3DMark Time Spy shows 2.8x average GPU performance increase over 8 years

UL released their newest non-raytraced benchmark yesterday, now available for all 3DMark users, even those on the Basic version (now called “Demo”). This benchmark is a successor to one of the most popular GPU benchmarks, Time Spy. Time Spy was released in 2016 and has amassed 48 million results over its history— that’s a staggering number of tests.

The team behind the benchmark compiled those 48 million results into a single graph, showing how GPU performance has changed since 2016. The data reveals a performance gain of around 2.8 times over the period (note: in the PR, 3DMark says it quadrupled). However, this doesn’t tell the full story.

Similar to the Steel Nomad, the 3DMark Time Spy is a non-raytraced benchmark. It uses modern DirectX 12 (actually UL’s first for this API), but it doesn’t take advantage of newer DX12 Ultimate technologies like mesh shaders, variable rate shading, or sampler feedback. It’s no secret that GPU vendors have been focusing heavily on these technologies, as well as ray tracing acceleration.

3DMark TimeSpy 2016-2023, Source: UL/VideoCardz

The TimeSpy performance graph shows a massive increase in average GPU scores in 2020 and 2022 when both NVIDIA and AMD launched their new GPU architectures. There was a clear pause after the RTX 40 and RX 7000 series came out, even a small dip, possibly contributed by the high price of the new GPUs at the time. The trend has now entered a new flat line as users are no longer using Time Spy as often.

3DMark was updated with Port Royal and later Speed Way, which are ray-traced benchmarks and are now preferred by users with RTX/RX graphics cards. Given the fast pace of technological development in the gaming industry, particularly with upscaling and frame generation technologies, it’s likely that future 3D benchmarks may not span as many years as current ones.

Source: 3DMark PR