See the whole market before you commit
Every connected GPU, what it serves, how fast it measured, whether it is free — beside what customers asked for and could not get. Public, live, anonymised.
Most networks ask you to join before they show you anything. Ours is open: the supply and the demand, on one page, without an account.
If you have a GPU
The page opens on requests nobody could serve — demand that already exists and is going unanswered. Load one of those models and you are earning from the first hour, rather than guessing what might be wanted.
You also see what every other card is serving, at which quantization and how fast it measured. That tells you where you would be one of many and where you would be the only one.
If you are building
You see how many machines are free right now for the model you want, what they measured, which quantizations are live, and which regions they are in. No sales call, no quota conversation — the capacity is either there or it is not, and you can look.
Free means free, not "enrolled"
A card counts as free only when its agent is connected and asking for work and holds no job. There is no availability flag anyone maintains, so there is none to be out of date: the queue already knows.
Measured, never estimated
A speed appears only where that machine benchmarked that build itself. We publish the median across hosts rather than the fastest, because a single quick card should not make the network look better than your request will actually get.
Anonymised
A row says "an AMD 24 GB card in Belgium serving Qwen3.8 at IQ4_XS, 58 tok/s, free". It never says whose card it is: no identifier, no owner, no hostname, no address.
Related
Model catalogue · Download mahout · How the whole system works · Blog