A dark pool for GPU compute. Orders go in sealed and clear at one price per batch, so nobody sees your hand.
On an open GPU market, every job is a signal.
A public order for 512 H100s tells everyone what you are building. Change the order below and watch what each side gives away.
reserve(H100 SXM × 512, 72 h, ≤ 2.40 $/h)
…
This is the real commitment format the pool contract checks at the reveal, computed in your browser with a random 32-byte salt. Nothing is sent anywhere.
The whole idea, without a word of jargon.
What leaks on an open market, what a sealed order hides, and how everyone in a round ends up paying the same price.
Commit. Reveal. Clear at one price.
Post a hash and a 1 USDG bond. The order stays in your browser.
Open the order. Funds lock, the bond comes back.
Bids and asks cross at one price for everyone.
The batch sets the price.
The open-market median is shown only as a yardstick.
| Class | VRAM | Open-market median | Offers |
|---|---|---|---|
B200 B200 NVIDIA · frontier | 192 GB | $10.00 /h | 19 |
H200 H200 NVIDIA · frontier | 141 GB | $4.61 /h | 14 |
H100 H100 SXM NVIDIA · datacenter | 80 GB | $3.79 /h | 12 |
H100P H100 PCIe NVIDIA · datacenter | 80 GB | $3.32 /h | 4 |
A100 A100 SXM4 NVIDIA · datacenter | 80 GB | $0.845 /h | 31 |
L40S L40S NVIDIA · workstation | 48 GB | $0.801 /h | 28 |
5090 RTX 5090 NVIDIA · consumer | 32 GB | $0.574 /h | 64 |
4090 RTX 4090 NVIDIA · consumer | 24 GB | $0.513 /h | 64 |
3090 RTX 3090 NVIDIA · consumer | 24 GB | $0.198 /h | 64 |
Clearing rule, bonds and windows in full in the docs.
What stays sealed.
- GPU class
- Rent or provide
- Size and price
- Machine access, always encrypted
- Vault deposits
- That you committed
- Fills and lease streams
The most private GPU is your own.
Hold more, get in first.
Lock the token to reach a tier. No yield, no fee share: access.
Priority at equal price
When two orders name the same price, the higher tier fills first.
Larger orders
The order size cap grows 4×, then 16×, then disappears.
Dark Rooms
Holder-only books, and first entry into every new GPU class.