00:00Peggy 16
00:04Hey Planetside 2 players, my name is Kyle Ryan
00:07I'm the Vice President of Technology here at Sony Online
00:09And I'm here to talk to you about the Planetside 2 optimizations
00:13So, I know you guys have seen the tweets from Smed
00:16You guys have heard some of the noise that's been going on
00:19As we've tried to do Command Center
00:22I cannot wait to find out what happens on next week's episode
00:25Of Operation Make Faster Game
00:28Holy s***
00:29I think it's probably going to make the game faster
00:33What we're going to do is we're going to document this
00:37Just like we were talking about in Command Center last time
00:39We're going to expose a little bit of ourselves
00:42What we do and how we do it
00:43Talk about some of the specifics about the changes that we're making
00:47And how we're going to make Planetside run faster
01:11We want to talk to you separately
01:17You know what this is about
01:23My name is Ryan Eelm
01:24I am the Technical Director for Planetside 2
01:28Basically the guy that goes to the scrums every day
01:30And listens to what's going on
01:32What the guys are working on
01:33The problems they're running into
01:36Channel those around to the various teams
01:38There are a lot of teams working on this
01:39They're all over the company
01:42And we've got to basically marshal all those forces together
01:44Make sure everyone's working on the thing that they should be working on
01:47And if they run into any problems
01:49Make sure that we can beat those problems out of the way
01:52All right, let's start with you
01:53What's going on?
01:55We are continuing our optimization efforts
02:00In various fronts
02:02One thing that I found after looking at the NUAM profiler
02:05Is we are doing input processing on the HUD
02:08Even when the mouse is not visible
02:09It's trying to find the topmost entity
02:12So it's probably using some sort of recursive situation
02:15To do that every frame
02:17And sometimes the hit can be as much as 0.3, 0.4 milliseconds
02:20We don't need to do that
02:21So I'm going to work with Josh
02:23On kind of only doing some of those checks
02:26On the Swift Center open if the mouse is out
02:29So with a live game a lot of times
02:31It's hard to actually find the time to break the leg
02:35Before you can set it again
02:35You know, because essentially you have a game
02:37Where people are in it all the time
02:39They expect it to work all the time
02:41Every change has to be bookended
02:43So that it can be out and live quickly
02:45And that just wasn't going to fly
02:47For some of the things that we needed to actually fix in this game
02:51So we took a step back
02:52We looked at where the frame rate was
02:54We looked at the things that we needed to fix
02:56And we said this is going to take serious time
03:00And at that point Smed said
03:02You guys take the time
03:03Everyone on the team
03:04Get behind making the game faster
03:07This is Mitch J everybody
03:08Hey
03:09You got to make it fast too
03:11Like usually people take like 20 minutes
03:13I should have shaved
03:14I should have shaved
03:16Alright so
03:17I was looking at some of the spikes in the overhead
03:20Kind of timing block
03:23And it turns out a lot of the spikiness that we think we see
03:27Is actually just things that are in that big overhead
03:30Pile of stuff
03:31Happening not at frame level
03:33But at like 10 frames or 100 frames or 1000 frames
03:37So you'll get that kind of stuff as these things kind of kick off at different times
03:41So I try to break out that a little bit to smooth out the different timings
03:46And a lot of the spikings just went away
03:50So we're here today in the fishbowl
03:52Where we often meet with the programmers
03:54To talk about the progress
03:56What tasks are working
03:57What's not working
03:58What we need to do next
03:59So today we're doing the last step of the packet analysis profiling tool
04:04Where we're going to convert the packet IDs that we have
04:07Into a readable stream
04:08So it's easy if it pops up
04:10You can tell exactly what packet is causing an issue
04:12So also I'll put some telemetry tags and stuff like that
04:15Rob had some suggestions there
04:17But then that should be usable by today hopefully
04:20Sweet
04:22Both the telemetry and the visual?
04:23Well yeah, because telemetry tags are really easy to add
04:25Yeah, it's like a single line
04:26Sweet
04:26That would be really nice
04:28So at this point the programmers are working on
04:31Several different tasks
04:33Things that come up on our profilers
04:35Like rendering and physics
04:38The user interface
04:39And the time that it takes
04:40And I did notice
04:42Maybe that's what I saw yesterday
04:43Something's creating shaders a lot
04:46Like something within scaleform
04:48Is causing D3D to create a shader
04:51When you do different things
04:53So maybe that's what's creating the itching
04:54Is for some reason it thinks it needs a whole bunch of shaders
04:57So one of the major parts of this optimization
04:59Is to go through every single system
05:01And figure out how we can better use the multiple cores on your machine
05:05Right now we do have things like sound, physics, and animation
05:09And loading in their own threads
05:12But a lot of times the main thread ends up just waiting on those operations to finish
05:16Which results in it looking like nothing's really going on on your CPUs
05:20And also a reduction in frame rate
05:22I was working on the UI yesterday so you can load in an XML
05:26So you can have a load in which instruction set you want to go in
05:30You can switch between either the dumb brain or the instruction brain
05:33So going through the various systems that take time on the main thread
05:41We have found areas that were going to be a significant improvement in frame rate
05:46But we're going to also take a significant amount of time to get done
05:49And those are a lot of the things that now that we have the time we're doing
05:54Yeah I've gotten a quick look at our cache coherency so far
05:58And it looks fairly good
05:59I mean there's not crazy amounts of cache
06:01We're not like doing the stupidest thing ever
06:03And looping the wrong way around a matrix or something like that all the time
06:08Most of our L3 cache stuff is pretty minimal
06:11We're hitting cache when we need to it looks like
06:13Nothing is ridiculously crazy
06:16But I want to see if I can get an export of that as a baseline
06:20So we can see if we're changing that behavior as well
06:21So anyway we're going to have a lot more of these videos
06:23You guys are going to see all aspects of this process
06:26We know you've been very, very patient
06:28We want to reward that patience with sort of a glimpse inside how we get this done
06:32We want there to be no more secrets
06:35We want you guys to be able to discuss facts instead of opinions
06:37I mean it's just in all of our best interest to get this information out to you
06:41And now would you guys please go away
06:43And now would you guys please go away
06:57Sony
Commenti