The full lineup
Catalog
Every price, spec, and wattage below is a real, current, sourced number — the same figure appears everywhere this product shows up on the site.
NVIDIA GeForce RTX 5090
The on-ramp: a single desktop card for prototyping with open models
- GPU memory32 GB
- Power draw575 W
- GPUs in unit1
NVIDIA RTX 6000 Ada Generation
A quiet workstation card built to run flat-out, 24/7
- GPU memory48 GB
- Power draw300 W
- GPUs in unit1
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
The biggest single card money can buy — no server room required
- GPU memory96 GB
- Power draw600 W
- GPUs in unit1
NVIDIA H100 SXM
The GPU that started the modern AI datacenter buildout
- GPU memory80 GB
- Power draw700 W
- GPUs in unit1
NVIDIA H200 SXM
Same compute as the H100, with far more memory to work with
- GPU memory141 GB
- Power draw700 W
- GPUs in unit1
NVIDIA B200
NVIDIA's current-generation Blackwell datacenter GPU
- GPU memory192 GB
- Power draw1,000 W
- GPUs in unit1
NVIDIA DGX H100
Eight H100s, pre-cabled into one appliance
- GPU memory640 GB
- Power draw10,200 W
- GPUs in unit8
NVIDIA DGX H200
The same trusted DGX chassis, upgraded to H200 memory
- GPU memory1,128 GB
- Power draw10,200 W
- GPUs in unit8
NVIDIA DGX B200
Eight Blackwell GPUs in NVIDIA's current flagship server
- GPU memory1,536 GB
- Power draw14,300 W
- GPUs in unit8
NVIDIA GB200 NVL72
72 GPUs and 36 CPUs, wired into one liquid-cooled machine
- GPU memory13,824 GB
- Power draw120,000 W
- GPUs in unit72
NVIDIA GB300 NVL72
NVIDIA's largest rack-scale system, tuned for serving reasoning models at scale
- GPU memory20,736 GB
- Power draw140,000 W
- GPUs in unit72