Search options (Page 49 of 55)
One thing to note is that a setting of 6 threads now is resulting in the problem... it seems to be very inconsistant because originally 4 threads worked fine, but yesterday when I tried it 4 threads resulted in said problem and yet today 4 threads is fine...
Mystery wrote:When you get flickering, is CPU capped at 100%?
Just tested it right now with a ~30 second clip and my CPU utilization was ~85% with an SVP index of 1.02x.
Mystery wrote:And when you watch the same video with old DLLs, is CPU usage lower in those cases?
Definitely no - with the stock SVP 3.1.6 + hotfixes DLLs I had to have my CPU overclocked at ~3.8GHz, but with the new DLLs I can run my CPU at its stock 3.2GHz without changing any SVP profile settings and still retaining an SVP index of 1.0x.
Mystery wrote:Does LAV Filter 0.65 include these optimizations?
Well shoot, I've done it again - made recommendations of trunk builds without realizing that a new stable version has been released. 
(I made the same mistake with MPC-HC since it was at v1.7.7 for so long)
So yes you could just use LAVfilters v0.65 instead.
EDIT:
Chainik wrote:It can be a know bug for the AVS plus, not for SVP libs.
Uhh, this doesn't make any sense because all I've literally done is take a stock copy of SVP 3.1.6 + hotfixes and replaced the svpflow1.dll and svpflow2.dll files with newer ones...
Did you make an actual profile that is 1440x1080@24? You did not say if you did or not...
Also, try toggling and/or doing various combinations of the two options under "Profile selection method"
Yup, that's it.
And Chainlink, Microsoft is a corperation, not a software developer. You really think they don't use yyyy-mm-dd in-house?
Want even more of a boost? Use the newest trunck revisions of LAVfilters (if you're not already).
For two smaller CPU utilization reductions (more noticable on weaker hardware), if you're not using madVR, you can use "D3D Fullscreen" . Also if you're using MPC-BE, then select the "MPC Audio Renderer"; if you're using MPC-HC, then you can download the MPC-BE standalone filters and add it to MPC-HC's "External Filters" list manually (make sure you set it to "Prefer!".)
In total, if you were using a stock copy of the MPC-HC that came with SVP, you could see a 10-20% CPU utilization decrease by doing all three things.
James D wrote:try to manually set # of threads to other values.
Yup, setting 6 threads or less on my dual-core Pentium G3258 works well; by comparison setting 7 threads or more causes the flickering/shimmering effect that I saw with "Auto".
I'm guessing this is a known bug?
EDIT: And now I get to pick on you Chainik with your Doom9 post, what kind of software developer doesn't use ISO 8601 for version dating, that is YYYY-MM-DD? And you're not allowed to pull the American card because both of my parents and myself have lived in northeast Ohio our entire lives.
If you're downscaling to 1080p, then make absolute sure you're getting 4:4:4 chroma since being able to get a full chroma resolution is one of the benefits of downsizing 4k video to 1080p.
Easiest way to do this is to open the following test image in MSpaint:
http://forum.doom9.org/showthread.php?p … ost1640299
Chainik wrote:And for the record - it's up to 38% faster. Which means from zero to 38%.
There's a reason I used the words "could" and "maybe".
Besides, I got a wierd shimmering and/or flickering when using the exact same settings I've always used that doesn't seem to occur on the official 3.1.6 + hotfixes, so for actual video use I'll continue using that. For reference, said shimmering/flickering seemed to be greatly reduced when I enabled "by two with global refinement" if that makes any difference.
Mystery wrote:Yes! I just replaced the 2 DLLs. As a start, video initialization is considerably faster.
Which DLLs in the SVP program folder are these?
(assuming I don't find it myself first...)
EDIT: Whelp, found them. They're in "Program Files\SVP\Plugins"
I'm definitely going to look into this and use it as a baseline for performance in my undervolting.
Well for one, I did not know that SVPflow used newer backends than SVP 3.1.6.
So 38% faster? Wow, I could get away with maybe a 2.7GHz clockrate then!
Um, I think we are talking about two very different things here.
I'm asking what kind of performance improvement could I expect from this fabled "new version of SVP" compared to the current v3.1.6 + hotfixes?
Actually yes, because I can't find an actual official dedicated thread for SVP on Doom9 - only related projects that use SVP (SVPflow and InterFrame in particular) and generic threads where people are having issues with SVP and/or are asking for help.
MistahBonzai wrote:Chainik wrote:MistahBonzai
EVR-CP is not good for 4k@60. Plain EVR works much better.
And so it does 
Thanks.
If it's anything like 1080p on old low-end Intel iGPUs, D3D Fullscreen may also work much better than EVR-CP.
You said that LAVfilter optimizations don't count. (besides, I'm already using trunk LAV builds)
Or are you implying that there's an actual SVP thread on Doom9...?
So uh... just how much more performance will this new version of SVP have? The only daily reason I have my CPU overclocked is for SVP, so I've been stability-testing some lower-overclock settings during this recent wave of cold air in the NorthEastern US and summer hasn't hit yet, but I am unsure just how much I can lower my clockrate without also lowering my SVP settings...
Currently I run my Pentium G3258 at 3.9GHz @ +0.007v and that gives me ~90% CPU utilization in SVP for a typical intensive use-case of mine, and so far I've fully stability-tested 3.8GHz @ -0.026v, 3.7GHz @ -0.049v, and 3.6GHz @ -0.063v.
Mystery wrote:For SVP on 8k video, just increase the pixel size from half pixel to 2 pixels.
What if you're using an 8k head-mounted display where you need as much resolution as you can get?
Chainik wrote:Still I admit that 8K video definitely will need x64 
Therefore it's better have a 64bit version now so that the bugs can be ironed out before such a thing is actually necessary for normal usage.
Of course, I wouldn't be surprised if SVP on 8k video will need some crazy hardware that that rumored 200-300w TDP HPC Zen APU with HSA + HBM memory.
Mystery wrote:An integrated chipset will pretty much only handle SVP.
That would imply that the iGP on the A10-7850K isn't enough for madVR...
Mystery wrote:Yes. First, madVR requires a Radeon or AMD to run properly.
Uhh, did you mean to say it requires Radeon/AMD or Geforce/Nvidia?
I don't really have the issue in question unless SVP is running at less than 1.0x performance, and even then it's not audio delay but rather video delay relative to the audio - hence why I was suggesting that maybe it's just a performance issue?
MAG79 wrote:Nintendo Maniac 64
Maybe it is the problem of certain video file?
Uhh, what I meant is that interpolating up to 144/150fps seems to take considerably more grunt than interpolating up to 120/125fps, especially when you consider that 120/125fps doesn't seem to need that much more grunt than 90/96/100fps
The lightboost thing is just being able to get the most out of your lightboost monitor if you're going to be running at a lower refresh anyway (lightboost itself does not support 144hz - it maxes out at 120hz).
I can confirm from my own experience that 144fps and 150fps requires considerably more performance than you would expect over even 120fps.
Therefore it may be wise to instead use something like MPC-HC's built-in resolution opitons and set it to 120hz for 24fps & 30fps content and 100hz for 25fps content.
If you are hesitant to give up the full 144hz, then I recommend seeing if your monitor is "lightboost" compatiible which will improve your monitor's motion resolution considerably:
You should clarify that the performance of a graphics card does not reduce or increase the visual quality of madVR, but rather having a better GPU allows you to set madVR's options to higher-quality settings.
Also make sure you're:
1. upscaling rather than downscaling
2. Not using Intel graphics over HDMI (has a tendency to output with reduced chroma)
Apologies for double-posting, but I just came accross a 24--in->25fps video. I never mentioned such a thing because I didn't think it would actually exist, but it would seem that I was wrong.
Example:
https://www.youtube.com/watch?v=2i-t0C7RQWM
Could this be related to SVP using more CPU when a profile is set to "Off" than when SVP is closed completely? I particularly noticed this with 60fps 1080p content on a 2.4GHz Penryn Core 2 Duo (I'd get full performance with SVP closed but some infrequent slow-down with SVP set to "Off")
Posts found: 1,201 to 1,225 of 1,368