Trails of Cold Steel 3 for PC will be released in a couple of days, but before that happens, we wanted to provide a bit of insight into the specifics of the PC version. [h1]Returning Favourites[/h1] Just like the ports of its predecessors, ToCS3 on PC adds a number of graphical and convenience features unique to the platform. These returning ToCS PC favourites include: [list] [*]A [b]timed autosave[/b] feature with custom save intervals. [*][b]Configurable turbo modes[/b] with separate factors for battles and other gameplay. [*]Arbitrary framerate support [b]up to 480 FPS[/b]. [*]A large range of Field of View adjustments. [*]Arbitrary resolution support (tested up to 8k), with all secondary rendertargets scaling correctly with the main resolution. [*]Several [b]MSAA [/b]modes, and optional transparency supersampling. [*]High-quality ambient occlusion in the form of [b]HBAO+[/b]. [*]A high-quality postprocessing effects option. [*]Flexible input rebinding for both keyboard/mouse inputs and gamepads, with the correct re-bound prompts showing up in all in-game tutorials. [*]An unlimited drawing range option for high-end systems to eliminate pop-in. [*]Support for utrawide 21:9 aspect ratios. [/list] [img]{STEAM_CLAN_IMAGE}/36616022/c2bc92729c913882114e814740f2d8bdf8a51e27.jpg[/img] [h1]Asset Compression[/h1] One of the new challenges with ToCS3 is the sheer size of its assets. The Trails of Cold Steel games are long and intricate, and the third entry pairs this large amount of content with significantly more detailed assets designed for more modern hardware. With just normal texture compression, the 3D model assets and textures alone (excluding e.g. movies and all sound/voice acting) take up over [b]60 GiB[/b] of space -- and that's only for the Japanese version. While Steam and GoG would reduce the download size a bit, that would still be a rather large chunk of space to dedicate to a single game. As such, early on in the port development we investigated and implemented many options for asset compression. Good compression in this use case needs to fulfill several criteria: (i) it should be [i]lossless[/i], so that we can apply it to arbitrary assets without worry; (ii) it should [i]not increase load times significantly[/i], even on lower-end systems; and of course, (iii) it needs to actually [i]meaningfully decrease the storage space[/i] required. Some options, like LZMA compression, failed early on the load time criteria. Other simple options did not provide a sufficient reduction in size. [img]{STEAM_CLAN_IMAGE}/36616022/edda1e5ff96e6d4a2188d3c483c8eb2f4485ab26.png[/img] Ultimately, the field narrowed to [b]LZ4[/b] and [b]ZSTD[/b] as compression schemes, with the addition of our engine-specific [b]PKA[/b] format which primarily performs hash-based deduplication on game assets. As the image above shows, measuring loading performance correctly for PC games is not trivial. Windows caches all file accesses in RAM as long as any unused space is left, which means that repeated ("[b]hot[/b]") loads will be much faster than initial ("[b]cold[/b]") loads. Also, for cold loads, the HDD/SSD access and read time is a significant factor, which is not the case for hot loads. During gameplay, players will encounter a mixture of hot and cold loads depending on how they traverse maps and even how much available RAM their system has. ZSTD-PKA reduces the asset file size to ~16 GiB from over 60 GiB, but roughly doubles hot load times, and also decreases cold load performance compared to ZSTD without packaging. Based on this result, we changed the sorting of data within the PKA (PKA-S), which achieved improved cold load times without any other drawbacks. Finally, we actually took a step back to LZ4, to trade some compression effectiveness for greatly improved hot load times. Ultimately, LZ4-PKA-S gives us a very acceptable asset package size of [b]~18 GB[/b], with the best cold load times of all options -- a huge [i]improvement in cold load times[/i] over no compression at all! -- and very acceptable hot load performance. We hope you enjoy the work that went into this aspect while playing the game! [h1]Screen-space Anti Aliasing[/h1] For previous ToCS games on PC, various levels of MSAA (with optional transparency supersampling) were sufficient for most use cases. These games were built to run on a PS3, and as such most PCs had ample headroom for such high-quality, but relatively performance-intensive anti-aliasing options. Trails of Cold Steel III was originally built for PS4, which means that its baseline resource requirements are significantly higher. Therefore, on many current PC systems MSAA might not be the ideal trade-off between image quality and performance. [img]{STEAM_CLAN_IMAGE}/36616022/ea4e5a287ca39f1658f923b14e3a2c60f17e67c2.png[/img] As you can see above, at "high" game settings 8xMSAA with transparency supersampling alone can reduce performance by more than half. Even 4xMSAA can have an almost 30% performance impact. 2xMSAA is cheaper, but its effect is rather weak for long and high-contrast aliasing. To provide more options for these use cases, we've also implemented an FXAA option into the game. This implementation is tuned for a minimal impact on overall image quality while providing good anti-aliasing for most obvious geometry edges, with a minor performance impact. [h1]Other Improvements and Final Notes[/h1] We also made several other improvements during the development of ToCS3 for PC, such as [b]higher-quality shadow filtering[/b] with deterministic screen-space dithering. On a more obscure note, I added some logging output for [b]speedrunning[/b] purposes, as I was asked to do so by members of the ToCS speedrunning community. I would like to take this opportunity to remind everyone that ToCS3 was originally built for a platform almost an order of magnitude faster than the target hardware of its two predecessors. [i]As such, please do not expect to get the same performance at the same settings which you were able to achieve in ToCS1 and 2 -- given the improvements in asset detail and quality as well as rendering technology, that is an unrealistic goal[/i]. Finally, I would like to [b]thank our beta testers[/b] for their in-depth testing, reporting and general responsiveness during the long testing period of this game. - Peter "Durante" Thoman, PH3