVMe hot-swap bays
900GB/s GPU-to-GPU bandwidth with NVIDIA NVLink
4+4 3000W 80 PLUS Titanium redundant power supplies
Specifications shown are reference configurations—final BOM validated by NTS architects for your workload, facility power, and security baseline.
Frequently asked questions
A GPU server pairs host CPUs with high-bandwidth accelerators for parallel AI/HPC. CPU clusters remain right for transactional and lightly threaded services.
- Size PCIe/NIC/memory to feed GPUs
- APEX for dense HGX
Follow software stack, export posture, and second-source needs - not headline TFLOPS. NTS delivers side-by-side BOMs mapped to your frameworks.
- CUDA vs ROCm
- Facility kW called out
Air suits many PCIe nodes; DLC is recommended for sustained HGX training and high rack density.
- Clock stability
- CDU planning
SEWP V, ITES-4H, GSA MAS, and SLED cooperatives such as OMNIA, NASPO, TIPS, and Texas DIR when in scope.
- Catalog quotes in days when silicon available
- Custom clusters use milestone schedules
