devtrackers.gg
almost 6 years
ago -
Hunt: Showdown
-
Direct link
Transcript (by Youtube)
| 2s | [Music] |
|---|---|
| 4s | we spoke to Chris auti |
| 6s | level design director and Sebastian |
| 9s | Laurent technical director about |
| 11s | performance update 24.2 when we released |
| 16s | 4.0 there was a number of things that |
| 18s | have been changed in the game there were |
| 19s | a number of things in the background |
| 20s | that had happened and this course some |
| 22s | performance drops for a lot of people |
| 24s | the community were very vocal about it |
| 26s | as right they should have been and this |
| 28s | was one of the things that we wanted to |
| 30s | make sure we fixed for the performance |
| 32s | update and not only fixed it and |
| 34s | improved it but added on top of it made |
| 36s | the stores go down lower people's CPU |
| 39s | usage and other things so players will |
| 43s | get some improvements to the threading |
| 45s | because that's one of the main causes |
| 47s | for this tools that we've identified |
| 49s | overall we are going to reduce the |
| 52s | pressure on your multiple CPU cores you |
| 55s | should also see a slight decrease in the |
| 56s | memory usage we see maybe 15% memory |
| 59s | reduction there people in lower end |
| 61s | hardware should also see a frame rate |
| 62s | increase as well and we always said that |
| 67s | it's better for a game to have a smooth |
| 69s | 30 than it is to have a spiky 60 when it |
| 72s | comes to performance like you're trying |
| 73s | to aim at someone you're trying to take |
| 75s | someone's head off with it with a |
| 76s | Winchester and the second you got to |
| 78s | pull the trigger there's a few frame |
| 79s | drop and you missed the shot is now |
| 81s | that's a devastating thing to a player |
| 82s | and it's something that we obviously |
| 83s | want to avoid what we found was in the |
| 86s | background there was a process that was |
| 88s | happening in run time that should not |
| 89s | have been happening in run time it was |
| 91s | something that was disabled you know at |
| 93s | some point in development of hunt but it |
| 95s | was only disabled in our internal bills |
| 98s | in the release versions of the game this |
| 99s | feature was still actively running in |
| 102s | the background and when we turn that off |
| 103s | we noticed that there was you know a big |
| 105s | improvement in the smoothness of the |
| 106s | game so that is something that will |
| 108s | benefit literally every player out there |
| 110s | it will be you know a much smoother |
| 112s | experience because of that being |
| 113s | disabled now so yeah this background |
| 115s | process that we figured out was also |
| 118s | especially memory hungry |
| 119s | so for machines that have lower amounts |
| 122s | of RAM like typically eight gigabytes or |
| 126s | so it was especially expensive to have |
| 130s | it running additionally to this we also |
| 133s | our physics thread was now having a |
| 136s | higher workload than our main thread so |
| 138s | we had to synchronize properly the main |
| 141s | thread and the physics thread to make |
| 143s | sure that you don't render multiple |
| 144s | frames with the same physics result |
| 146s | because even if your frame rate would |
| 148s | stay high in that kind of context you |
| 150s | would still end up with a very stuttery |
| 152s | feeling typical users have about four |
| 155s | cores but we have many more threads that |
| 158s | need to run that need CPU time to |
| 160s | execute so like typically we have |
| 162s | multiple threads running for physics |
| 164s | like physics workers we have a job |
| 167s | system that is splitting all the workers |
| 169s | of the gameplay logic and of some other |
| 171s | systems onto multiple threads we have a |
| 173s | main thread we have a render thread we |
| 174s | have multiple earlier threads and all of |
| 176s | this like fights to get some cpu time so |
| 180s | even if we just reduce the workload on |
| 184s | the physics thread it's not just |
| 185s | affecting physics it's also giving some |
| 188s | breathing room for every other process |
| 190s | to run and get CPU time in addition to |
| 193s | that we found a few places where we were |
| 195s | loading objects in runtime which is a |
| 197s | big no-no we don't want to be doing that |
| 199s | we found an issue specifically in |
| 202s | Cemetery normally when you're in these |
| 203s | underground sections you don't render a |
| 205s | lot of the stuff outside and around you |
| 207s | we found a bug in that where it could be |
| 209s | rendering a lot more than it should do |
| 210s | so we fix that so players might see |
| 212s | perhaps 10 to 20 frame improvements in |
| 215s | the underground sectors especially |
| 216s | fighting bosses which we think will |
| 218s | obviously help people a lot when they're |
| 219s | doing that users might see that in the |
| 224s | debug information that they can activate |
| 227s | that the CPU starts waiting for the GPU |
| 230s | if you ever see this something that you |
| 233s | might want to take a specific look into |
| 235s | is resolution scaling because the |
| 238s | biggest cost we have on the GPU is |
| 240s | typically the the rendering side of it |
| 243s | which is directly linked to the number |
| 245s | of pixels that need to be processed so |
| 247s | the resolution scaling is a good way to |
| 249s | keep your you ice crisp and still lower |
| 251s | a bit the load that you have on the GPU |
| 253s | side also please make sure that if you |
| 256s | can afford it you're in exclusive |
| 259s | fullscreen because that might also help |
| 260s | for various reasons like vos allows |
| 264s | hunter to have more |
| 265s | resources dedicated to it when you're in |
| 268s | that mode what we haven't done is |
| 272s | combined everything that we were going |
| 273s | to do for 5.0 and 4.2 into one patch |
| 275s | reason being is then you wouldn't get a |
| 278s | patch for quite a long time so we wanted |
| 280s | to make sure that we addressed it |
| 282s | current issues and smoothed out the |
| 284s | experience for people as much as we |
| 285s | could as soon as possible because |
| 287s | obviously you know players have been |
| 289s | waiting several weeks now for |
| 290s | performance update so we wanted to make |
| 291s | sure that they have something now that |
| 293s | they could see and they could get some |
| 294s | effects for to offset the issues that we |
| 296s | saw on 4.0 so that was a huge priority |
| 298s | for us we're focusing on asset side |
| 302s | optimizations for 5.0 and we want to |
| 305s | make sure that you know from a GPU side |
| 307s | that there'll be improvements there as |
| 309s | well |
| 309s | the reason why we don't have all these |
| 311s | optimizations shipped into 4.2 is linked |
| 315s | to the way we structure patches all of |
| 318s | our minor patches will have like smaller |
| 321s | chunks of perhaps content but mostly |
| 324s | code improvements whereas it's only in |
| 327s | major patch releases that we can feature |
| 329s | like perhaps level refactorings or |
| 333s | changes on the way we approach |
| 334s | animations or things like this like |
| 336s | riskier changes have to wait for larger |
| 340s | content updates from our side we want to |
| 343s | say thank you guys for your patience |
| 344s | especially over the last few weeks when |
| 346s | it comes to performance on hunt and we |
| 347s | appreciate any feedback you give us on |
| 349s | this patch which should now be live for |
| 351s | all of you we will keep on working on |
| 353s | performance and there's gonna be more of |
| 355s | these to come so thank you thank you |
| 360s | [Music] |
| 363s | achieved with CryEngine |