3DMark Port Royal viewed – ComputerBase

– UL has expanded the 3DMark to include the “Port Royal” scenario. The new test uses real-time ray tracing according to Microsoft DXR to display reflections and shadows. BitcoinMinersHashrate looked at the benchmark and tested all four available GeForce RTX graphics cards.

The 3DMark Port Royal

At the launch of the GeForce RTX graphics cards in September, UL, owner of Futuremark, provided an early and abbreviated version of the 3D Royal test sequence Port Royal. She uses ray tracing on a GeForce RTX and wants to beautify reflections and shadows. At CES 2019, UL Port Royal completed and published. After Battlefield V, 3DMark is the second software that can use the new capabilities of the Nvidia graphics cards.

If you only have the free version of 3DMark (download), you won't be able to try Port Royal yourself. The test is only available to owners of the paid Advanced or Professional version of the benchmark software. All others can watch the demo sequence from Port Royal (recorded on a GeForce RTX 2080 Ti) in the video on BitcoinMinersHashrate.

In the following, the editors take a quick look at the test sequence, which as a pure benchmark as before has no direct informative value for the performance of a computer in games. This becomes clear again in this case.

There's so much ray tracing in it

UL states that it uses its own and further developed engine for the test sequence, which was developed in collaboration with AMD, Intel and Nvidia, among others. Port Royal uses Microsoft's DXR interface for real-time ray tracing and thus DirectX 12 automatically. According to UL, DirectX 12 with feature level 12.1 is used without the developers having to go into detail.

In order to reproduce the benchmark, the graphics card must have at least a DXR backend. At the moment it can only run on a GeForce RTX (Turing) or Titan V (Volta). Nvidia's Pascal graphics cards also have the backend (although ray tracing is then calculated using the normal shader units), the test currently only spits out an error message on the graphics cards.

Ray tracing for reflections and shadows

The actual benchmark scene uses real-time ray tracing like Battlefield V for reflections (reflections with correct perspective) and for shadows (pixel-perfect hard shadows). Both are effects that are not possible or only possible with the classic rasterizer tricks. If you want, you can also switch off both effects individually. Then a classic shadow map is used for the shadows and reflections are only faked.

However, in the case of reflections without ray tracing, UL states that the traditional integration is not at the level that it would be in a classic game. As a result, Port Royal offers only very limited meaningfulness in terms of performance and quality comparisons of ray tracing with rasterization. This is also the reason why the article explicitly does without screenshot and video comparisons.

Port Royal also uses more ray tracing, but not in the benchmark but in the demo sequence, which is optically a bit nicer but does not allow performance measurement. The demo uses so-called “ray traced photon mapping” for the “Global Illumination”, ie “Global Illumination” is rendered by ray tracing instead of a rasterizer effect. Why the effect only found its way into the demo remains a mystery. Also exclusively implemented in the demo is a fluid simulation in a scene, which, however, is not calculated using ray tracing.

DLSS will be added later

Port Royal should receive further features in the future. At least Nvidia has already announced the DLSS AI anti-aliasing for the benchmark. When it will be included in the test is still unclear.

Test system and test results

An Intel Core i7-8700K overclocked to 4.7 GHz serves as the processor for the benchmarks. Two 16-gigabyte DDR4 modules (dual-rank) are operated with DDR4-3200 at the timings 16-16-16-38-1T. Windows 10 version 1809 is used as the operating system. The GeForce 417.54 and thus the launch driver of the GeForce RTX 2060 is used as the graphics card driver. All graphics cards work with the standard clock rates.

Port Royal is always calculated in the resolution 2,560 × 1,440 regardless of the monitor. The quality of the reflections is set to a medium level of detail in the benchmark by default. Only the demo sequence uses high details. The test shows the results both as points and as frames per second. Only the FPS results are used in the test.

Benchmark with RTX 2060, RTX 2070, RTX 2080 and RTX 2080 Ti

Only the GeForce RTX 2080 Ti manages a fairly smooth reproduction of slightly more than 35 FPS in Port Royal. All other Turing graphics cards are a good deal or significantly below. The GeForce RTX 2080 delivers 25 percent fewer images per second than the flagship. The GeForce RTX 2070 is another 22 percent slower and the GeForce RTX 2060 another 19 percent. The distances are larger than on average in classic games.

At the upper performance limit, it is conceivable that the theoretically clearly superior ray tracing performance of the GeForce RTX 2080 Ti compared to the GeForce RTX 2080 (72 to 48 RT cores) will do more in 3DMark than the higher shader computing power in normal games. With the GeForce RTX 2060, the TU106 GPU (30 to 36 RT cores), which is only 17 percent cropped and has a higher clock speed than the GeForce RTX 2070, does not explain the distance of 23 percent.

But there is another relevant difference: the memory reduced to 6 GB on the GeForce RTX 2060. VRAM measurements can confirm the assumption, because the memory of the new Turing entry-level model is shoveled in the benchmark. That alone is no confirmation of the lack of memory, but in connection with the benchmark result, the assumption is obvious.

Comparative tests on the RTX 2080 Ti

If results are compared with and without ray tracing, it becomes clear that ray tracing costs surprisingly little speed in Port Royal. While in Battlefield V the performance is almost halved by RT reflections alone, in 3DMark it is just 15 percent loss in RT reflections and RT shadows on a GeForce RTX 2080 Ti FE.

At first glance, there is no reason for this. In Port Royal, the entire spaceport, including the spaceship, was polished to a high gloss, so there are numerous reflections there too. In connection with UL's statements that the classic reflection and shadow techniques are not really well integrated in Port Royal, the reason is likely to be found in the poor implementation of the rasterizer variants. Raytracing is most likely not going well, but rasterization is going particularly badly.

A question of performance: how often are rays reflected?

A further contributing factor to the phenomenon could be the fact that UL apparently only allows a shot to be reflected once – at least the menu item "Max reflection Sample count" indicates this. This is set to "1" in the standard settings, but can be increased to "4".

Higher values ​​improve the quality of ray tracing, but also cost a lot of performance on a GeForce RTX 2080 Ti. A maximum of two reflections already reduce the speed by 14 percent. It is 25 percent for three reflections and 33 percent for four reflections. That would amount to 43 percent less power than without ray tracing.

conclusion

With Port Royal, 3DMark will be the second software in early 2019 that will use real-time ray tracing via Nvidia Turing and Volta via Microsoft DXR. If you want to see the test on a GeForce RTX graphics card, you have to have the paid version of 3DMark – Port Royal is not available in the free basic version.

Since UL itself admits that the rasterizer alternatives to the raytracing reflections and shadows used are not at a high level in terms of performance and quality, performance and quality comparisons between the two modes with Port Royal are, however, not meaningful.

If GeForce RTX graphics cards are compared with each other in Port Royal, the image from Battlefield V is more likely to be found again, even if the GeForce RTX 2080 Ti can stand out better from the RTX 2080 and the GeForce RTX 2060 falls back a little further.

Port Royal cannot shake off an important finding: the performance in their games is relevant for players, not that in synthetic benchmarks.

This article was interesting, helpful, or both? The editors appreciate any support in the form of deactivated ad blockers or a subscription to BitcoinMinersHashrate. More on the topic of ads on BitcoinMinersHashrate.