GTX 1080 vs RTX 3070

The FP32 figure of 20.3 TFLOPS on the RTX 3070 exceeds the 8.9 TFLOPS on the GTX 1080. This difference translates to greater throughput for training and inference operations that rely on single precision arithmetic. Memory bandwidth at 448 GB/s versus 320 GB/s permits larger batch sizes through accelerated data movement between memory and compute units. The RTX 3070 also lists dense FP16 at 40.6 TFLOPS while the GTX 1080 provides no published FP16 value. Dense FP16 comparisons therefore apply only to the RTX 3070 against itself when sparsity is not used. With sparsity the RTX 3070 reaches 81.3 TFLOPS FP16.

The GTX 1080 suits scenarios that prioritize its 180W TDP over the 220W TDP of the RTX 3070. Systems limited to PCIe 3.0 interconnect benefit from the GTX 1080 specification without requiring PCIe 4.0 support. Workloads that operate solely on the published 8.9 TFLOPS FP32 avoid any need for the additional FP16 dense figures available only on the RTX 3070.

The RTX 3070 fits workloads that require its 20.3 TFLOPS FP32 and 448 GB/s memory bandwidth. Tasks that utilize dense FP16 at 40.6 TFLOPS gain from the Ampere architecture numbers absent on the GTX 1080. Higher INT8 dense performance at 162.6 TOPS further supports inference pipelines that scale with these exact specifications.