GTX 1080 vs RTX 3090

The RTX 3090 FP32 figure of 35.6 TFLOPS exceeds the GTX 1080 figure of 8.9 TFLOPS by a factor of four. Dense FP16 performance reaches 71 TFLOPS on the RTX 3090 while the GTX 1080 figure remains unpublished by the vendor. This gap means training and inference steps complete faster on the RTX 3090 when dense FP16 operations apply. Memory bandwidth of 936 GB per s on the RTX 3090 versus 320 GB per s on the GTX 1080 supports larger batch sizes before memory limits appear. Sparse FP16 on the RTX 3090 reaches 142 TFLOPS and sparse INT8 reaches 568 TOPS allowing further acceleration on supported workloads. The GTX 1080 TDP of 180W remains lower than the RTX 3090 TDP of 350W so sustained operation costs differ in power constrained settings.

The GTX 1080 suits environments where power limits cap consumption at 180W and workloads fit within 8 to 11 GB VRAM. Scientific computing tasks that rely solely on the published 8.9 TFLOPS FP32 complete without need for higher figures. Older PCIe 3.0 interconnects match existing host systems that lack NVLink support.

The RTX 3090 suits workloads that scale with 24 GB VRAM and 936 GB per s bandwidth. Dense FP16 at 71 TFLOPS accelerates training steps beyond the GTX 1080 capability. PCIe 4.0 and NVLink options enable faster multi GPU scaling when interconnect bandwidth matters.