Geekbench
We are using Geekbench 5 and its Compute Benchmark which runs through a suite of compute-based synthetic tests. This benchmark tests OpenCL, CUDA, and Vulkan. It outputs a total score, and higher is better.
In the graph above OpenCL is being tested between the two video cards. The OpenCL Score is quite a bit higher on the GIGABYTE GeForce RTX 3090 GAMING OC compared to the Radeon RX 6900 XT. The RTX 3090 is 23% faster than the AMD Radeon RX 6900 XT in OpenCL testing according to this benchmark. That’s a significant difference.
In the graph above Vulkan is being tested between the two video cards. The gap widens very wide between the video cards in these Vulkan tests in these benchmarks. The RTX 3090 is 57% faster than the AMD Radeon RX 6900 XT according to this test. That is much greater than OpenCL, and therefore according to these tests, Vulkan has a huge advantage on the RTX 3090 versus the RX 6900 XT.
Only the GeForce RTX 3090 is able to run CUDA. We have shown the CUDA result above. If we compare the GeForce RTX 3090’s CUDA result to its OpenCL result, we find that CUDA is 8% faster than running the tests in OpenCL.
LuxMark
We are using the latest LuxMark 4.0 version, which includes path trace testing. The Hall Bench utilizes Path Tracing plus a Global Illumination cache. The Food scene utilizes brute force Path Tracing.
In the graph above we are testing the Hall Bench scene using OpenCL GPU only. The result is that the GIGABYTE RTX 3090 is much faster than the AMD Radeon RX 6900 XT rendering this scene. The RTX 3090 is 25% faster than the AMD Radeon RX 6900 XT at Path Tracing with Global Illumination cache.
In the graph above we are testing the Food scene using OpenCL GPU only. Once again we find that the GIGABYTE RTX 3090 is on top, it is 21% faster than the AMD Radeon RX 6900 XT rendering this scene with brute force Path Tracing. This scene uses heavy path tracing, so it shows there is a big difference in this type of rendering between the RTX 3090 and RX 6900 XT.
HandBrake
In HandBrake 1.4.0 we are testing the transcoding of a 10 minute long H.264 4K 60FPS video. We are using the Fast 1080p60 preset in HandBrake. For the AMD Radeon RX 6900 XT, we are encoding using the H.264 AMD VCE encoder. For the RTX 3090, we are using the H.264 NVIDIA NVENC encoder.
In the graph above we are looking at the time, it took to render the video, and lower time is better. This result was interesting, as it shows that both video cards took an equal amount of time rendering the video, 8 minutes and 14 seconds on the RTX 3090 and 8 minutes and 13 seconds on the RX 6900 XT. Only a 1-second difference, therefore practically the same. For comparison, using the H.264 CPU encoding (no GPU assistance) takes a total time of 9 minutes and 33 seconds on our Ryzen 7 5800X CPU. Therefore the GPUs are both speeding this process up, but neither one is really faster than the other or provides any advantage in this specific workload.