[Dave Haynie]

Typically, you want CBR for a streaming medium, like online video, satellite television, etc. and VBR for finite-storage-based media like DVDs and BDs.

There’s no sense to “2-pass CBR”… didn’t even think it was possible. The whole point of two-pass is optimizing VBR. In VBR, you basically have a “bit-budget”… you can spend bits here or there, and there are various strategies.

One example of is CQ… sort of a constant quality, but it actually implies that the same amount of data will be tossed filtered after the DCT. This makes the video very consistent, but the resulting size is hard to control — you only get the average bitrate you select when you have average video (eg, nothing too dynamic).

The 2-pass basically lets the encoder “cheat”… it can scan the entire video, qualify the whole thing in terms of dynamics, then apply just the right range of encoding to optimize that range of movement while hitting your bitrate target.

CBR, of course, always hits your bitrate target… and unless you’re encoding at maximum possible bitrate, it won’t look at good as VBR on most video. Sure, on a perfectly static talking head, there’s not likely much difference, but for real world video, you have sections on a DVD that look just dandy at 3.5Mb/s, and others than would really like to go past 9Mb/s.

In professional encoding, there’s a mastering engineer with realtime encoding tools who can “eyeball” these things. And even play other tricks. For example, one of the worst since in MPEG encoding is leaving visible macroblocks. That’s when the compression is so high, frame to frame, that they blocks basically don’t line up anymore, and your eye sees that mis-match as block lines. You tend to see this in fact motion, where the whole motion prediction thing kind of breaks down. If you apply a global low-pass filter to those sections of video, there will naturally be less high frequency data to toss out of the DCT, but no visible macrocells, since the filtering was global.

But I digress. If the Main Concept encoder actually does anything with the 2-pass flag and CBR, I’m at a loss to say just what. One thing to check: a CBR encoding at the same bitrate is always going to be the came length (that whole “constant” thing). Try with and without, see if you get the same file size, or even identically encoded files.

-Dave

[Dave Haynie]

[Carl Alessi] “Right now, I’m having a debate with our tech guys about whether it’s necessary to make sure that I have a dedicated drive in the box just for rendering: they believe that at transfer speeds of 6 GB per second, having everything on SATA RAID 0 would be sufficient (they’re not sure that the Dell Optiplex they’re going to purchase will support RAID 0 plus a third, stand alone SATA drive internally).”

It’s not NECESSARY to have a second drive. It’s optimal, however.

And here’s the reality… the link speed of SATA 3.0 is 6Gb/s (six giga-bits per second) = 750MB/s. And even at that, a dual drive RAID0 isn’t going to perform anywhere close to that speed. You might get 100MB/s, maybe even 200MB/s. Probably not more.

Second issue… that’s “straight-line” speed. So if you have just one file, let’s say you’re getting 200MB/s. But one file isn’t very interesting. If you read one and render to another, that’s two files.
You can read at 200MB/s or write at 200MB/s, but not both to the same drive… so that’s a throughput of 100MB/s. Add a third file, and now you have to read two tracks, write one… now you’re 66MB/s throughput. And so on.

Only, no.. no soup for you. You’re not going straight-line anymore. Two files means seeking before each chunk of a file is read. So let’s say you have a two of those new WD 3TB drives, which haev an average seek time of 15ms. So for every full read/write (both files), you have 30ms of seek to factor in. So that 100MB of throughput is actually going to take 1 second + 0.030 second… so the throughput is actually 97MB/s, assuming 100MB buffers. But if you only buffered 10MB, that would be ten seeks per 100MB… 1 second + 30ms * 10 => 77MB/s.

What if you had three inputs and one output file being processed. Now it’s 50MB/s raw throughput, with four seeks per cycle… 1 second + 60ms => 47MB/s into a single 50MB buffer. But if you had only 10MB buffers, now it’s 1 second + 60ms * 5 -> 38MB/s.

And that’s only due to the RAID. RAID helps quite a bit in straight line performance. If one drive can read at 100MB/s, two ought to be close to 200MB/s… as long as they don’t exceed the link speed, which as pointed out first, they don’t. And better still, since you’re reading twice as much data before seeking within a file, even seeks are less of a factors.

But when you’re not straight-line… go into seeking between files. When a RAID seeks, all drives have to seek. And the data can’t be returned until they all finish seeking. So the seek time of the RAID is the worst-case of all drives at any given time. It’s almost certainly a little slower at seeks than a single drive.

The big complaint about single-drive systems in media editing goes way back to the early days of multimedia. In these days, memory was tight, and you could very easily have bits of a program or chunks of your own data paging from virtual memory to/from disc storage. Programs do virtual loads, DLLs get demand loaded, etc. Lots of possible other stuff going on. Put your data there too, and now the data, application, and OS stuff are all seeking all over the place. Better to move the speed critical stuff elsewhere.

Ok, but it’s 2011 now, almost 2012. Some of this isn’t important anymore, some it. You shouldn’t be paging. Sure, VM is turned on, but you have enough memory (or you can, anyway)… I saw a deal a couple of weeks ago for 16GB worth of DIMMs for under $60. But programs can still do virtual loads, DLLs can still load on-demand, and there’s other possible, normal OS activity. I don’t think it’s a big deal anymore, but you definitely want a second drive. Or more… I have a “project” drive, which is actually a drive bay. Drop a SATA drive in there, and it’s a video project, a backup drive, some archival stuff, etc. But I digress.

If you have a Solid State Drive, this is far less important. And even if you don’t.. while I have multiple drives, I often render to C:, and that’ll definitely go faster (assets on D:, outputs on C:) in pretty much any case than rendering one drive back to itself.

-Dave

[Dave Haynie]

[andy abulafia] “Funny you should mention NT4… I’d be still running it if only it had USB support!

Naaa… it was missing too many things. Also, it did graphics in user-mode. Good for security and crash resistance, but kernel drivers made things go fast.

I went to NT as a nice middle-ground. I was mostly telecommuting to a job that ran its stuff on either OS/2 or NT. And I was trying to use the same PC for music.

But that’s actually what got me into Vegas. I had been running Windows something-or-other on my home PC. When I updated to NT, my applications actually worked ok (those that counted, anyway) but I lost my Turtle Beach sound card (Tahiti) and my MOTU MIDI interface.

Then I found that Pete Haller, independent software developer, had written drivers for both of these devices. Furthermore, the MOTU driver actually worked way better than MOTU’s drivers in Windows.

I read PC-DAW regularly, and offered my hardware system designer expertise as needed (pretty often, actually… PCs were terrible, terrible machines back in the day, and it was hard enough to keep audio and MIDI working, forget about video). One day, there was some big argument about audio bottlenecks in multitracking… Pete on one side claiming it was the hard drive, and nearly everyone else claiming it was CPU. And this was before plug-ins started sucking up CPU time. I replied with a verbal kotegaeshi, explaining things like seek time vs. buffer size vs. latency, etc. The other guys shut up.

Awhile later, I got comp versions of Vegas 1.0 and Acid 1.0. Actually, a very fine investment on Sonic Foundry’s part… I’ve owned nearly every version, and recommended them both, almost exclusively (I’m still up to date on Cakewalk Sonar, but Acid is usually my go-to, particularly given the “quick, I need music for video” nature of most projects these days).

[andy abulafia]
And rest assured – in 10 years time I will be clinging on to W7 like grim death!”

From what I’ve seen of Windows 8, I’m right there with you, buddy.

There’s a lesson in all of this, that Microsoft seems to learn, but only for one major revision of Windows. There are many reasons to make a new OS. But Microsoft has a big one: OEMs (HP, Acer, Dell, etc) want a new version of Windows every three years or so. Many users buy new PCs, and won’t venture an OS upgrade on their current HW.

But given that, they have other motivations. The one I want to see: a really good reason for me, as a user, to upgrade. In short, you give me something I want, that I don’t have now, that’s worth the upgrade price.

And then there are motivations I don’t care about. For example, building the Microsoft empire. Vista was all about that… changing APIs, adding in DRM, etc. Nothing much for me.. and many others felt the same way. Microsoft did learn from that, so Windows 7 was very much consumer oriented… even the $50 intro upgrade price.

Windows 8, so far, is all about Microsoft. And in fact, all about the fact that Microsoft failed to realize that both smartphones and tablets today as also personal computers. And likely to overtake PCs at some point. TVs are probably going to personal computers soon, too (I tried that one at a startup company called Metabox, back in the late 1990s). But on one really wants MS’s cellphone OS, and today’s Windows isn’t functional on a tablet without a mouse.. or at least a stylus. From all views, Windows 8 is putting a tablet/cellphone on Windows proper, in order to hopefully leverage the might of the desktop onto MS’s shrunk-down Windows for tablets and phones.

Here’s the thing.. even if this was the best cellphone/tablet OS in the history of the multiverse, it doesn’t belong on my desktop. Touch interfaces are put on highly mobile devices because they’re small and I’m travelling. Neither of these things are true when I’m at my desk, with my real Bella keyboard (Vegas edition), high precision mouse, and pen-based graphics tablet. I have no use for a substantially inferior interface device (eg, my fingers), nor a UI optimized to make that bad UI device usable.

-Dave

[Dave Haynie]

Yup. There’s definitely some serious production talent there. I can figure out how much of it was done, and probably do similar. Though I see a good week’s work or more there, just editing the cool video-wall effect.

The low-light stuff could well be a digital cinema camera. I could so that with my Canon 60D and a fast prime lens, for sure, but there’s no way on this planet I’d get that low noise and shallow DOF with my camcorders. The reason folks in most of our income brackets found HDSLRs to be revolutionary — even given so many video-hostile aspects to the use of such beasties — is precisely the fast that a ~$1000 device can product that class of video.

Of course, if I was regularly making video that good looking, I’d probably stop looking for a new job in electronics engineering and jump with both feet into full-time video! Like more everything else, you learn to make something that good by doing it worse on your first 1000+ projects.

As I’ve actually had a few video and photo jobs this fall for my current employer (the company is leaving the HW business), I’ve taken an exercise from my musical practice into video these days. When you’re learning to play guitar (and like most instruments, you’re either learning it or not playing it), you find something you can’t do and then just keep trying. Eventually you get close, maybe even right there. Same with video or any other artistic expression.

I’m never going to play “Purple Haze” like Hendrix (particularly on the Martin :-).. but I might learn new things trying to get that much closer. And as an engineer, I put a little of that engineering mindset into figuring out just how it was done. Did he use a weird tuning? I learned to play the Beatles’ “Yesterday” a few years ago, and it’s kind of difficult… I only recently leaned that McCartney (yeah, it’s him and only him, on an Epiphone “Texan” acoustic) down-tuned his guitar.. the song’s in F, but he was playing much easier note patterns in G. Try de-constructing the video, and figuring out how the various things might have been done. There are bound to be “magic tricks” like McCartney’s… only in video/graphics.

-Dave

[Dave Haynie]

[Jan Vozobule] “I have received some advice from Sony support
When you are playing back your footage, what is the preview quality you are using? Please try right clicking on the preview window and make sure that “Adjust Size and Quality for Optimal Playback” is NOT selected. Then try changing your preview quality to Good(Full), and then try playing back your footage. “

They’re doing what you’d normally recommend a person do to optimize their system. That’s good, and it’s the right answer from a support person.

What I’m doing when I benchmark is looking for the limits. So I have the display mode set to best possible, and the resolution 1:1 on-screen vs. the resolution (eg, probably 1280×720 or 1920×1080). Things will definitely go faster when you lower the quality, run at 1/2 or 1/4 resolution, etc.

And that can be a legit benchmark, too. But on a fast PC, you’re likely to get to full speed pretty soon, as you downrez and downgrade the quality. That makes it difficult to a reasonable system-to-system comparison… I can only know if someone’s faster than me if there’s some headroom. That’s why my own benchmarks were projects my system could barely preview (at full quality).

-Dave

[Dave Haynie]

[Tom Owen] “Dave, I tried ConCat/Split from JD Designs. It did the concatenation, but soften the files unacceptably. What was crisp, turned fuzzy. “

All that does is concatenate the files… it doesn’t even understand the file structure, much less change anything about the video. I claim it can’t make the video fuzzy — something else has to be wrong.

TSMuxer is also totally unable to change anything about the video, but it does re-mux the media stream. Which is also not a problem, as long as the end of file condition doesn’t trip it up. The real question I have for this merge operation is whether TSMuxer really handles abrupt EOFs properly.

-Dave

[Dave Haynie]

[andy abulafia] “Still… I am yet to be convinced regarding W7 … i need to do some research and figure out what I can disable to get rid of some of the unecessary fluff. Also, I used to be able to play my hi def videos using VLC – they now stutter on W7. Going from perhaps 15 processes (XP) to about 50 seems (W7) to be the price to pay for progress :-)”

VLC is a pretty awful video player. Don’t get me wrong… it has many capabilities, it’s open source, it runs independent of system CODECs. But even the latest has weak GPU acceleration, and it seems to have file I/O problems.

On Windows 7, you ought to see extremely smooth playback of pretty much any AVC file, even under the usually weak Windows Media Player. I can play back 1080/60p video under Windows 7 on my dual-core, 2.4GHz Core2 laptops, at full 60p, even in WMP. The reason here is Windows 7. First, it supports the DXVA 2.0 APIs introduced in Vista.. that’s video-specific GPU acceleration, which is way more efficient than using something like OpenCL or CUDA for this kind of work. And second, Microsoft added their own H.264 decoder in Windows 7 (which VLC doesn’t use… it’s using x264 I think), which uses DXVA for acceleration.

Even better, something like Splash Lite will give a much better looking video playback. Now sure, if there’s some special reason you need VLC… RTP streaming or something, then maybe you just need VLC. The latest does seem a bit better in Windows 7 than it did a year ago, so make sure you update.

Processes themselves take up memory, but they don’t use CPU unless they’re actually running. This is true of all modern operating systems. Most system processes are waiting on something, and sitting on a wait queue somewhere until that thing happens. They don’t burn CPU. I upgraded my desktop directly from XP to Windows 7, and found Windows 7 faster and superior in all ways. That’s me, but I’m sensing you haven’t really given it a real-world trial.

When I upgraded from NT 4.0 to Win2K, and Win2K to XP (in theory it was this same machine, though I’m not sure, with over 12 years of upgrades, there are any original pieces left), I always installed the new OS on a separate partition. I bought three copies of Win7 at the $50 special upgrade price, and put one immediately on my laptop — it shipped with Vista, so really, could it get worse? It didn’t, and as a result, it was a no-brainer: XP had to go.

-Dave

[Dave Haynie]

Ditto here… my Samsung DLP definitely cuts off a bit. My LG LCD less so, but I did tweak it. And yeah, doesn’t make any sense… a digital display is support to have 1:1 pixel correspondence with the input video (well, at least at 1080i/p). There’s no picture tube aging, either, so what’s the point of overscan? I dunno, but it still seems to be there. I suspect there’s a way to get full scan, but unless this is for purely personal use, you’re going to have to accept that some people are still going to be chopping off your edges).

-Dave

[Dave Haynie]

I always build my own system. No problems with that, IF you’re comfortable, and/or have the tech resources to do a bit of system-level debugging, if it comes to that. That’s far less common than it used to be, but it can still happen. But it’s definitely the way to get what you want, and at least as you go high-end, a good way to save money IF you have the time and expertise. Or just refuse to settle.

My system is about half the performance of the one you’re looking at (AMD PhenomII x6 1090T CPU, 16GB RAM, AMD HD6970 GPU) and it’ll tear through a track or two (composited) of Canon video (from my 60D, pretty much the same CODEC as the 5D Mark II) at full speed, even with a few FX on ’em. So you’re good to go with that system. The Canon CODEC is actually a bit less CPU intensive that regular AVCHD/AVCCAM video, since it’s IP only, not IPB…. I’ll see a little drop in frame rate on full quality previews compositing two similar AVCCAM tracks from my HMC40.

From the Canon, the bitrate’s about twice that of a typical camcorder, so you will win with a fast HDD. It’s not the raw bitrate, which is pretty mild for today’s HDDs, but the fact that, once you’re streaming a few tracks, your source drive will be seeking all the time, which has a very large impact on effective drive throughput.

-Dave

[Dave Haynie]

[Jan Vozobule] “If I understand it right, that means, that for my configuration it is better to buy more CPU cores than invest to better GPU card? Is it right?”

That is correct. This is pretty much always the case. GPU is not yet a general purpose enough computing resource to replace a decent CPU. Once you have enough CPU (not necessarily fastest on the market, but at least fast enough to let the GPU be useful), the GPU can help out in specific circumstances.

[Jan Vozobule] “Other thing that I don’t understand is that you are writing:
I get something like 5-6 1080p60 streams playing at full speed these days on my AMDx6 plus AMD HD6970

That means that with 6 footages in timelines simultaneously you get 29.75FPS?”

No. I have video at 1080p60 (59.94fps). Using a player application, I can play five to six of these at full frame rate. Dramatically faster than what I see in Vegas. This is somewhat of a cheat — there is a separate video decoding API, called DXVA 2.0, which is part of Windows. This gains some performance by graphics cards that can translate color spaces in hardware (eg, directly display YUV rather than translate in software to RGB), but it’s still very impressive, compared to what you can see in Vegas, even with the fast GPU.

[Jan Vozobule] “And at your benchmark you write that with 4x 720p footages you get 2.4FPS or 5FPS with HD6970?”

Right. For a single stream at 720/60p, I have no trouble playing at full speed, full quality. But I tried compositing four streams in two different ways, and both plain Vegas and Vegas with the GTX570 did 2.4fps. Vegas with the HD6970 did 5fps. I was really just looking for something that didn’t run full speed, after I first got Vegas 11, so that I could judge any speed-ups in compositing and video playback.

[Jan Vozobule] “Another thing is that I don’t really care about Sony’s benchmark testing videos because I have my own one 😉
“

Thing is, no one else has your benchmark. So there’s no easy way to translate your result to anyone else’s. The main point of any standard benchmark (and simply by posting it, Sony’s made their demo a standard benchmark) is comparison of system to system doing the same work. Since you have my numbers on the benchmark, if you run it on your system, you’ll have some idea of how much faster my system performs, at least on that specific benchmark. There are plenty of others, like RenderTest. There are a bunch of CPU and GPU benchmarks out there.

What do you do for an upgrade? What’s you budget? If your motherboard and power supply support a higher spec Phenom II processor, that’s probably the lowest cost upgrade. I see my processor, the AMD Phenom II T1090 (six cores, 3.2GHz) on NewEgg for $170… so that’s much less than buying a whole new MB + CPU, much less a whole computer. That’s a 125W peak CPU, though, so you must be certain your system can take it (your is speced at 80W).

Once you have six cores, though, you may start to find that 4GB of memory is limiting you. I have never found any Vegas activity to really need more than 8GB, and I ran with that for the last few years (I upgraded to 16GB last summer, I needed it for photo editing of all things… stitching 30-40 18MPixel, 48-bit photos together can take up a bit of RAM).

Your GPU is a pretty good one. First CPU, then RAM, I’d say. Now, it’s quite possible to get a high-end Intel i7 system that’ll about double the performance of my machine, if you have the cash. Of course, next year, that’ll cost half the price…

-Dave