How much has GPU performance evolved in 10 years? Comparing the GTX 980 to the RTX 5090
From the NV1 to the RTX 50 series, NVIDIA has released many desktop GPUs over the past few decades. A comparison of major products from the GeForce GTX 980 released in 2014 to the GeForce RTX 5090 released in 2025 has been published on the technology site Phoronix.
NVIDIA GeForce GTX 980 Through GeForce RTX 5080/5090 GPU Compute Performance Review - Phoronix
https://www.phoronix.com/review/nvidia-maxwell-to-blackwell
First, here is a comparison of the time it took to render the 'Classroom' scene in Blender 4.3. When using NVIDIA OptiX, the GTX series without ray tracing cores is noticeably slower, and there is a significant difference between the RTX 20 series and the RTX 40 series.
Even using Blender's CUDA backend rather than OptiX, we see incredible performance improvements from the GTX series to the RTX series.
When rendering the 'Barbershop' scene with CUDA, the RTX 5080 takes just 7% of the time of the GTX 980.
Power efficiency is also improved as the generation advances, with the RTX 40 series outperforming the RTX 50 series. The horizontal axis is the average joules per run, the lower the better.
Using Blender 4.3's OptiX renderer, the RTX 5080 can complete a render in 5% of the time of the GTX 980.
The Barbershop/OptiX combination also puts the RTX 40 series at the top of the power efficiency list. Phoronix noted that 'NVIDIA's longstanding support for GPU acceleration has made the company a leading choice for Blender users. In recent years, Intel oneAPI and AMD ROCm/HIP have been supported, but the mature hardware support seen from NVIDIA is still largely missing.'
When the fluid analysis software 'FluidX3D' was run with FP32-FP16S compute, the performance difference between the GTX 980 and the RTX 5090 was 9.4 times. The horizontal axis is
Synthetic performance tests for GPU memory bandwidth using OpenCL also show the RTX 5090 outperforming.
In a processing speed test using Llama's large language model, the RTX 4090 outperformed the RTX 5080. The horizontal axis shows the number of tokens processed per second, and the longer the bar, the better the processing performance.
Power efficiency was roughly equivalent between the RTX 50 series and the RTX 40 series.
'Given that the most demanding games for the RTX 5080/5090 cannot be run on the GTX 980 4GB graphics card, and that older games are heavily influenced by the CPU, it is difficult to accurately compare performance in games, so we passed on them this time and tested only GPU computing performance,' said Michael Laravel, who conducted the study.
Related Posts:
in Hardware, Posted by log1p_kr