Synthetic Benchmarks
We are including a new section in our video card reviews we are calling synthetic testing and GPGPU and compute testing. The goal is to represent and show the performance of video cards for other workloads than gaming. Let’s admit it, video cards are used for a lot more than just gaming, and that side should be explored and compared. We will either be adding or subtracting to this section over time as we hone in on the right benchmarks and tests that our readers want to see. Our goal is to represent real-world workload scenarios that people use GPUs for. Feel free to leave a comment in our forums on your suggestions for improving this section and telling us what you want to see.
3DMark
First we will start with 3DMark, which is a popular synthetic benchmark. Our Timespy score is 9885 on the XFX Radeon RX 5700 XT THICC III Ultra, which is 4% higher than the AMD Radeon RX 5700 XT. It is also 6% faster than the RTX 2070 FE.
Another test we are going to run today is the PCI Express Feature Test. This test measures the bandwidth of data across the PCI-Express bus. Since we have PCI-Express 4.0 support on our testing platform, and XFX Radeon RX 5700 XT THICC III Ultra we wanted to see what this looks like. This test shows 25GB/s of bandwidth on the two AMD video cards, and 13GB/sec on the PCIe 3.0 GeForce RTX video cards. This just shows the bandwidth advantage of PCI-E 4.0.
Blender Benchmark
In this next real-world workload we are going to compare the Blender Open Benchmark tool running Blender rendering on the GPU. The two scenes we chose, which take the longest to run, is pavillon_barcelona and victor. We use these same scenes in our CPU testing. We made sure that these are running on the GPU. The time it takes to render is recorded and lower is better.
In our first graph is pavillon_barcelona you will see how much faster the GeForce RTX 2070 SUPER FE is compared to every other card, even the RTX 2070 FE. It completes in 4.47 minutes; our only theory is that the increased compute units are causing this awesome reduction in time to completion. Otherwise, the GeForce RTX 2070 FE completes in 7.24 minutes, which is still slightly faster than the two AMD video cards. The XFX Radeon RX 5700 XT THICC III Ultra does complete slightly faster than the AMD Radeon RX 5700 XT though, thanks to the faster clock speed.
The scene Victor seems to really appreciate CUDA rendering with the GeForce RTX video cards. We actually aren’t entirely sure the AMD GPUs are doing anything hardware wise to improve rendering time. The 40+ minute time is what we usually get when doing CPU rendering. We will have to spend some more time on this benchmark at another time to determine what is going on. At any rate, the RTX 2070 SUPER FE is faster than the RTX 2070 FE. We also see that the XFX Radeon RX 5700 XT THICC III Ultra is a bit faster than the AMD Radeon RX 5700 XT, so maybe it is doing something very slightly.
HandBrake
Lastly we are going to transcode a video using HandBrake and AMD’s VCE or NVIDIA’s NVENC.
Video transcoding time is very close between all the video cards here. The XFX Radeon RX 5700 XT THICC III Ultra matches the time of 9.55 minutes of rendering time of the GeForce RTX 2070 SUPER FE. This is faster than the AMD Radeon RX 5700 XT time of 9.58 minutes and faster than the RTX 2070 FE. For reference, using CPU rendering this same video takes almost 12 minutes to transcode. Therefore, the GPU shaves off almost 2 minutes of time. In a much longer video this time would add up.