Skip to main content

GPU comparison

Use the following table to compare the number of GPUs, RAM, and storage installed in CoreWeave’s GPU compute instances. Select an instance type to view full specifications.
1 Local disks are assembled into a RAID array whose layout depends on the number of drives. On a mirrored array (RAID1 or RAID10), usable space is about half the total capacity listed; an instance with a single drive has no RAID and the full capacity is usable. See Local storage for details.

Choose the right GPU for your workload

Not sure which GPU type fits your model? See Select an instance for a decision guide that covers:
  • GPU memory estimation by model size and precision (FP16, INT8, INT4)
  • Inference vs. training and fine-tuning memory overhead
  • NVLink vs. InfiniBand guidance for multi-GPU workloads
  • Example models that fit each GPU type
To check which instances are offered in each region, see the Instance availability matrix.
Last modified on June 29, 2026