Performance delta from GCN to RDNA

In my last post, I documented upgrading my desktop’s graphics card (GPU) from an older GCN (graphics core next) architecture to a modern, RDNA-based GPU: an AMD FX 5700 XT. The more interesting part of the upgrade is that I replaced the blower-fan air cooler on the 5700 XT with an all-in-one (AIO) water cooler. The water cooler was initially designed for CPU cooling, but with a 3D printed bracket, I was able to mount it to the GPU’s printed circuit board (PCB). The radiator on this AIO is actually larger than the original Corsair H100 I am still using for my CPU, and is very effective for cooling the new GPU.

To get a sense of both the performance improvement due to upgrading the GPU, and the improvement in cooling from installing the AIO on the graphics card, I ran Unigine Superposition. Superposition is a graphics-intensive gaming benchmark that stressed both my old and new cards to their limits.

I ran two configurations: the medium 1080p preset, and a custom 1440p setting to max out the resolution of my monitor. Because I had to run the same benchmark on each GPU, the texture settings I ran were constrained by the amount of VRAM available on the older HD 7970 GPU. Therefore for the 1440p custom test, I set extreme shaders but only medium textures. Between tests, I allowed the GPU under test to return to a steady-state temperature.

Results on 1080p Medium Preset for Unigine Superposition

This first table gives the results for the 1080p medium preset. The table shows the performance in frames per second (FPS) as well as temperatures and recorded GPU utilization during the benchmark run. The fourth column shows the score given by the Superposition benchmark. Clearly, the 5700 XT performs much better than the 7-year-old 7970. The minimum FPS was improved by between 1.4x and 1.6x, depending on the cooling configuration of the 5700 XT. The maximum and average FPS improvements were at least 1.9x, and again this gets even better with GPU on water cooling. The effect of the AIO is evident in the maximum temperatures reached: on water, the 5700 XT reaches a maximum temperature near that of the steady state, unloaded temperature on the original air cooler! The 5700 XT scores higher on water because it is able to maintain higher clocks and reach higher overall utilization.

Card FPS Min FPS Max FPS Avg Score Max Temp (c) Max GPU Utilization
HD 7970 37.40 58.20 45.64 6102 74 98%
RX 5700 XT 53.35 124.69 86.88 11615 72 81%
RX 5700 XT, Water 62.77 133.30 92.42 12356 54 95%

Results on 1440p Custom test for Unigine Superposition

The second table shows the performance, thermal, and utilization metrics under the custom 1440p settings. Both GPUs have a much harder time rendering with extreme shaders at this resolution; clearly even the new 5700x XT is not up to no-holds-barred 1440p rendering. This is pretty evident since the GPUs all reach 99% utilization (reported by MSI Afterburner), where in the 1080p test the 5700 XT had more idle time. But, it is more than 2x faster than the older 7970, and with the AIO, runs much cooler and quieter to boot!

Card FPS Min FPS Max FPS Avg Score Max Temp (c) Max GPU Utilization
HD 7970 7.08 10.02 8.46 1131 78 99%
RX 5700 XT 18.07 24.54 21.32 2850 81 99%
RX 5700 XT, Water 18.38 21.87 25.28 2932 57 99%

Closing Remarks

While my chosen method of watercooling the 5700 XT is unorthodox and requires some DIY know-how and tools, the results show modest improvements in performance and greater improvements in peak GPU core thermals across both tests. All in all, it was fun to put this upgrade together and the reductions in thermals and fan noise are worth it to me.

My remaining concern is that the VRAMs are no longer directly cooled by a strip of metal from the main heatsink. However, I did affix small heatsinks to each VRAM which had thus far done a decent job of controlling memory temperatures.