Real NVIDIA Hardware, Four Tiers
From desktop GPUs to complete AI servers. Every card shows The dedicated memory built into the graphics card. An AI model's entire set of "weights" (its learned parameters) must fit into this memory before the card can run it. If a model needs more VRAM than one card has, you need multiple cards working together, or a card with more memory. , Thermal Design Power — roughly how many watts the card draws under full load. Higher-performance cards need more electricity and generate more heat, which affects your electricity bill and cooling needs. , and Approximate cost to acquire the hardware. For consumer cards this is usually a fixed retail price. For datacenter-class hardware, list prices are rarely published — buyers negotiate directly with NVIDIA or its partners, so we show a typical range or say "contact for pricing" rather than inventing a number. with a plain-English explanation of why each spec matters for running AI models. Want to line products up directly? Try the comparison tool.
Consumer GPUs
NVIDIA GeForce RTX 5090
| GPU Memory | 32 GB |
|---|---|
| Power | 575 W |
| Price | $1,999 |
Hobbyists and developers experimenting with smaller open-source models on a single desktop.
NVIDIA GeForce RTX 5080
| GPU Memory | 16 GB |
|---|---|
| Power | 360 W |
| Price | $999 |
Budget-conscious entry point for running small quantized models locally.
NVIDIA GeForce RTX 4090
| GPU Memory | 24 GB |
|---|---|
| Power | 450 W |
| Price | $3,474 |
Still a popular high-VRAM consumer card for local AI work, though now only available secondhand at a premium.
Workstation GPUs
NVIDIA RTX PRO 6000 Blackwell (Workstation Edition)
| GPU Memory | 96 GB |
|---|---|
| Power | 600 W |
| Price | $11,636 |
Small teams needing large single-card memory for mid-size open-source models without full datacenter infrastructure.
NVIDIA RTX PRO 5000 Blackwell
| GPU Memory | 48 GB |
|---|---|
| Power | 300 W |
| Price | $6,472 |
A mid-tier professional card balancing memory, power draw, and cost for growing workloads.
NVIDIA RTX 6000 Ada Generation
| GPU Memory | 48 GB |
|---|---|
| Power | 300 W |
| Price | $6,800 |
Previous-generation professional card, still widely deployed, same VRAM class as the newer RTX PRO 5000.
Enterprise / Datacenter GPUs
NVIDIA H100 (SXM)
| GPU Memory | 80 GB |
|---|---|
| Power | 700 W |
| Price | $32,500 |
The GPU inside most 2023-2024-era AI training/inference servers. Sold as part of an OEM server (e.g. NVIDIA DGX), not standalone.
NVIDIA H200 (SXM)
| GPU Memory | 141 GB |
|---|---|
| Power | 700 W |
| Price | $35,000 |
H100's larger-memory successor, better suited to today's 100B+ parameter open models. Sold as part of an OEM server.
NVIDIA B200 (Blackwell)
| GPU Memory | 180 GB |
|---|---|
| Power | 1000 W |
| Price | Contact for pricing |
NVIDIA's newest datacenter GPU, the building block of the DGX B200 system below.
Full AI Server Systems
NVIDIA DGX A100
| GPU Memory | 640 GB total (8× 80GB) |
|---|---|
| Power | 6500 W |
| Price | $199,000 |
A complete, previous-generation 8-GPU AI training/inference server — the most affordable entry into full-server-class infrastructure.
NVIDIA DGX H100
| GPU Memory | 640 GB total (8× 80GB) |
|---|---|
| Power | 10200 W |
| Price | $390,000 |
A complete 8-GPU H100 server for organizations that need serious throughput today.
NVIDIA DGX B200
| GPU Memory | 1440 GB total (8× 180GB) |
|---|---|
| Power | 14300 W |
| Price | $387,500 |
NVIDIA's current flagship AI server, with enough per-node memory to run even the largest open-source models we sell without a multi-node cluster.