Claude.Rivet Posted February 6, 2017 Report Share Posted February 6, 2017 Good day, I am in the process of building new stations, I will post results and testing procedures when I am done as requested in another thread when I am done. However while testing HAP playback I stumbled upon an issue where my production computer was running out of video memory (as stated in the message panel). At 8GB I was not so surprised as I was playing 50x HD files however when playing those same files on my display computers I got the same message even though I have 16GB of video memory available. My production preview was set to video as thumbnail. Surprised I ran GPU-Z on my display and production computer while playing back those same files. I also ran AMD system monitor and Windows ressource monitor to get a bigger picture of what was happening. I logged to file when possible. GPU-Z reports a maximum of 304MB used on card... memory clock was topping and so was core clock but memory didn't seem to be even remotely used, telling me either GPU-Z was badly reporting or only the output frame buffer is reserved (HD 249MB) plus a little overhead and therefore GPU ram was indeed not used at all as all compositing was done prior. So why the video memory issue then? So I am getting this error where Watchout runs out of video memory even though all tells me there is still much memory left. Even if all meters are wrong (which is a possibility after all) there is still the odd result of running out of memory with the very same amount of data on both an 8GB card and 16GB one. Somehow I have a feeling I have a lot of unused power there . side NOTE: encoding HAP in chunk is not always a good idea btw, it helps to play back huge files but also greatly limit the amount of files you can play back simultaneously, in my production computer as soon as I play back 2 HD files encoded in 4 to 8 chunks (8cores processor) the playback as issues, with no chunks I reach 50x HD files. So be carefull about chunks, it helps, but only for huge files and only if you play back a minimum of file at the same time also, it is not the solution to my issue, it has been tested. 0 Quote Link to comment Share on other sites More sharing options...
Claude.Rivet Posted February 7, 2017 Author Report Share Posted February 7, 2017 Is it possible HAP files are decoded with the intel GPU rather than the amd pcie card? It would explain getting identical results from an 8GB and 16GB machine and the GPU-Z reading. 0 Quote Link to comment Share on other sites More sharing options...
Mike Fahl Posted February 8, 2017 Report Share Posted February 8, 2017 What's the exact error message you're getting? Can you provide a screenshot of the message? Your notes on chunked HAP videos are correct. Using chunks allows the used of multiple CPU cores working on the same video. This can help if you want to play one or two huge files. But it may actually hamper the performance when playing many smaller videos, as those would be spread out over multiple cores anyway. Possibly, WATCHOUT could be a bit smarter here in how it allocates threads to HAP chunks, taking the number of simultaneously played videos into account. Mike 0 Quote Link to comment Share on other sites More sharing options...
Claude.Rivet Posted February 8, 2017 Author Report Share Posted February 8, 2017 Here is the screenshot, long story short HD007 and HD008 are the 55th and 56th files added, usually is starts at around 51st and 52nd. http://imgur.com/a/FZUY4 0 Quote Link to comment Share on other sites More sharing options...
Mike Fahl Posted February 9, 2017 Report Share Posted February 9, 2017 Yes, those messages sure seem legit. Just to make sure I understand this properly. Are you playing 50+ videos simultaneously or sequentially? What codec is being used? There's a protocol command to check for available memory that may help you in tracking this one down. Please check with dataton support for the exact syntax here. The idea being that you could ussue this command repeatedly (using a telnet client or similar) while running those videos and then see what impact it has on memory from WATCHOUT's point of viw. Mike 0 Quote Link to comment Share on other sites More sharing options...
Claude.Rivet Posted February 9, 2017 Author Report Share Posted February 9, 2017 50+files simultaneously, introduced one by one with 1second gap between. I will ask support, thank for the tip! 0 Quote Link to comment Share on other sites More sharing options...
Claude.Rivet Posted February 27, 2017 Author Report Share Posted February 27, 2017 Still haven't got any reply from Dataton so I guess I won't get to test that. So basically Hap seems to be decoded in the CPU embeded GPU rather than on the AMD W9100 which only reserves the amount of memory needed for the frame buffers (or whatever it is named). If I used two output it goes from 302-304MB to 600ish and at 6 outputs the amount of reserved memory is around 1.8GB, so GPU-Z does report some stuff correctly now maybe some usage is not monitored and my assumptions are wrong but it does look like WO only takes ram on the GPU for it's output and decodes HAP elsewhere. My guess is if that get fixed (I will post on hap forums in case the problem is in the decoder) we will see a dramatic increase in playback performance as we will now use to pro card with 16GB of ram availlable for HAP decoding. Anyways, thanks for the help, hope to see some fix at some point. regards 0 Quote Link to comment Share on other sites More sharing options...
Claude.Rivet Posted February 27, 2017 Author Report Share Posted February 27, 2017 Hello guys, I posted the issue on Github and they answered saying it was a Watchout issue and to contact Dataton. So it IS a known issue, and it IS in the hands of Dataton, this is almost good news as it means we will soon see a dramatic increase in playback performance as soon as this is fixed. Thanks for the help looking forward to the fix regards 0 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.