01
Connect
Any Solana wallet is your account. No email, no KYC queue, no invoice net-30.
select wallet▮
Phantom● connected
Backpack
Solflare
7xKX…9fQa◎ 42.0810
● mainnet-beta / live[ click anywhere → shockwave ]
Rent RAM.by the second,
settled on-chainPay SOL
On-demand VRAM from 14,382 GPUs, metered per second and streamed from your wallet. No card, no contract, no minimum. Plug in, allocate, walk away.
- slot
- syncing…
- ◎/GB·hr
- 0.000190
- finality
- 400ms
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
B200192GBHBM3e
H200141GBHBM3e
H100 SXM80GBHBM3
MI300X192GBHBM3
L40S48GBGDDR6
RTX 509032GBGDDR7
[01]The thesis
Most of the world's VRAM sits idle — warm, expensive, waiting. We slice it into 1 GB gigabytes, price every one by the second, and let anyone with a wallet rent it. No sales call. No credit check. Just memory, streaming to whoever needs it, paid in SOL.
[02]Protocol — four steps, zero paperwork
Wallet
to GPU
in 2s
Scroll →
02
Allocate
Pick a card and carve out exactly the gigabytes you need — from 8 GB to a full 8×B200 node.
allocated0 GB
03
Stream
Payment flows per 100ms from your wallet to the host. You pay for memory you actually hold.
wallet
◎ 42.0810000
node
FRA-1 · H200
0.0000521/ 100ms
stop anytime — unused lamports never leave your wallet.
04
Run
SSH, Docker or Jupyter in under two seconds. Your weights, your stack, bare-metal speed.
root@fra-1-h200
$ vram alloc --gpu h200 --gb 141 --pay sol
✓ stream opened sig 4Rk2…xQ1
✓ node FRA-1 attached in 1.8s
$ nvidia-smi --query-gpu=name,memory.total --format=csv
NVIDIA H200, 143771 MiB
$ python serve.py --model llama-4-405b
loaded 405B params → 138.2 GB VRAM
ready on :8000 ▮
[03]Allocator
Drag. Price. Own it.
VRAM
282GB
NVIDIA H200
141 GB HBM3e
4.8 TB/s
8 GB1,128 GB · 8× H200
Duration
memory map36/144 blocks
GPU0
GPU1
GPU2
GPU3
GPU4
GPU5
GPU6
GPU7
you pay
1.624
lamports / sec
18,800◎ / GB·hr
0.00024[04]Live memory marketstreaming
0
GPUs online
0 PB
VRAM pooled
0ms
Settlement
0%
Cheaper than cloud
nodegpuregionfree vram◎/GB·hrping
iad-2-h200-018× H200IAD-2554 GB0.00023447msrent →iad-2-mi300x-024× MI300XIAD-2674 GB0.00015553msrent →sin-1-h200-034× H200SIN-1408 GB0.00023161msrent →syd-1-h200-042× H200SYD-1179 GB0.00020040msrent →sin-1-b200-052× B200SIN-1225 GB0.00028340msrent →fra-1-l40s-061× L40SFRA-119 GB0.00009828msrent →syd-1-b200-074× B200SYD-1402 GB0.0003117msrent →sin-1-h200-084× H200SIN-1424 GB0.00024872msrent →tyo-3-h200-098× H200TYO-3788 GB0.00028252msrent →