RTX 2070 vs RTX 4080

The dense FP16 rating of 29.9 TFLOPS on the RTX 2070 versus 97.5 TFLOPS on the RTX 4080 indicates that the RTX 4080 can sustain larger batch sizes during training and inference when both figures are treated as dense. The FP32 rating of 7.5 TFLOPS on the RTX 2070 versus 48.7 TFLOPS on the RTX 4080 produces a similar ratio for single precision tasks. Memory bandwidth of 448 GB/s on the RTX 2070 versus 717 GB/s on the RTX 4080 limits the maximum batch size on the older card before data movement becomes the bottleneck. The dense INT8 rating of 119.4 TOPS on the RTX 2070 versus 389.9 TOPS on the RTX 4080 follows the same pattern when sparsity is not applied. The RTX 4080 also lists a dense FP8 rating of 194.9 TFLOPS that has no counterpart published for the RTX 2070.

The RTX 2070 suits environments where power draw must remain at or below 175W. It also fits workloads that fit within 8 GB GDDR6 and require only 29.9 TFLOPS dense FP16 or 7.5 TFLOPS FP32. PCIe 3.0 interconnect further aligns with older host systems that lack PCIe 4.0 support.

The RTX 4080 suits workloads that require 16 GB GDDR6X and 717 GB/s bandwidth to accommodate larger models.