Additional information about the headset tested:
- Engine used: 50PPD QLED
- Latest firmware tested: V1.0.17
- OpenXR Runtime: Pimax OpenXR
NOTES
- I had some issues recording VR with OBS. I disabled QV, but the quality was still vastly worse than the Crystal Super itself. Again, I’m quite new to VR.
- To my future self: don’t record VR videos during a heatwave without cranking up the AC ever again…
A few weeks ago, Pimax sent me a Crystal Super. I will share my opinions in a later video, as today’s discussion is about a series of benchmarks I ran to understand how to get a solid experience.
Since I’m very much a new VR user with zero background knowledge, I did what I usually do in similar situations when playing DCS: I threw numbers at the problem.
First, let’s add some context or, better put, the hardware I used, so that you can have a better idea of the testing setup.
The Ol’ Reliable
I updated my PC in 2019, moving from a 3770k with 16GB of RAM to an AMD 3900x with 64GB of RAM. In the past seven years, I updated my CPU to a 5800x3D and the GPU to a 6900xt first and then to a 5080. The RAM is still the ancient 64GB 3200MHz CL16. Ergo, my hardware is not new at all, and I shelved my plans to upgrade the triumvirate mobo-CPU-RAM until the AI bubble hopefully bursts.
It is worth noting that I did not overclock my 5800x3D, but I tuned the PBO curves to minimise temperatures and drastically increase its performance.
Since the GPU is fairly modern, I fully expected my CPU to be the performance bottleneck. This is the point I attacked.
Scenarios and Measures
The scenario I put together is quite tough DCS-wise. As we know, the game weighs heavily on the CPU, and there are not many ways to work around that. Since the number of frames per second floats rapidly depending on the amount of AI and objects in the air, the scenario is meant to replicate both situations. It starts with many objects, but it quickly escalates as missiles start flying. Ergo, values such as the lowest 1% are fundamental to ensure that the experience is smooth. Otherwise, the VR experience becomes almost nauseating.
On the other side of the coin, numbers such as the average FPS are a bit pointless, as the moments of calm skyrocket them. It’s basically like using only the GDP to describe a country’s economy.
To ensure reliability, I saved a track circa 45 seconds long, and I used that across the tests. Since DCS is DCS and tracks can change, every data entry has been replicated multiple times, and eventually, I collected circa 700 datapoints. Moreover, to minimise the effects of eye-tracking and quadview, I focused my eyes on the cross at the centre of the DDD.
Lastly, the timing of the capture was automated to further enhance data reliability.
Parameters Collected
I used a program called CapFrameX to capture the following values:
- P95
- Average
- P5
- P1
- 1% Low Average
- P0.2
- P0.1
- 0.1% Low Average
The various “P”s are the Percentiles.
On top of that, CapFrameX captures additional data related to the frame times. From my understanding, CapFrameX monitors when the frame times spike above the average by a certain factor, possibly because the game is resource-starved.
- Smooth
- Low FPS
- Stuttering
- Smooth %
- Low FPS %
- Stuttering %
Results
The results are split into two parts, as a Pimax firmware update was released before I completed the tests. Those numbers will be discussed later.
The baseline test is conducted on my Ultrawide 34″ monitor, set to 3440×1440 with almost every possible graphical option activated and no frame generation.
Let’s have a look at the most interesting series.
Series I: Monitor
As you can see, my ancient PC handles DCS very well, easily averaging well above my monitor’s refresh rate. Even the lowest FPS recorded averaged around 40, which is more than playable.
Series II
Series III: Lab Refresh Rate
This set uses an experimental feature: the “Lab” implementation of the locked 72Hz refresh rate. The performance against the previous series improved, but several graphical artefacts hindered usability. For example, the Detail Data Display of the Tomcat becomes unreadable.
Series V and VI: Reducing DCS Quality
These series move from the previous preset called Custom 3 to Custom 2.
What I did not expect is the lack of any tangible impact. A possible explanation is that the GPU has spare computing power compared to my ancient 5800x3D, and a graphical downgrade is not particularly effective.
Series VII: CPU Cores
For this benchmark, I used an old practice that used to work quite well back in the Quad Core days, where DCS used de facto only two threads. With the x3D instead, no tangible effect is visible. In fact, it appears that the maximum FPS are actually lower. Perhaps the now-Multithreading DCS makes better use and benefits more from having multiple cores available.
Series VIII: Graphics settings downgrade
Since Series IV seems to be a good baseline, I decided to run a last test. For the occasion, I disabled the mirrors and reduced the image and QuadView quality. The performance data collected is actually worse than the previous series.
Old Firmware Data
What I did not expect is the lack of any tangible impact. A possible explanation is that the GPU has spare computing power compared to my ancient 5800x3D, and a graphical downgrade is not particularly effective.
Let’s have a brief look at the old data, which should be taken with a grain of salt, since software updates can have an impact on the results. Moreover, the update modified, merged, or removed some of the options previously available, making comparisons across the dataset impossible.
The modus operandi is the same, ergo multiple takes of the same sets to minimise unwanted fluctuations.
The most interesting comparisons here are the II against the III series, and the IV against the V.
Series II vs III
Series IV vs V
Once again, these benchmarks represent particularly resource-intensive scenarios, and the goal is finding and understanding what can help us players to make those situations playable.
Observations
Let’s have a look at some of the points raised by these numbers, keeping in mind that as I progress through the tests and I prepare the review of the Pimax Crystal Super, I will probably find better settings and options to improve the quality versus performance ratio. If that’s the case, I will update this discussion and the related article on FlyAndWire.
- Effects of Graphics Settings
In some cases, lower graphics settings actually reduce the performance. I wonder if this is caused by the engine re-prioritising the load from the GPU to the CPU. Something similar happens in Star Citizen as well. So, although this usually works, downgrading the GPU load in an already CPU-bound game becomes detrimental. If the issue is the lack of GPU memory, for example, then it might help instead. - CPU and Process Priority.
Settings such as process priority or core affinity did not change the situation in any appreciable way, at least in the tests I conducted on my PC. - Upgrade to 5080. Worth it?
The 5080 is more than capable of sustaining a high-quality headset such as the Pimax Crystal Super. If you plan to upgrade from an older series, it might be worth the cost. The 5090 is definitely superior, but I bought my Palit 5080 for slightly more than 1 grand, whereas the former was almost four times as expensive, between 3700£ and over 4000£, and prices are still going up. Personally, the cost versus performance ratio makes little sense, but you do you. - 5800x3D: Old but Good-ish
CPU-wise, DDR5-based options are available and capable, but the AI bubble has dragged the prices to levels unheard of before. In fact, the 5800x3D has apparently seen a resurgence, and it is still an adequate option for general gaming.
The combination of 5800x3D and 5080 is capable of sustaining DCS through the fantastic quality of the Pimax Crystal Super in all sans from the most complex and taxing scenarios.
Leave comments and feedback in the Discussion Forum thread.
You must be logged in to post a comment.