GTX 1070 vs RTX 3080

FP32 throughput on the RTX 3080 measures 29.8 TFLOPS against 6.5 TFLOPS on the GTX 1070 for a ratio of 4.58 times. The RTX 3080 also lists 59.5 TFLOPS dense FP16 which equals exactly twice its own 29.8 TFLOPS dense FP32 value. No FP16 number exists for the GTX 1070 so direct dense FP16 comparison remains unavailable. Memory bandwidth of 760 GB/s on the RTX 3080 exceeds the 256 GB/s figure on the GTX 1070 by a factor of 2.97. This bandwidth increase supports larger batch sizes in memory bound operations. INT8 dense performance reaches 238 TOPS on the RTX 3080 with no corresponding figure published for the GTX 1070. Sparsity doubles the listed INT8 value to 476 TOPS on the RTX 3080 while FP16 with sparsity reaches 119 TFLOPS.

The GTX 1070 suits environments limited to 150W power budgets where 6.5 TFLOPS FP32 meets baseline requirements. Its 8 GB GDDR5 configuration and PCIe 3.0 interface fit older server nodes that lack PCIe 4.0 support. Workloads that do not require tensor operations or high bandwidth avoid the 320W TDP of the RTX 3080 and stay within the GTX 1070 thermal envelope.

The RTX 3080 fits tasks that scale with 29.8 TFLOPS FP32 and 59.5 TFLOPS dense FP16. Its 760 GB/s bandwidth enables larger batches than the 256 GB/s available on the GTX 1070. Users needing INT8 dense throughput of 238 TOPS select the RTX 3080 over the GTX 1070 which publishes no INT8 value.