Thank you Traverse Research for sending us the Professional version of this benchmark to review!

Since the beginning of computing, there has always been a desire to quantify performance. How do I know if one CPU is faster than another model, other than to test them both? The same applies to the graphics processor as well. How do we test which ones are faster, or leads to a better game experience? While the best test is of course in the applications you use or games you play, applications more focused on testing can still provide a lot of value, especially if they drill down and test individual aspects of a system's performance. Evolve Benchmark is perhaps one of the newest entries in a long history of PC gaming benchmarks.

Like other benchmarks throughout the years, Evolve Benchmark is based around a visually stunning graphical demo, rendered with the latest technologies with cinematic flair. In this case, we get to see a safari where our friend in a large-brimmed hat walks through a Jurassic Park-style tropical paradise and gets to see dinosaurs up close. It really does look great, with ray-traced lighting and shadows. According to their technical documentation, it even uses GPU-based object state tracking. (GPU stands for Graphical Processing Unit.) It should probably be noted that Evolve Benchmark is above all a GPU tester. While the CPU does certainly play a part, it's more in how quickly it can submit commands to the GPU, rather than playing a part in physics modeling like most other benchmarks do.

The demo uses the latest in rendering techniques, and offers five different approaches to rendering the same scene. All benchmarks require ray tracing-capable hardware; without that, no benchmark will start. I tried launching it using the last GPU without ray tracing hardware, the GTX 1660, without success, despite NVIDIA offering basic software emulation on that model. On the other hand, my AMD Ryzen 6800U laptop with Radeon integrated graphics has basic hardware ray tracing support, and it can execute four out of five of the included tests. My other machines have NVIDIA RTX cards, and they can run them all.

The first two tests are called Ray Tracing benchmarks. The first one is called Inline, and the other is called Pipeline. These are both two different approaches to solving the same problem. The folks at Evolve wrote an excellent blog post explaining the difference here.

To summarize, the main difference is that Inline is where ray-tracing shaders are added to an existing rasterization workflow, whereas with Pipeline, multiple independent shaders are used to perform ray tracing operations. Both demos look fairly similar, though the Inline approach probably has the most stable image as it relies on tried-and-true rasterization for as much as possible. Ray Tracing Inline is also included in the free demo version of Evolve Benchmark, which I would encourage those interested to check out, though you do need to create an Evolve account for results submission, which some may object to. It might help in the future if Steam users had automatic account creation based on the Steam login info, but be aware it's not like that.

Interestingly, according to the above blog post, Inline and Pipeline perform fairly similarly on NVIDIA GPUs, while Pipeline performs much better than Inline on AMD Radeon. At this time all of my high-end GPUs are NVIDIA, so I can't really confirm those results with my experience. I was able to run both Ray Tracing benchmarks on my gaming handheld, the GPD Win Max 2 (the 6800U mentioned earlier), though the scores are quite low. Well, except for the Energy Consumption score, but more on that later.

The other major renderer type is the Path Tracer group of benchmarks. The interesting thing here is that they don't seem to use denoisers, or at least they don't fill in the gaps - you see each traced path and only the traced paths, so visually it looks a bit like there's a poor signal on an old analog TV (for those of up old enough to remember that). The gaps in between the rays show up as small white dots. I believe in most real-world use cases, AI denoisers are used.

When it comes to Path Tracer benchmarks, there are three different kinds. They are Inline, Pipeline, and WorkGraph. Unfortunately, there is less detail on their website defining those differences. Path tracing is in many ways the 'endgame' of ray tracing; instead of using ray tracing to enhance a rasterization-created scene, it instead uses rays to simulate all lighting. This article on the history of PC graphics rendering does a great job explaining the difference.

As explained in that article, the Evolve Path Tracer benchmarks are brutal on performance. Where the Ray Tracer benchmarks might offer eighty frames per second on an RTX 5080, you're lucky to see twenty on the same Path Tracer benchmarks. And WorkGraphs are the most strenuous of all. My Radeon integrated graphics laptop can run the Inline and Pipeline benchmarks - well at least they start. I found my system hard locked after a couple of frames, each of which took close to a minute... Regardless, WorkGraph is not available to run at all on that RDNA 2-based integrated GPU. However, it does run on higher end AMD GPUs like the 7900XTX (if the leaderboards are any indication) which is RDNA 3, so it could just be that my little laptop is underpowered for that workload. As a side note, I noticed while preparing for this review that there seems to be a bug where NVIDIA GPUs show as 'Not verified' when running the WorkGraph test, so while the test completes and you get results, they are not showing up on the leaderboard. Update: I reported this bug and it has been fixed.

Each benchmark is broken down into several of seven different score types, though I don't think any single benchmark type uses more than six. The first type is Acceleration Structure. It's used by all benchmark types. This score isn't explained all that well, other than to say that it represents the speed at which the GPU and driver builds the dependencies needed for ray tracing operations.

The second score type is Rasterization. This is used in the two Ray Tracer benchmarks, Inline and Pipeline. This is the classic rendering type, where triangles are converted to pixels. Modern GPUs are very good at this, as it's how games have been rendered for the last several decades.

The third score type is Energy Consumption. This one is interesting because it seems to be some form of ratio of performance to power usage. Incidentally, I happen to hold a world record (for now) on this test, as my aforementioned GPD Win Max 2's 6800U managed to top the chart on the Ray Tracing Pipeline test when forced down to 15W. I'm sure I could do even better with undervolting and such, but I'll take a win where I can get it. Regardless, it's interesting that this score has a wide range of different pieces of hardware represented on the top 25; while the performance ones are dominated by RTX 5090s, in the Energy Consumption score, there are desktop, laptop, and integrated graphics all in the top 25. It is unclear if CPU power usage is factored into the Energy Consumption score, or if it's entirely the GPU that matters.

The fourth score is a key metric, Ray Tracing. This score is present on four out of five benchmarks; it's not present on WorkGraph. As you might expect, it measures what it says on the tin: it measures the performance when utilizing the hardware-accelerated ray tracing on modern GPUs. According to their technical documentation, it includes six different forms of tracing, though these are not broken down in the results.

The fifth score is Compute. This score is also present in four out of five benchmarks, as it's not present on WorkGraph. This category covers a wide range of effects, and according to Evolve, it covers one of the largest sets of tasks. It includes rendering, animations, physics, world updates, and more. It also covers post-processing like anti-aliasing. The scores are interesting here; on one of the PathTracing leaderboards, you can see that Radeon GPUs are more competitive, especially on the Pipeline scores.

The sixth score is Driver, which is one of the few scores that applies to all benchmark runs. This one is the closest we have to a CPU test, but since the GPU driver has a large impact also (obviously), AMD Radeon cards do quite well on this test. This test measures how quickly GPU command operations are issued from the CPU side. According to the technical documentation, it currently measures primarily driver calls from the Acceleration Structures side of the benchmark.

The final score is the WorkGraph score, which is only used on the similarly-named benchmark run. In a WorkGraph run, all work that would otherwise be covered by Raytracing, Rasterization, and Compute work is combined in this score.

It would be kind of nice if Evolve Benchmark offered some kind of overall, aggregate, or weighted average score, as many gamers like having one big number to refer to when comparing systems. The scores as is are no doubt quite useful, as it can help us tweak, or spend time determining what unexpected bottlenecks there might be, and as each score relates to something different, it can be helpful having those extra data points.

Regardless, benchmarks like this really do help us a lot even as is. For example, the Path Tracer benchmarks helped me realize that my gaming laptop's GPU undervolt and memory overclock was not as stable as I thought. It's not because it crashed, but because I was seeing small graphical artifacts, or clumping of noise dots, that I didn't see on my other systems. That's one of the reason we love benchmarks so much - to help us determine if our systems are stable!

Generally, the benchmark has been easy to use, though I have noticed a few bugs since playing with it. I was grateful to be involved in reporting and fixing a leaderboard submission bug; while it's not 100% gone, it only happens occasionally now rather than quite often. I also noticed the bug above on WorkGraph submissions; I reported that, and it was fixed in a few days.

While I was writing this review, version 14.0 was released. That version added some great changes like saving benchmark runs to disk by default, instead of only submitting them online. They also added more features to the Advanced version.

Speaking of, the base version of Evolve Benchmark as sold on Steam is probably enough for most users who just want to see how their systems score. There is an Advanced version DLC that adds offline features and batch/command line running. I was given the Professional version for this review, which allows for commercial use as well. There are also data analysis features available for Professional users, as there is a GitHub repository with scripts and such. They are written in Python and are open source.

Evolve Benchmark is quite a promising new entry into the benchmarking space. It takes advantage of some of the most advanced features of modern GPUs, and breaks down results in ways that few (if any) others do. As a long-time PC benchmarking nut, I'm excited to see where Evolve takes this in the future. I would definitely, at minimum, recommend checking out the free demo, and if you like what you see, picking up the rest. I also can't wait for the promised Linux support to arrive, as I dual boot all of my PCs. Highly Recommended if you are a PC tweaker like I am!