Full Catalog
Every GPU, CPU, memory module, server, rack, and support plan we carry — from a $249 Intel Arc card to a $6.25M GB300 NVL72 rack. Filter by category, vendor, and market tier.
41 products
DGX OS
Included free with every DGX system and NVL72 rack.
CUDA Toolkit
Free, and required for anything with an NVIDIA GPU in it.
Arc B580
The cheapest real GPU on the shelf — Intel's Battlemage value champion.
Arc A770
First-gen Arc with 16GB — a cheap way to have more VRAM than a 4060.
DDR5 64GB RDIMM
The baseline server memory module for any host build in this store.
Ubuntu Pro
Canonical's long-term security and compliance layer for your OS image.
RTX 5070
The affordable on-ramp to Blackwell — enough VRAM for quantized 7B–13B models.
DDR5 128GB RDIMM
Double the density for memory-hungry preprocessing and RAG pipelines.
RTX 5070 Ti
The mainstream Blackwell pick — most of the 5080's memory for less.
RTX 5080
The sensible sibling to the 5090 — still a capable local-inference card.
RX 7900 XTX
AMD's consumer flagship — big VRAM per dollar if you're ROCm-friendly.
RTX 4090
Last generation's flagship — still a very real local-AI workhorse.
ConnectX-7
The nervous system between your GPUs and everything else.
1yr Standard Support
Business-hours support and next-business-day parts replacement.
L4
The entry point. Cheap, low-power, and plenty for small inference jobs.
RTX 5090
The fastest consumer card NVIDIA makes — a hobbyist's on-ramp to local AI.
DDR5 512GB Kit
A full 8-DIMM kit — one line item instead of eight.
AI Enterprise
The supported software layer — optimized frameworks, containers, and SLAs.
EPYC 9654
96 Zen 4c cores of host CPU to keep your PCIe GPUs fed.
L40S
Ada Lovelace's do-everything card — fine-tuning, inference, and graphics.
3-Phase PDU
Because an 8-GPU server pulling 10kW is not going into a standard outlet.
Grace Superchip
144 Arm cores with NVLink-C2C — the CPU half of every GB200/GB300 rack.
3yr Priority Support
24/7 priority response with a 4-hour SLA, for hardware you can't afford downtime on.
A100 PCIe
The Ampere legend, still training real models for a fraction of Hopper cost.
Xeon 8592+
Intel's flagship Emerald Rapids host CPU for enterprise GPU nodes.
RTX PRO 6000
96GB of workstation VRAM — the biggest single card you can put in a desktop.
MI300X
AMD's answer to H100/H200 — more memory, a real second sourcing option.
A100 SXM4
NVLink-connected Ampere for real multi-GPU training pods.
Spectrum SN4700
The top-of-rack Ethernet switch for management and storage traffic.
5yr Mission-Critical
A named TAM, 4-hour onsite response, and rack-level spares on-site.
H100 PCIe
Drop-in Hopper for standard servers — no SXM baseboard required.
Rack CDU
The liquid-cooling heart every rack-scale system needs to survive.
H100 SXM5
The Hopper workhorse. Still the backbone of most training clusters.
H200 SXM5
H100's memory-hungry sibling — built for large-context inference.
B200 SXM6
Blackwell has arrived. The current frontier-training standard.
Quantum-2 Switch
64 ports of 400G NDR InfiniBand — the fabric spine for a real cluster.
Flash Storage Node
~1PB usable of NVMe, fast enough to keep GPUs from starving.
DGX H100
Eight H100s, one NVLink domain, zero integration headaches.
DGX B200
The Blackwell-generation DGX — 3x the training throughput of DGX H100.
GB200 NVL72
72 Blackwell GPUs and 36 Grace CPUs acting as one giant GPU.
GB300 NVL72
Blackwell Ultra. The most powerful rack NVIDIA has ever shipped.