Project CARS: Über 40% höhere Framerate bei Project CARS - Entwickler nimmt Stellung zu den Veränderungen und nennt Details

In der vergangenen Woche zeigten die Kollegen von Eurogamer ein Vergleichsvideo zu Project CARS, dass die Xbox One- und PS-Fassung des kommenden Racers zeigte und durch starke Schwankungen in der Bildrate auf sich aufmerksam machte. Kurze Zeit nach der Veröffentlichung wurde das Video zurückgezogen. Der Grund: die Redakteure des Onlinemagazins verwendete ältere, nicht zur technischen Analyse freigegebene Versionen des Spiels, die laut Aussagen des Entwicklers nicht das fertige Produkt repräsentieren. Siehe auch: Project CARS : Vergleichsvideo von Project CARS basiert auf alten Builds - Eurogamer zieht Artikel zurück



Nach der Auslieferung der gezeigten Konsolenfassungen kam es bei Project CARS zu zahlreichen Optimierungen und Anpassungen, die zu einer Performanceverbesserung von bis zu 41% führten. Darüber hinaus wurde auch die anisotrope Filterung (AF) von 4x auf 8x angehoben, was zu einer wesentlich schärferen Darstellung der Texturen führt.

Auch die Art der Reflexionsberechnung wurde komplett abgeändert. In der zu sehenden Build wurde noch eine sehr rechenintensive Cube-Map-Variante verwendet, in neuen Builds schwenkte man auf Screen Space Reflection um, was ebenfalls eine massiven Performanceverbesserung führte.

Auch was die Kantenglättung der Xbox One-Version angeht, wies der Artikel von Eurogamer Fehler auf. Anstatt der im Bericht genannten Post Processing FXAA-Kantenglättung kommt ein angepasstes Multisampling Antialiasing (MSAA) zum Einsatz. Konkret handelt es sich hier um AMD's EQAA (Enhanced Quality Anti Aliasing) mit 8 Fragmenten und 4 Samples, was dem 4xMSAA auf dem PC entspricht. Darüber hinaus greift Project CARS für Xbox One jetzt auch auf den siebten CPU-Kern zu, der kürzlich von Microsoft für alle Spielentwickler zu 50%igen Nutzung freigegeben wurde.


The preview was based on an older build of the game and several of its observations are unfortunately either wrong or due to bugs or items which have subsequently been addressed by the development team during the finaling process. Notably:

- The level of Anisotropic filtering was increased from 4x to 8x on both consoles, significantly improving the general image quality and sharpness of the road.
- Both consoles are set to use motion-blur at the equivalent of the PC medium setting – the differences seen on PS4 were due to bugs which were transient with on-going optimisation and not because PS4 uses object based motion-blur.
- The shadow differences highlighted were largely down to the various slope-scale DX11 issues we had at this time - many WMD users had reported these on PC builds and we were able to fix these bugs to more closely align the shadow rendering across platforms.
- The build reviewed was also lacking some significant "render bridge" camera work, which significantly improved frame pacing and the general feeling of smoothness especially when cornering.

In the article it is mentioned that Project Cars on Xbox One is using post-process anti-aliasing which is equivalent to the PC's higher FXAA settings. This is incorrect, the Xbox One version uses MSAA, or more precisely AMD's EQAA (Enhanced Quality Anti Aliasing) with 8 fragments and 4 samples, which is equivalent to MSAA 4x on PC.

- In the older build that was analysed it was possible to become CPU bound with very high numbers of AI, maxing out all 6 cores on XBox One. However, Microsoft had recently opened up 50% of the 7th core to developers : in later builds the development team was able to offload work such as the audio mixing, engine sound synthesis and detailed grass generation onto this core, fixing this problem of becoming CPU limited.
- Tracks that had water elements (sea/lakes etc) in the analysed build used an expensive cube-map method for reflections leading to inconsistent frame rate. In late March this was addressed with us switching over to use a screen space reflections method - A WMD community member did some comparison benchmarks on PC when this new technique was introduced:

"B967/B968

Azure Coast 62.5 fps/82 fps +31%
Azure Circuit 61.5 fps/ 73 fps +19%
Laguna Seca 55 fps/76 fps +38%
Sakitto GP 61.5 fps/86.5 fps +41%
Nürburgring GP 57.5 fps/76.5 fps +33%

Looks identical visually, big improvement."

The relative scaling there applied equally to the console versions.

If you incorporate our above corrections to the article, these points and some more general optimisations that have subsequently been worked on, it's easy to see why we might feel this review isn't very representative of what the development team has achieved.