Want to help? I've found my PC's matchup: 4:4:4 video playback.

My PC is fast enough to run 4K video smoothly with a single Radeon 7950 GB and a six-core Phenom II PC, but with a few caveats that some of the next-gen videos files use today. The website Hd-trailers uploads free HD video clips movie trailers. 1080p files are often available, but they've offered a handful of 4k video, all of which have run smoothly on my PC, until now, when they uploaded two more recently: After Earth and Samsara here: http://www.hd-trailers.net/ultrahd/ These files use 4:4:4 color space instead of the typical 4:2:0 "compression." As a film geek, I've played some open-source video files in 4:4:4 and uncompressed color space/sampling before, and they've also run surprisingly smoothly in 1080p in YUV format. Some samples here: http://media.xiph.org/ Here, I wonder if there is either just an inefficient codec that was used, or there is some bottleneck somehere in my PC like my Phenom II when running nearly lossless compression codecs. Needless to say, if you have an i5, i7, 6 or 8 core AMD processor and would like to help out with a little 2 minute video test, try out these two 1GB mp4 files from Hd-trailers and let me know if you're able to run them smoothly or not, along with your specs- whether it's an intel, amd, DDR4 or DDR3, amount of RAM, and type of SSD or HDD, and whether it's SATA II, III, or M.2, etc. A 4k monitor is not required to run the video; if it runs smoothly, it will run smoothly on almost any monitor.
http://videos.hd-trailers.net/SAMSARA_TLR-2-UK_EN-XX_OV_51-4K-HDTN.mp4 (Edit: make sure to right click the file and press "save as" to download. It won't work via web player (and it would likely run slower than from a native video playback program like VLC).
http://videos.hd-trailers.net/AFTER-EARTH_TLR-D_GEN_EN-XX_INTL_51-4K-HDTN.mp4

I did compress one of these files using Handbrake to 60% compression using a default profile in only 12 minutes (thanks multicore CPU) and was able to run it smoothly in 4k. It changed the file to 4:2:0, however. The output file size was around 600MB.

Thanks for reading!

Ok, it will be later tonight before I can download the samples. I'm on a 4790K with twin HD Radeon 4670 (I'm not a gamer, a developer), DDR3 RAM (1600).

However, you can investigate this some yourself on your own machine while I, and others, are obtaining sample video.

You're either GPU limited (if you're enabling GPU rendering of video), CPU limited or disk throughput limited.

For GPU...well, that's a given. Either the feature is enabled in your viewer, or it's not...and the GPU is either at 100% or it's not.

For the CPU it's a little trickier. If the viewer has limited or no threading, you should set affinity to the number of cores applicable and see if that's pegged at 100% for those cores. On a 6 core AMD, if the viewer is not threaded, I'd expect the CPU shows under 20% utilization. What you need to know is if the single thread demand is overcoming the CPU, and using affinity and the windows task manager or resource monitor, you can figure that out.

If you're disk limited, you might not expect so. You might have drives that can provide, say, 120 MBytes/sec sustained and expect that should suffice. It can be, however, that Windows filesystem is not well tuned, and you're not getting a sustained supply of data.

In case that matters, disks for video should probably be formatted with large allocation blocks (I personally think all large disks should use large allocation blocks in Windows).

Thanks. I was able to make a few tweaks with small improvements. The first one only mitigated the issue. I changed the VLC playback speed->slower and it allowed a finer grain transition at the expense of real-time lapse. The bittrate didn't appear to be the issue, nor 4:4:4 alone, but I went into my Bios and discovered my drives were still set to Native IDE and changed it to ACHI. I have a 6.0Gb/s Seagate 5900rpm with 64mb cache on a SATA 3.0Gb/s motherboard. The sustained reads have gone below 60MB/s, which I think may be the cause, if not random read or other large sector benchmarks. I think it could also be the clock speed of the CPU- setting it to minimum 2,2ghz in AMD Catalyst Controll Center and in the Biostar BIOS utility to overclock to V6, leaving all cores on maximum turbo boost might do the trick. Perhaps the variable bittrate peaking suddenly to over 225Mbps may have something to do with it, although I was able to get nearly smooth speeds in linux and booting my other SSD, (a Plextor M5 Pro, a very fast drive that gets over 550MB/s read speeds), when using the bundled Totem player instead of VLC in Linux Mint. The realtime bittrate in Totem is less detailed, climbing steadily with GPU and/or CPU fans increasing to around 100,000kbps whereas VLC reports finer grained VBR spikes to 250,000kbps. I played an uncompressed .y4m file I had of Tears of Steel (a 185GB file which VLC says is 4:2:0 and tops over 1GB/s with moderate stuttering (not as bad as I was getting before with either 4:4:4) so I think it could be all three- CPU, SSD, and GPU. So steady improvements. This Plextor was formatted for ext2 or ext3. Maybe that might have something to do with it, although my Harddrive for windows doesn't appear to be defragmented- it's NTFS and was recently analyzed with no issues.

ext2 & 3 are superior at managing large files, and NTFS can do it well if tuned.

1 GBytes/s - that's gonna take RAID 0 SSD...if that's not G bits per second.

If 1 Gbits per second, then sustained 120 Mbytes per second is required, which is more "doable".

185 Gbytes for a 2 hour video would be sustained average 26 Mbytes/sec, which shouldn't be a problem on it's own.

With NTFS, I've had situations where the directory ends up with hundreds of thousands or even millions of entries, mapping out the location of a large file. When that happens there can be significant delays just finding the next chunk of data, which shows up as CPU usage in Windows while it figures out the directory (briefly), and thrashing of the drive (rotational) or lots and lots of individual (non-burst) requests from an SSD.

I would think a viewer with significant read ahead buffer, and sufficient RAM, would help a lot, but then that moves onto your CPU issue.

it's surprisingly more than that. A 185GB file for a 12 minute short film, if 'm not mistaken averages around 960MB/s. Uncompressed stills average 20-40MB per TIFF still, times 24 frames a second, and VLC has a built in y4m player (best to unzip the file first: http://media.xiph.org/tearsofsteel/tearsofsteel-4k.y4m.xz)

My Seagate has the AF large block section feature Seagate SmartAlign. I'm not sure if it's being used though.

VLC seems to play up to about 100MB/s stable. I have a 4k short film of Sintel that stays under 80MB/s and is the best quality video that has no issue of a complete sustained 15 minute run, thus it wouldn't have an issue if it were 2 hrs long. I have a couple short 2160p clips that are 50 frames a second instead of 24 for 6-7 seconds (see "crowd_run" or "ducks_take_off") and they have similar issues- they stutter after a second of marginally smooth footage and isn't repeated for another 5 or 6 seconds since the bittrate is around 850MB/s.

I checked resource monitor. It's utilizing all 6 CPU cores but only at 67% but with 99% maximum frequency. I've noticed best performance when my CPU and GPU fans start making noise. I just don't know what triggers it to spin besides the heat. I also checked Catalyst for GPU acceleration and they all appear to be checked.

I hadn't realized Raid 0 increased speeds. You may be right. I thought RAID was just better for handling larger files when hard drives were limited in size. And backups of course.

Xenohitsu said:

it's surprisingly more than that. A 185GB file for a 12 minute short film, if 'm not mistaken averages around 960MB/s. Uncompressed stills average 20-40MB per TIFF still, times 24 frames a second, and VLC has a built in y4m player (best to unzip the file first: http://media.xiph.org/tearsofsteel/tearsofsteel-4k.y4m.xz)

My Seagate has the AF large block section feature Seagate SmartAlign. I'm not sure if it's being used though.

VLC seems to play up to about 100MB/s stable. I have a 4k short film of Sintel that stays under 80MB/s and is the best quality video that has no issue of a complete sustained 15 minute run, thus it wouldn't have an issue if it were 2 hrs long. I have a couple short 2160p clips that are 50 frames a second instead of 24 for 6-7 seconds (see "crowd_run" or "ducks_take_off") and they have similar issues- they stutter after a second of marginally smooth footage and isn't repeated for another 5 or 6 seconds since the bittrate is around 850MB/s.

I checked resource monitor. It's utilizing all 6 CPU cores but only at 67% but with 99% maximum frequency. I've noticed best performance when my CPU and GPU fans start making noise. I just don't know what triggers it to spin besides the heat. I also checked Catalyst for GPU acceleration and they all appear to be checked.

I hadn't realized Raid 0 increased speeds. You may be right. I thought RAID was just better for handling larger files when hard drives were limited in size. And backups of course.

Click to expand...

Working backwards

Yes, Raid 0 works by causing both drives to work in tandem, leaving the access time the same but doubling the read/write data rate. A pair of drives sustaining 120 Mbytes/sec each can sustain 240 Mbytes/sec in Raid 0.

Without affinity, threads bounce around. If I run an application that executes on 2 threads, consuming 100% of the available CPU computer power all the time, I don't usually see 2 cores at 100% with the rest quiet. What I see on a quad core is all 4 cores at 50%. 67% is an odd number to hit by accident. It sounds like a 4 thread process sucking 100% of what a 6 core CPU can give it. In other words, IF you could set that to 6 cores, you'd have another 33% of compute power applied to the task.

I think we may have some confusion betweet bit and bytes in communicating speeds, but I'm not sure.

For example. 185 Gbytes / 720 seconds = 256.944 MBytes per second. Video is often quoted in bits per second, because of the way source data is provided, but drives are usually reckoned in bytes per second (at 8 bits per byte).

I did think the 185 Gbyte file was a 2 hour film.

Even so, 256 MBytes per second is fine for a wide range of SSD, tough even for a Raid 0 rotational hard drive setup.

Yes, you're right. 185Gbytes for 12 minutes would be 256MBps.

I changed the affinity to 4 cores but saw no increase in cpu usage. I wonder if it's limited by an antivirus or some other background program.

Xenohitsu, I haven't been able to get a download of the sample video yet, sorry. I've had about 58 Gbytes on download schedule these past several days, updating things from Android OS source code to XCode 6.1.1, OS X updates, Android NDK updates...man, this stuff adds UP - I'm gonna need a bigger pipe.

If CPU usage is still 67% at 4 core vs 6 core, then I'd expect all 4 cores used to be 100%, the other two idle.

If so, that indicates it's a 4 thread process utilitize all that the CPU has to offer. If you cut affinity to just 3 cores, you'd likely see 50% overall usage and the video would show more stuttering.

This hints to a CPU bottleneck. It might be alleviated if a viewer were able to utilize all 6 cores, but it appears the software isn't doing that.

You can get a little deeper view from Resource Monitor, where it can show the number of threads in use, but that isn't full detail.

I don't know of other tools which show threads and usage for each, perhaps Spy++ (part of Visual Studio), but this is becoming clearer.

AV isn't going to limit CPU usage, other than imposing it's own slice out of the aggregate compute power.

What you can try, in order to zero in on the question, is one of two approaches.

You can increase priority, which could help and could identify that you're riding right on the margin of the CPU's ability with that software.

You can also try various video sources in slightly decreasing demand (like 4:4:4 vs 4:2:0), monitoring CPU usage until you notice the usage drop below this 67% sustained consumption. That would identify, coarsely, what level is within the CPU's power, and when you're consumption is crossing into the range where the CPU can't keep up.

Also, resource monitor can show each drive's usage, something like CPU usage, in Windows 7. Windows 8.1's task manager also shows this for each drive. That can help identify data source as a bottleneck.

You may be experiencing both, probably very close to their maximum margin of sustained throughput.

I sense the CPU has unused compute power, and like gaming software, could rise to the occasion if only the software utilized it.

No worries. I have noticed my video speeds on 1080p clips slow down between under 2.9ghz on my other quad core PC. I get smooth playback around 2.6ghz, suggesting speed may have to do with it. Your i7 4790 is a great chip- it seems capable of powering smooth playback at 4ghz if the GPU is not limiting it. I've never used a PC at that speed, but it's probably twice as fast as mine. Whenever I do decide to upgrade, I'll go with at least a 6-core Intel or AMD.