Dedicated Mac Studio machines, by the day.
128 GB of unified memory, full SSH access, paid upfront — no queues, no sharing.
- From $20 a day
- Book 1 to 5 days, up to 8 weeks ahead
- Wiped clean after every reservation
$ ssh user@the-last-in-pipeline.res-4ad0a522.ssh.gputolease.ai% system_profiler SPHardwareDataType | grep -E "Chip|Memory" Chip: Apple M5 Max Memory: 128 GB% Available to reserve
A whole machine for you alone, for 1 to 5 days.
Mac Studio M5 Max · 128 GB
Apple M5 Max
A dedicated Mac Studio with the M5 Max chip and 128 GB of unified memory, reserved by the day.
- Chip
- Apple M5 Max
- CPU
- 18 cores
- 6 super + 12 performance
- GPU
- 40 cores
- Memory
- 128 GB unified
- 614 GB/s bandwidth
- Storage
- 1 TB SSD
- Dedicated physical machine — never shared
- Full SSH access with your own key
- 128 GB unified memory — run 70B-class models locally
- Optional startup script runs when your machine is ready
- Wiped clean at the end of your reservation
Coming soon
More machines on the way. They can't be reserved yet.
Mac Studio M5 Ultra · 96 GB
96 GB unified memory @ 1.2 TB/s
1 TB SSD
Mac Studio M5 Ultra · 256 GB
256 GB unified memory @ 1.2 TB/s
2 TB SSD
NVIDIA DGX Spark · 128 GB
128 GB LPDDR5x unified memory @ 273 GB/s
1 TB NVMe (self-encrypting)
How it works
From sign-up to a shell on your machine in four steps.
- 1
Pick a machine
Choose a machine from the catalog. Each one is a dedicated physical computer, never shared.
- 2
Add your SSH key
Paste your public key once. It's installed on every machine you reserve, so you log in with it.
- 3
Choose your days
Pick 1 to 5 days in a row on the calendar, up to 8 weeks ahead. Green means free.
- 4
Pay and connect
Pay upfront by card. When the reservation starts, copy the SSH command from your dashboard.
Why unified memory
Large models need memory more than anything else. A Mac Studio has one large pool that the CPU and GPU share.
One pool, no copying
The CPU and GPU read the same memory. Model weights load once: there's no separate graphics card with its own, smaller memory to fit them into.
Room for 70B-class models
128 GB holds a 70-billion-parameter model quantised to 4 bits, about 40 GB of weights, with room left for a long context.
Bandwidth sets the pace
Each generated token reads the model's weights from memory, so speed follows bandwidth: 614 GB/s on the Apple M5 Max.
From $20 a day
You pay for the days you book, upfront. No subscription, and every day costs the same.
Questions
Get a machine of your own
Pick your days, pay upfront and connect over SSH when your reservation starts.