NVIDIA A100 40GB PCIe is the lower-capacity variant of the famous A100 Ampere data center GPU, with 40GB ECC HBM2e memory. It supports multi-precision Tensor Core calculation including TF32, BF16, FP16 and INT8.
MIG function allows splitting one GPU into up to seven isolated GPU instances, perfect for shared-server multi-tenant inference. Passive cooling, PCIe 4.0 dual-slot design, compatible with regular rack servers.
Compared with the 80GB version, it delivers similar core compute performance at a lower procurement cost, ideal for teams running 7B~34B model inference and mid-sized simulation jobs. No display outputs.