Skip to content
sulimko.com

NVIDIA DGX Spark

NVIDIAon sale

Memory, GBmaker's claim
claimed
Memory speed, GB/s
On this page

What it can run

The agent count is a memory estimate for this model and context. The speed check below is a separate estimate, not a measured rate per agent.

How it is calculated

How many agents of each open model fit in its memory at once, each with room for about 32 thousand tokens of conversation — a typical coding agent. The agents share one copy of the model, so the first one needs the most memory and each next adds only its own conversation. Our estimate from memory size; open a row to try other settings.

Available to models: 96 GB of 128 GB. A quarter of the memory is left to the system.

Measured speed

Public tests by the community: how many tokens per second one agent gets.

  • Tokens per second
    4-bit, as released · uses toolsOct 14, 2025
  • Tokens per second
    8-bit, closer to the original · uses toolsOct 14, 2025
  • Tokens per second
    4-bit, as released · uses toolsOct 14, 2025
Fit for agents, of 100our score
no confirmed price in our data yet
Price, USDno confirmed price in our data yet

Specification

Memory
Memory speed
claimed
Power draw
Combines with others
yes
Price
no confirmed price in our data yet

How the fit score is made

Each part compares this computer with the five others on the page: 100 is the best of them, 0 the weakest. The weight says how much the part counts.

  • Memory speedcounts 35%10
  • Memory sizecounts 20%100
  • Measured speedcounts 20%the only measured computer, so it gets the top mark100
  • Speed per wattcounts 10%the only measured computer, so it gets the top mark100
  • Price (cheaper is better)counts 10%no data — left out
  • Combines with otherscounts 5%100

Sources