GPU Fleet

From cost-effective T4s to cutting-edge B300s

GPU Memory / GPU Per Node Multinode Interconnect
B300Blackwell · Latest
192 GB
8 up to 48 NVLink + EFA NEW
B200Blackwell
192 GB
8 up to 48 NVLink + EFA NEW
H200Hopper · High Mem
141 GB
8 up to 48 NVLink + EFA POPULAR
H100Hopper
80 GB
8 up to 48 NVLink + EFA POPULAR
A100Ampere
40 GB
8 up to 48 NVLink + EFA
RTX PRO 6000Blackwell · Desktop
48 GB
4 PCIe NEW
L4Ada Lovelace
24 GB
4 PCIe BUDGET
T4Turing
16 GB
4 PCIe BUDGET
CPUARM (c7g) + x86 (c7i)
DEV/BUILD
+ MIG slices available on H100 and B200  ·  Spot instances available (~70% cheaper)  ·  Custom Docker images supported