|
On February 24 2013 16:29 Cyro wrote:
Thats not good, what FPS do you have in the game? I would see if you could impose some kind of FPS limit or something - your CPU usage is too high as it is, i would hesitate even for your current settings (you are clearly maxing CPU cores very often)
I don't have a way to see my fps in game without downloading Fraps, but it feels relatively smooth so I would estimate about 40-45 fps not streaming, and barely above 30-35 when streaming. Alas, I am tied to a laptop. 
Other games like DoTA2 or Diablo are fine for me; I'm just seeing if I can improve, or if I should even continue streaming Blacklight.
|
United Kingdom20326 Posts
Its just that you have like 75% CPU usage without even streaming and you are on a relatively weak dual core, its difficult
|
I'm getting lots of disconnects lately. Like every 7 min. or 20 min. it has to reconnect. Anyone else having the same issue on the current and beta versions?
|
can anyone help me with streaming on daily motion?
|
I am running into this error when trying to start the stream.
Cannot initialize desktop audio
Anyone know why this is happening?
|
Russian Federation81 Posts
Good day, gents. I can't believe I haven't found this thread earlier...
So I'm a shitty diamond player, and I like to show my play to a couple of my friends once in a while. The problem is, I am not quite happy with the game performance I get when doing so!
My PC specs are:
Intel i7 950 Bloomfield OC'd to 4.2 Ghz, HT enabled
12 GB DDR3 RAM (OC'd to 1800Mhz I believe but I might be slightly wrong here, not that it matters)
HIS 7970, again OC'd to the limit
60 GB OCZ Vertex 3 SSD with Win 7 Ultimate x64 installed on it
I play with the lowest settings and get 200+ FPS. That feels very comfortable and I like it.
Now when I start streaming, FPS drops. Around 100 at start and to 40-60 with 200 supply armies (recent example). Mind you, on twitch it looks like everything's fine. But when you're actually playing a game and have to control that 200 army it feels choppy. And seeing people (pros and not) stream mid to high graphic settings, 1080p even, makes me question if I'm doing something wrong?
Here are my CPU usage graphs:
While streaming
![[image loading]](http://i.imgur.com/G6zl6pMl.jpg)
No streaming
![[image loading]](http://i.imgur.com/EO1xuJzl.png)
OBS settings:
Quality : 10
Max Bitrate/Buffer Size : 3000/3000
Base Resolution : 2048x1152
Resolution Downscale : 1.5(1364x768)
FPS : 30
Windows Aero : Disabled
x264 CPU Preset : veryfast (superfast doesn't make any difference)
I have also switched from capturing the whole display to capturing SC2 window (which appears to have not done much?). I've seen people mention capturing the process, is it the same thing or am I missing something?
I would also love to be able to switch to streaming mid/high settings at some point, but the choppines there only becomes worse. So my question is, is my CPU simply too old and slow for that, or am I missing some stupid detail?
EDIT:: Looking at the CPU graphs.. shouldn't encoding take advantage of HT and use all 8 cores?
|
United Kingdom20326 Posts
Your encoding is not nearly stressful enough on CPU for it to bother with it, you could put 3-4x as much load on CPU in terms of encoding ability.
Its pretty normal what you are experiencing, ESPECIALLY the lategame part + Show Spoiler +Sc2 does not scale well, there are no CPU's that will keep FPS up in more stressful situations - this is a 2500k benchmark > results differ person to person quite a lot, but massive performance losses are pretty much expected to happen. UPDATE: I repeated the experiment with a late game ZvZ replay with maxed armies:
3300 MHz:
Avg: 41.925 - Min: 27 - Max: 58
3700 MHz:
Avg: 46.625 - Min: 32 - Max: 63
4400 MHz:
Avg: 56.025 - Min: 39 - Max: 77
You just lose a ton of performance in the screen capturing process, before encoding (which is clearly fine on CPU load) which is why tripling CPU load by going from Veryfast to Medium preset is often completely invisible on the FPS meters, for example. You cant measure it in OBS seperately, because you have to go all-or-nothing, you either preview the game, captured, encoded, everything, or you dont. Here's a chart with just capture from Xsplit gamesource i made a couple days ago - same CPU at 3.8ghz, reproduced almost exactly by a friend with yet again the same CPU
+ Show Spoiler +
I have much higher base FPS, id suggest updating drivers etc to try and get it up - friend was stuck far lower (100-250ish i think?) until he did that, even though he was on a pretty recent driver, though it was an nvidia card (we both use nvidia cards - me 580 and then 260, him 670)
You can see there increasing resolution does not hurt you too much, 540p30<1080p30, 4x the resolution but only doubles the performance loss, so its a lot less expensive than increasing framerate - that might not be true for OBS methods, i didnt check everything very extensively myself and see conflicting reports of performance all the time.
![[image loading]](http://i.imgur.com/UGIKuVV.png)
This is easily reproducable, it drops as the game goes on, some parts of the map give higher FPS, some lower (and not due to GPU load) with pretty far over 300fps expected on any map in the first minute
Your goal should just be to have as much FPS as possible at the start, get a reproducible scenario like i did, a paused replay at 1 second (not 0 sec, it bugs) and look at the same part of the map, and then find settings you are comfortable with - the FPS loss from OBS/Xsplit is basically just a multiplier on whatever FPS you would have anyway, there does not seem to be a lot of evidence that it behaves very differently, at least not yet, so your goal is just to get performance as good as possible, its going to be pretty bad whatever you do if you are performance sensitive
Its unfortunate that you have an odd resolution, I'd suggest setting up GPU drivers so that you can have small black bars around bottom, top and sides of screen, so that you can play in 1920x1080 and stream from that - you wont lose any viewing area in sc2, it's clamped to 16 : 9 resolutions so it will be identical. CPU-wise, high end for the CPU and kind of overclock you want would be 1920x1080, veryfast, 45fps, ive done pretty extensive testing on that, but if you want the best performing game, you could try something like 20fps, it should run a lot better - If you only have 200fps at the start of game on low though, something is wrong, i can double that and then some if i try with a system that is inferior in every way - that probably translates to doubling midgame performance, too. All the numbers i have there are in the same paused replay at 1 second at the same camera location, solid and easily reproducable - but in many places of the map, id have like 340fps, so everything else scales down a bit as well - and in a really stressful lategame situation, FPS drops from that by an order of magnitude - maybe even slightly more. Thats just how sc2 runs.
R1CH/Jim etc, if any of this info is wrong, please correct me. I think it should be generally ok
|
Russian Federation81 Posts
Okay I am sorry for feeding you the wrong information, I simply took the number 200 out of my head without thinking it might be of any significance. I only knew it wasn't lower than that. Actual testing showed this
![[image loading]](http://i.imgur.com/47oJpKNl.jpg)
So I guess that is fine...
I just did some excessive testing with different resolutions and modes and here's what I can conclude. Using 1920x1080 instead of my native resolution has _almost_ no impact on performance: I used medium graphic settings and had 70 FPS with 1080p+ and 75-80 FPS on 1080p at the very start of the game. Later I built a 300 ling army and had around 30 FPS staring at it on both resolutions. so while it does affect the performance at start, it doesn't when it matters (lategame).
What I was really surprised to read and actually see in practice is:
You just lose a ton of performance in the screen capturing process, before encoding Streaming 720p@30fps vs 1080p@30fps seemed to have absolutely no performance difference. This fact really blows my mind. I mean, clearly the CPU graph shows that my CPU can handle more, why does it bottleneck :/
I am also going to play with HT and see if it helps. Since those cores aren't doing anything, might as well turn them off right?
EDIT:: Turning HT off did nothing. Sad
|
United Kingdom20326 Posts
I mean, clearly the CPU graph shows that my CPU can handle more, why does it bottleneck :/
Because you are bottlenecked by other things, not by CPU load. The cores will be loaded more and HT will be active when you need them, when you use far more intensive settings (or at least they should) 1280x720@30fps@veryfast is something like a quarter of the load that CPU can handle - if you see four cores at like >70% while playing the game but HT cores are completely idle, maybe its cause for concern, but the more cores you use the lower encoding efficiency you have (by a relatively small margin) so i wouldnt be suprised if its just x264 being smart when you are not putting much load on the CPU.
Can you give me a number for that same scene when you start streaming at 1280x720, 30fps and 1920x1080, 30fps with OBS game capture, nothing else going on?
|
My laptop has an i7-3630 QM processor and a Nvidia Geforce 650M. Would streaming LoL / random PC games be smooth?
|
United Kingdom20326 Posts
Depends what you call smooth - it should be obvious that there is some performance loss
|
Russian Federation81 Posts
I feel really stupid now.
After you've mentioned game capture I actually bothered to look it up. I have no frigging idea how is it possible that I've been missing it for MONTHS I've been using OBS ...
1080p+@30FPS - I am currently at work so this is what I've been able to put together through TeamViewer. 150-160 FPS at start, staring at workers. This is incredible. Another video I've recorder yesterday using display capture, not sure if it's 720p or 1080p, but it's clearly doing two times worse than game capture in terms of FPS. I'll do more testing this weekend on ladder games, but as long as it doubles the display capture performance I'll be more than happy!
|
Streaming workers doesn't really mean anything, you need to actually stress it and see whether you get playable frame rates in late game scenarios.
|
Russian Federation81 Posts
Cyro suggests that performance scales linearly, so if I get double performance at start in theory I should be doubling lategame too. Obviously I will test it once I get the chance, just throwing the info out here.
|
Russian Federation81 Posts
Full game 2048*1152@30 FPS, mid graphic settings. Basically what I've been hoping for is true. Around 50 FPS while moving a 250ish ling/bane army (13:00 and till the end). It drops down to 30ish when I'm moving it into the opponents army but the important thing here is that the choppiness that's been pissing me off before is absolutely gone. This particular video has some dropped frame issues, just ignore them. I'll see if I bother to fix that or if I won't since I wasn't originally going to stream more than 720p. Anyway, thank you Cyro for giving me that small hint about game capture! It's exactly the fix I've been looking for.
|
United Kingdom20326 Posts
Cyro suggests that performance scales linearly, so if I get double performance at start in theory I should be doubling lategame too
Well what i meant is that performance at one point in the game is indicative of performance at another - if you have more FPS with the same settings at one point you probably will at all of them
Streaming workers doesn't really mean anything, you need to actually stress it and see whether you get playable frame rates in late game scenarios.
I always do quick checks on that map position on ohana with nothing going on, because it's easily reproducible and always gives the same results that seem to scale throughout the game in terms of % framerate lost for X/Y settings
|
Keep in mind at the start of a game you're almost always GPU-bound (on a good system) and mid to late game you become CPU bound.
|
United Kingdom20326 Posts
Keep in mind at the start of a game you're almost always GPU-bound (on a good system) and mid to late game you become CPU bound.
Not on low, thats why i do some of my testing like that
|
On March 03 2013 02:15 R1CH wrote:
Keep in mind at the start of a game you're almost always GPU-bound (on a good system) and mid to late game you become CPU bound.
This is most certainly true (at high settings anyway). I have the game completely maxed @ 1920 x 1080 on a 7950 (overclocked to 1050mhz). I get about 90-120 FPS (depnding on the map), and maintain an FPS above 60 until the mid-late game.
I just got my 3570k stable at 4.6ghz @ 1.21v on a Hyper 212 + which I'm super happy with. I think I may be able to get 4.8, which would be amazing! This is super helpful for lategame scenarios, I can't really ask for better performance than that.
|
United Kingdom20326 Posts
Doubt you'll be able to get 4.8 on a 212+ (though it all depends on temps), and how thoroughly did you stability test the 4.6, 1.21 (also what's your load vcore in cpu-z?)
it seems a bit low, you might have just got a little lucky with the chip though
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/G6zl6pM.jpg)
![[image loading]](http://i.imgur.com/EO1xuJz.png)
![[image loading]](http://i.imgur.com/C6SMeY6.png)
