Quote:
OK, back to the testing.
I tried the Google docs link again from my apt and it worked normally. It took about 15 seconds to download the zipped project and I was able to produce to "H.264 M2TS, AVC 4K 3840x2160/30p 50 Mbps, with the hardware encoder" in 4:54 on my system using PD14 v2109 and the newest nVidia driver 358.50.
My 780Ti card has 3072MB on board and GPU-Z showed roughly 2800MB in use. The GPU load was typically only at 30% while my CPU was pegged at 100% the entire time. The produced video looks and plays normally as far as I can tell, although I only have 2 HD monitors and can't watch it in 4k.
I ran the same test using the nVidia 337.88 driver, which usually gives me the fastest results - but it actually took 5:11 to produce, which is 17 seconds longer! The GPU memory was pretty much fixed at 2003MB the entire time, and that was noticeably lower than with the 358.50 driver. In this case, it seems the higher the GPU memory usage, the quicker the video is produced.
Thanks for trying. Did you look at anything else in GPU-Z, like GPU load, memory controller, video engine, bus load, by any chance ?
It's become quite to me clear that NewBlue titler is the main culprit here - not so much as the various versions of the nVidia drivers.
Jeff (JL_JL) was very helpful in creating a version of the project that uses the default title object instead of Newblue title.
I just did a whole bunch of tests and recorded tons of info during the rendering. Another sleepless night for me.
I will post the results in the thread about NewBlue at
http://forum.cyberlink.com/forum/posts/list/45843.page
What I did note was that there was fairly different baseline GPU RAM usage after starting Windows with just PD, between the different versions of nVidia drivers - sometimes different by more than 200MB . The 337.88 seems to have lower baseline memory usage. But not better rendering performance, in most cases - except when in the cases where PD happens to run out of memory, when using NewBlue.
Also, merely having a few browser windows opened increases the GPU RAM usage significantly, especially since I'm running Aero. While re-running tests without any browser running, I didn't see any "GPU out of memory" error with any of the drivers, even with NewBlue.
I also tested which formats PD14 would let me use the HE on using AVC H.264, and I was able to select any progressive M2TS format and use HE. All HD, 2k and 4k progressive profiles could use HE.
After doing one more DDU driver uninstall and reinstall, I'm seeing the same exact choices you are being offered - all progressive choices OK for M2TS for hardware encode, and all MP4.
Not quite sure why it was glitchy yesterday when I took the capture. Go figure
So at least in my case, the newer nVidia driver is better (faster), although neither showed the "no video" issue with the RTM version of PD14.
FYI, that "audio-only" bug was only occurring with later versions of the drivers, like the 350-355 series. If you weren't on one of these, you wouldn't have seen that bug. It might also have been specific to the 750Ti , while you have a 780Ti.
347.88 was OK AFAIK as having both audio & video in the M2TS . The latest PD14 2130 patch seems to have fixed it for all driver versions I tried , but I didn't retest all of them
BTW - according to my data, with drivers 337.88, doing a hardware encode, my rendering time was 276 seconds - ie . 4:36 - somewhat better than your 5:11 . This is test #7 in the table I will post in the other thread.
With drivers 347.88, the rendering time was 250 seconds, somewhat better - or 4:10 . This is test #14 in the same table.
With drivers 358.50, the rendering time was 246 seconds, ie. 4:06, compared with your 4:54. This will be test #19 in the same table.
So, at least for me, the newest drivers seem to have performed better for this project, when nothing else was opened on the system.
My CPU also was no peaked during any of my rendering tests for the project that included NewBlue. It never went beyond 90%.
And in some (anonmalous) cases, dropped to the 30s% range.
What I'm wondering about here is why you are reporting slower times than I'm, given your 780Ti with 3GB vs my 750Ti with 2GB.
And also, your CPU is an Intel Haswell i7-4770k which cost much more (up to twice as much on Newegg, twhen that Intel part was still available) and was released in June 2013, and which I thought would perform better than my FX-8350 which was released in October 2012 and whichI acquired right upon release. All the independent benchmarks reviews I read pointed to the Intel chip being significantly faster, and the FX-8350 being closer to i5 performance of the time.
However, I did spend a lot of time stabilizing the 4.6 Ghz OC on my FX CPU, so perhaps I really did manage to get a manage that outperforms your Intel chip significantly. But it does surprise me to see such difference given that your chip includes Quicksync, while mine does not, and Cyberlink touts its multi GP-GPU technology. Perhaps that's what explains the higher CPU utilization on your machine, and maybe it's not actually all it's touted to be
I was planning on perhaps getting an Intel chip in the future for the Quicksync, but maybe it doesn't make sense just to do it for this reason, and I should wait to at least double the number of cores for my next upgrade to make sense.
Or maybe Cyberlink improved the performance between the RTM build of PD14 and patch 2130, and forgot to tell anyone
Julien
This message was edited 2 times. Last update was at Oct 10. 2015 11:34
MSI X99A Raider
Intel i7-5820k @ 4.4 GHz
32GB DDR4 RAM
Gigabyte nVidia GTX 960 4GB
480 GB Patriot Ignite SSD (boot)
2 x 480 GB Sandisk Ultra II SSD (striped)
6 x 1 TB Samsung 860 SSD (striped)
2 x LG 32UD59-B 32" 4K
Asus PB238 23" HD (portrait)