Lorem ipsum, dolor sit amet consectetur adipisicing elit. Voluptatum eum tenetur, porro optio magnam esse maxime accusamus sint similique nulla.

DGX Station GB300

The third-generation DGX Station is built around the Grace-Blackwell Ultra architecture, providing ground-breaking development performance in a desktop format system.

• NVIDIA GB300 Superchip with Grace CPU and Blackwell Ultra GPU
• 72 Neoverse V2 CPU cores / TBC GPU cores
• 288GB ECC Reg DDR5 system memory
• TBC NVMe SSDs system drives
• TBC NVMe SSDs storage drives

DGX H200

The fourth-generation DGX AI appliance is built around the Hopper architecture, providing unprecedented performance in a single system and unlimited scalability with the DGX BasePOD and SuperPOD enterprise-scale infrastructures.

• 8x NVIDIA H200 Tensor Core GPU
• 16,896 CUDA cores & 528 TF32 Tensor Cores per GPU
• 80GB memory per GPU - 640GB total
• 2x Intel Xeon Platinum 8480C – 112 total cores
• 2TB ECC Reg DDR5 system memory
• 2x 1.92TB NVMe SSDs system drives
• 8x 3.84TB NVMe SSDs storage drives

NVIDIA DGX H100

The fourth-generation DGX AI appliance is built around the new Hopper architecture, providing unprecedented performance in a single system and unlimited scalability with the DGX POD and SuperPOD enterprise-scale infrastructures.

• 8x NVIDIA H100 Tensor Core GPU
• 6,896 CUDA cores & 528 TF32 Tensor Cores per GPU
• 80GB memory per GPU - 640GB total
• 2x x86 CPUs - TBC
• 2TB ECC Reg DDR4 system memory
• 2x 1.92TB NVMe SSDs system drives
• 8x 3.84TB NVMe SSDs storage drives

NVIDIA DGX A100

The third-generation DGX A100 appliance introduces double-precision Tensor Cores, providing the biggest milestone since the introduction of double-precision computing in GPUs for HPC. This model was superseded by the DGX H100 version in 2022.

• 8x NVIDIA A100 Tensor Core GPUs
• 6,912 CUDA cores & 432 TF32 Tensor Cores per GPU
• 80GB memory per GPU - 640GB total
• 2x AMD EPYC 7742 - total 128 cores / 256 threads
• 2TB ECC Reg DDR4 system memory
• 2x 1.92TB NVMe SSDs system drives
• 8x 3.84TB NVMe SSDs storage drives

DGX Station A100

The second generation of the DGX station appliance introduces A100 GPUs with double-precision Tensor Cores, providing the biggest milestone HPC and AI training in a desktop server format that doesn’t require a datacentre environment.

This model was discontinued in 2022.

• 4x NVIDIA A100 Tensor Core GPUs
• 6,912 CUDA cores & 432 TF32 Tensor Cores per GPU
• 80GB memory per GPU - 320GB total
• AMD EPYC 7742 - 64 cores / 128 threads
• 512GB ECC Reg DDR4 system memory
• 1.92TB NVMe SSD system drive
• 7.68TB NVMe SSD storage drive

NVIDIA DGX-2

The second-generation DGX-2 appliance was the world’s first 2 petaFLOPS system integrating 16 NVIDIA V100 Tensor Core GPUs for large-scale AI projects. It was launched in 2018 and superseded by the DGX A100 in 2020.

• 5,120 CUDA cores & 640 Tensor Cores per GPU
• 32GB memory per GPU - 512GB total
• 2x Intel Xeon Platinum 8168 - total 48 cores / 96 threads
• 1.5TB ECC Reg DDR4 system memory
• 2x 960B NVMe SSDs system drives
• 8x 3.84TB NVMe SSDs storage drives

NVIDIA DGX Station V100

The original DGX station appliance introduced V100 GPUs in a desktop server format that doesn’t require a datacentre environment, aimed at delivering AI development and training for workgroups. This model was superseded by the Station A100 in 2020.

• 4x NVIDIA V100 Tensor Core GPUs
• 5,120 CUDA cores & 640 Tensor Cores per GPU
• 32GB memory per GPU - 128GB total
• Intel Xeon E5-2698 v4 - total 20 cores / 40 threads
• 256GB ECC Reg DDR4 system memory
• 4x 1.92TB SSDs

NVIDIA DGX-1

The original DGX-1 appliance was the world’s first dedicated AI supercomputer, featuring eight GPUs connected by NVLink technology. It was launched in 2016 and superseded by the DGX-2 in 2018.

• 8x NVIDIA V100 Tensor Core GPUs
• 5,120 CUDA cores & 640 Tensor Cores per GPU
• 32GB memory per GPU - 256GB total
• 2x Intel Xeon E5-2698 v4 - total 40 cores / 80 threads
• 512GB ECC Reg DDR4 system memory
• 4x 1.92TB SSDs