Loading
Loading
Prices are our ballpark as of 2026-09-27, not a live store. Memory is what the planner uses.
One DGX Spark holds about a 200 billion parameter model after it is compressed. Two is the size most people finish. Four reaches the largest class and usually needs a fast network switch.
4 of 4 machines
Memory
96 GB VRAM
Price
Workstation money, roughly $8–12k
What it can run
70B with a long context, or a 100B-class quant, still on one card.
Best for
A single workstation that should not become a cluster.
Memory
Up to 128 GB unified
Price
$1,500–2,500 for a mini PC or Framework Desktop
What it can run
Large MoE and 70B-class GGUF. Bandwidth is the limit, not a missing GPU slot.
Best for
One box, lots of memory, no NVIDIA driver story.
Memory
Grace Blackwell workstation class
Price
Official quote. No authored street range.
What it can run
A lab workstation, not a desktop GPU swap.
Best for
A lab that already buys NVIDIA workstations.
Memory
Rack scale
Price
No street price. This is not a listing.
What it can run
A cluster. Not a model you park under a desk.
Best for
Knowing the ceiling exists.