Answer a few questions and get a recommended GPU tier with an estimated monthly range. All figures are placeholders until pricing is finalized — pre-orders get final pricing first.
This estimator uses placeholder rates so you can gauge scale. Final pricing is set later and shared with pre-order and waitlist members first.
32 GB: many 7B–13B inference workloads, depending on precision, context and batch size. 96 GB: larger inference workloads, including many quantized 70B-class models. 141 GB+: full-precision or high-context models, larger batches and memory-intensive scientific workloads. Fine-tuning and training can require substantially more memory than inference.
Reserved means dedicated GPUs, always yours, billed monthly. Flexible is cheaper per hour but capacity can be shared. Not sure? Start flexible.
Send us your workload and we’ll sanity-check the estimate and recommend a setup before you commit to anything.
Get a recommendation