Smart Launch
Say what you run.Get the GPU that fits.
Type a model. The real engine runs in your browser.
Smart Launch · the real engine, in your browser.
$ llama 3.1 70b 4-bit 32k
Llama 3.1 70B Instruct · 4-bit · 32,768 tokens · 1 at a time
Model weights 32.9 GiB
KV cache 10.0 GiB
Overhead margin 8.6 GiB
CUDA runtime 1.0 GiB
Needs 52.5 GiB
B300 270 GB fits
B200 180 GB fits
H200 141GB SXM 141 GB fits
RTX PRO 6000 96 GB fits
H100 80GB SXM 80 GB fits
H100 80GB PCIe 80 GB fits
A100 80GB SXM 80 GB fits
L40S 48 GB fits on 2
L40 48 GB fits on 2
RTX A6000 48 GB fits on 2
RTX A4000 16 GB fits on 4
Medium confidence. Same input, same answer.
Also: Llama 3.3 70B Instruct
Your turn: type a model, or help.
The answer
A calculation. Never a guess.
Ranking
Measured performance per dollar first.
Reasons
Why it fits, its limits, its confidence.
Location
Only the countries you choose.
Environments
Ready at launch. Pinned by digest.
- JupyterLab
- JupyterLab 4
- PyTorch
- PyTorch 2.9.0 · CUDA 12.8
- vLLM
- vLLM 0.11.0 · CUDA 12.8
- Ollama
- Ollama 0.12.6
Launch with a budget.
Never billed past it.
Request access.
Tell us what you will run. Your access arrives by email.