Product Catalog
Prices and power draw are approximate, based on publicly reported specs -- actual pricing varies by configuration, region, and negotiation.
| Product | Price | Power Draw |
|---|---|---|
NVIDIA RTX 4090 Desktop GPU |
$1,599 About the price of a high-end gaming laptop. |
450 W About the same draw as 4 hair dryers running at once. |
NVIDIA RTX 6000 Ada Workstation GPU (48GB) |
$6,800 About the price of a used sedan. |
300 W About the same as running 3 hair dryers at once. |
NVIDIA L40S Datacenter GPU (48GB) |
$11,000 About the price of a brand-new economy car. |
350 W Enough power to run a home oven continuously. |
NVIDIA A100 80GB Datacenter GPU |
$10,000 About the price of a small new car. |
300 W Similar to running central air conditioning nonstop. |
NVIDIA H100 80GB SXM Datacenter GPU |
$30,000 About the price of a luxury car. |
700 W More than half the power draw of an entire average home. |
NVIDIA H200 141GB Datacenter GPU |
$35,000 About the price of a top-trim luxury car. |
700 W Same as H100 -- more than half an average home's power draw. |
NVIDIA DGX H100 Complete AI server (8x H100 inside) |
$300,000 About the price of a house in many US cities. |
10,200 W ≈ 8.5 average homes' worth of power, running around the clock. |
NVIDIA GB200 NVL72 Rack Full datacenter rack (72 GPUs) |
$6,500,000 About the price of a private jet. |
120,000 W ≈ 100 average homes' worth of power draw, all at once. |
Open-Source AI Models (100B+ parameters)
Our rule of thumb for minimum GPU memory: parameters (in billions) × 1GB, plus 20% headroom for the model to actually run.
| Model | Parameters | Min. Memory Needed | Minimum Hardware |
|---|---|---|---|
Llama 3.1 405B Meta Meta's largest open-weight model -- a strong general-purpose reasoner and coder. |
405B | 486.0 GB 405B × 1GB × 1.2 (20% extra) |
7x H100 80GB or 1x DGX H100 system(s) |
DeepSeek-V3 DeepSeek AI A mixture-of-experts model with 671B total parameters, competitive with top closed models. |
671B | 805.2 GB 671B × 1GB × 1.2 (20% extra) |
11x H100 80GB or 2x DGX H100 system(s) |
Falcon 180B Technology Innovation Institute (TII) One of the largest fully open-license models, built for research and commercial use. |
180B | 216.0 GB 180B × 1GB × 1.2 (20% extra) |
3x H100 80GB or 1x DGX H100 system(s) |
Customer Path: Small AI Startup
You're a small team building a chatbot or fine-tuning smaller open models. You need real GPU power without a datacenter budget.
Recommended Build: 4x NVIDIA RTX 4090 workstation
Power draw of 1,800 watts is like running 1.5 average homes at once (1 home ≈ 1200 watts). Run continuously for a day, that's 43.2 kWh -- about 0.48 full EV battery charges (1 EV charge ≈ 90 kWh).
Customer Path: Mid-Size Company
You're running a production AI product -- for example a support assistant powered by a 70B-180B parameter model -- serving real customer traffic every day.
Recommended Build: 1x NVIDIA DGX H100 system (8x H100 GPUs)
Power draw of 10,200 watts is like running 8.5 average homes at once (1 home ≈ 1200 watts). Run continuously for a day, that's 244.8 kWh -- about 2.72 full EV battery charges (1 EV charge ≈ 90 kWh).
What's a Cluster?
Get a Quote
Tell us who you are and which build you're interested in -- we'll follow up.