NVIDIA DGX Systems
The Gold Standard for AI Infrastructure
Purpose-built AI supercomputers from NVIDIA. Up to 72 PetaFLOPS of AI performance, 2.3TB of HBM3e memory, and 800Gbps InfiniBand networking. Deployed by our CMMC-RP certified team.
NVIDIA DGX Lineup
From a desktop supercomputer to a full rack server, there is a DGX for every scale of AI ambition.
NVIDIA DGX Station GB300
$94,231
A complete AI supercomputer in a desktop form factor. 20 PetaFLOPS of AI performance on your desk -- no data center required. Ideal for individual researchers, AI teams, and compliance-sensitive environments where data must stay on premises.
AI Performance
20 PetaFLOPS
GPU
GB300 Grace Blackwell Ultra Superchip
CPU
NVIDIA Grace CPU
Memory
748 GB Coherent
Networking
ConnectX-7
Form Factor
Desktop Supercomputer -- standard office power, quiet cooling
NVIDIA DGX B300
Contact for Quote
72 PetaFLOPS AI Performance
NVIDIA DGX B200
Contact for Quote
72 PetaFLOPS AI Performance
NVIDIA DGX H200
Contact for Quote
32 PetaFLOPS AI Performance
DGX System Comparison
All four DGX systems at a glance.
| Specification | DGX Station GB300 | DGX B300 | DGX B200 | DGX H200 |
|---|---|---|---|---|
| AI Performance | 20 PFLOPS | 72 PFLOPS | 72 PFLOPS | 32 PFLOPS |
| GPUs | 1x GB300 Superchip | 8x Blackwell Ultra | 8x Blackwell | 8x H200 |
| GPU Memory | 748GB Coherent | 2.3TB HBM3e | 1.4TB HBM3e | 1,128GB HBM3e |
| Memory Bandwidth | -- | 64 TB/s | 64 TB/s | 4.8 TB/s |
| CPU | NVIDIA Grace | 2x Xeon 6776P (128c) | 2x Xeon 8570 (112c) | 2x Xeon 8480C (112c) |
| Networking | ConnectX-7 | 800Gbps IB | 800Gbps IB | 400Gbps IB |
| Form Factor | Desktop | Rack Server | Rack Server | Rack Server |
| Starting Price | $94,231 | Contact Us | Contact Us | Contact Us |
Why Choose NVIDIA DGX
DGX is purpose-built by NVIDIA for AI -- not a generic server with GPUs bolted on.
NVLink Interconnect
900GB/s GPU-to-GPU bandwidth. Impossible to replicate with PCIe. Enables training models that span multiple GPUs as if they were one.
NVIDIA Validated
Every component tested and validated by NVIDIA engineering. No compatibility issues, no performance surprises, no guesswork.
AI Enterprise Software
Includes NVIDIA AI Enterprise, Base Command, NGC Catalog, NIM microservices, and enterprise support directly from NVIDIA.
Linear Scalability
Scale from 1 DGX to a SuperPOD cluster via InfiniBand. Performance scales linearly. PTG designs and deploys multi-node clusters.
Data Sovereignty
All processing stays on your premises. No data leaves your facility. Essential for HIPAA, CMMC, defense, and regulated industries.
Predictable TCO
One-time purchase, unlimited usage. No per-token fees, no egress charges, no surprise bills. DGX pays for itself within months for heavy AI workloads.
PTG DGX Deployment Services
We handle every aspect of your DGX deployment -- from site assessment to production.
Site Assessment
Power capacity analysis, cooling evaluation, rack space planning, and network readiness assessment before hardware arrives.
Physical Deployment
Rack installation, power connections, InfiniBand cabling, and environmental monitoring setup. Same-day for Raleigh-Durham clients.
Software Configuration
NVIDIA Base Command, AI Enterprise, PyTorch, TensorFlow, CUDA, and your custom AI frameworks pre-configured and tested.
Compliance Hardening
HIPAA/CMMC configuration by our 4-member CMMC-RP certified team. Encryption, audit logging, access controls, and documentation.
Frequently Asked Questions
Both deliver 72 PFLOPS, but the B300 uses Blackwell Ultra GPUs with 2.3TB HBM3e (vs 1.4TB on B200) and newer Intel Xeon 6776P processors with 128 cores. The extra memory on B300 allows larger models without multi-node parallelism.
A desktop AI supercomputer delivering 20 PFLOPS with the GB300 Grace Blackwell Ultra Superchip and 748GB coherent memory. It sits on your desk, runs on standard power, and uses quiet cooling -- no data center required. Starting at $94,231.
DGX delivers NVLink interconnect with 900GB/s GPU-to-GPU bandwidth -- impossible with PCIe. It is also NVIDIA-validated, includes AI Enterprise software and NVIDIA support, and scales to SuperPOD clusters. Custom servers cannot match this integration level.
DGX Station GB300 starts at $94,231. DGX B300, B200, and H200 are priced by configuration -- call (919) 348-4912 for current quotes. PTG also offers financing to make DGX accessible. For a more compact option, see the DGX Spark.
Yes. Our four-member CMMC-RP certified team specializes in compliant AI deployments. DGX keeps all data on premises. We configure encryption, access controls, audit logging, and security hardening for HIPAA, CMMC, NIST 800-171, and other frameworks.
Absolutely. DGX systems scale linearly via InfiniBand to form SuperPOD clusters. PTG designs and deploys multi-node clusters including the InfiniBand fabric, NVIDIA Base Command cluster manager, and distributed training software.
DGX is a complete, NVIDIA-validated system with integrated software, support, and cluster management. HGX uses the same GPU baseboard but is integrated by third-party server OEMs, offering more customization options at a different price point. See our HGX page for details.
Deploy NVIDIA DGX with PTG
From a DGX Station on your desk to a multi-node DGX SuperPOD in your data center. Our CMMC-RP certified team handles site assessment, installation, configuration, and ongoing support.
Call now for a free consultation and current pricing. Financing available.
Or schedule a call at a time that works for you
Petronella Technology Group | 5540 Centerview Dr, Suite 200, Raleigh, NC 27606 | Since 2002