Another in my series of text blog posts:
In the last week or so I’ve been putting together the spec for a new desktop video editing PC, and I’ve learned a fair bit about what’s (supposedly) really needed for a video editing PC. Here is the story:
I currently do most of my video editing at home on a HP DV7 notebook with a 4 core 2GHz i7 2630QM processor, 16GB of DDR3 memory, and ATI Radeon graphics chipset, and a 7200rpm 2nd internal SATA hard drive. But I’m not using the ATI chipset (too many driver issues with Sony Movie Studio), just the internal Intel integrated graphics (which actually pretty good). I also run a 2nd external monitor for dual screen. The chipset is a HM65. In short, it’s not bad as far as notebooks go, and it’s pretty zippy at video editing. It has a PassMark CPU benchmark of 5643
However, my lab PC is an old dumpster dive 4GB single 7200rpm Dell XPS 420 Quad core Q6600 2.4GHz machine isn’t nearly as zippy at video editing, and I do occasionally do some editing at the lab. That PC has a PassMark rating of 2961
So the plan is to replace the home notebook with a new video editing machine, and move the notebook to the lab as the secondary editor.
And then the fanboy wars started on the EEVblog forum… :->
Actually the response was brilliant when I asked about specing this machine on the forum, everyone jumped into the action and there was much research and discussion. 18 pages worth in fact!
And that doesn’t include a thread before that when I asked about video rendering speeds.
The result of that original thread (and on Twitter) was the suggestion that I should switch to a Mac Pro and Final Cut Pro. To cut a long story short, the Mac pro is massively slow at editing. And even a $500 hardware H264 accelerator card I was looking at would likely only bring it up to the same speed on my current HP DV7 notebook. So much for moving to Apple, sorry to all the Jobs fan boys, there was just no major incentive to do it.
So, the decision was to stick with my Windows and my current workflow and get a new WinTel machine.
I haven’t bought a desktop for at least a decade, as I’ve found notebook more flexible, but really that’s not good solution for a proper video editing machine.
So many people ask about my video editing workflow, so here it is:
I shoot on a Canon HF G10 in 12Mbps 1440×1080 AVCHD. (Have just switched to 17Mbps 1920×1080 in the last few videos)
I shot everything in sequence if at all possible, even if it means moving the tripod 20 times and changing the macro lens 10 times. That saves a HUGE amount of time in the video editing process.
I try and get the audio level right in-camera, so there is usually not much need to tweak or level audio while editing, is possible.
I then direct edit these .MTS files in Sony Movie Studio Platinum 12 (not Vegas). Most of my editing simply involves trimming the start and end of each clip, with maybe some overlay text or 2nd video track if I’m getting really fancy.
I then render from Sony to AVC, using the Sony AVC CODEC at a fixed 12Mbps rate (to match the input). This give me a top quality but HUGE intermediate file.
An aside on Constant Quality
Why don’t I use the MainConcept codec and output compressed Mp4 directly? The reason is simple and IMO, very important, and this is a point most peope do not understand and are quite confused by. I want the absolute best quality H264 HD video quality with the smallest possible file size, for every video regardless of what content is in that video (fast moving images, stills, etc). The codec like MainConcept that have a variable bit rate do not allow allow this easily. You have to know what type of content you have, and know the optimum average and peak video rate for that type of video content. If you get it too high, your file size is too big, and if you get it too low, your quality suffers. If your video contains a big mix of stuff, you have to compromise.
Why do I care about file size? Simple, Youtube!
Youtube cap the upload rate to 1Mbps or under (who knows why), even on my 2.4Mbps Telstra cable. This means it takes forever to upload long HD videos to youtube. Something the average person doesn’t really worry about. But to someone who does this daily, it’s really important. Optium file size vs quality is everything.
This is why I do the intermediate step in Sony, and use a 2nd final step with a program called Handbrake. Handbrake creates fantastic highly optimised H264 files using a codec technology called Constant Quality mode. It actually uses the x264 codec, Handbrake is just the nice GUI and command line interface to x264. There are others available like MeGUI.
What Constant Quality mode does (in a single pass!) is to optimise the bitrate to give you a consistent quality, and hence the most optimal file size, regardless of your video content complexity. So there is no need to worry about anything, just pick the quality you want (I use an RF value of 22) and it does everything. MAGIC!
This is something that regular (even 2 pass) Variable Bit Rate (VBR) codecs cannot match. So if you are using VBR, your videos are likely not a optimal quality.
And there is no video editor on the market (that I am aware of) with a CQ codec option.
Back to the Workflow
So I pass the big rendered video file into Handbrake (I have a command line script on my desktop, so just drag’n’go) and it produces the final 1440×1080 (or 1920×1080) H264 video I upload to youtube. I call this process transcoding. I also use Handbrake to produce a 640×360 iTunes podcast version.
Sony MS on my notebook renders in better than “real time”, so a 1 hour video maybe takes 45-50min to process. Handbrake take about ‘real time” as well. So all up a 1 hour video takes me maybe 2 hours all up to render before I can upload to Youtube.
I Feel The Need For Speed
So “real time” actually isn’t bad for rendering in a general sense, so my current system is no slouch, but if I can double or maybe triple my rendering speed, then I might get to bed earlier. As I usually edit when I get back form the lab, which maybe takes a hour or two at most, and then I need to do that two step render process before I can upload to Youtube before I head to bed.
So can a new PC be that much quicker? We’ll see the end of this week when my new PC comes in…
Now, for all the gory details and fanboy fights, feel free to visit the 18 page forum thread. So here I’ll cut to chase of what seems like the general consensus from what people are telling me, and what I’ve researched myself.
You don’t need a GPU!
Joe Average will tell you that a hot shit video editing machine will need to blazing fast video card and GPU. But the fact is, Handbrake won’t use it, and while Sony Movie Studio can use it for rendering, it’s really designed for playback with a dozen different timeline items. I don’t have a dozen different timeline items, I have one video, one audio, and maybe one text. My internal chipset integrated graphic works just fine and smooth as silk in playback on dual screens, thank you very much.
Also, Sony Movie studio can use Intel QuickSync technology which is inside the latest i7 chipsets. My notebook has this, but I’ve tried using it and it’s actually slower than without it. But maybe my new machine will be different. Also, there is some question that using it can be lower quality than CPU only processing, but I have confirmed this. That’s one potential reason to chose Intel instead of AMD though.
You don’t need a Solid State Drive!
Joe Average will also tell you that a shit hot video editing machine needs an SSD drive, as the hard drive is going to be a bottleneck. Wrong!
Compressed video rendering speed is almost entirely is driven by the CPU, number of cores, and any associated CPU features that may help. Any old 5400rpm hard drive will easily keep up with the writing speed of rendered video output. You don’t need a fast hard drive or SSD! Although my machien will have a 7200rpm 2nd hard drive to write video too, and an SSD for the operating system, just for general machine speed.
You Don’t Need Much Memory!
Joe Average will also tell you that a shit hot video editing machine need lots of RAM. Wrong!
Some video editors may, but I can run 3 instances of Sony Movie Studio rendering 3 videos at once, along with a dozen Chrome windows open and still not get anywhere near my 16GB of memory. I can also run 8 instances of Handbrake transcoding 8 videos at once, with plenty to spare. 4GB is actually enough to do video encoding entirely in memory. Other video editors may vary of course! Video rendering works in small chunks, so generally not much RAM is required at all to process it.
But my new machine will have 16GB anyway, just because I can.
Intel vs AMD
Cue the fanboys. Both make their case using all sorts of benchmarks to show that one is better than the other, but in the end I relied on the only test that really matters, video rendering. And as it happens Toms Hardware has a nice comparison table using Handbrake for video rendering:
It turns out the sweet spot in terms of bang-per-buck is either the Intel i7 3770K, or the AMD 8350. The AMD 8350 is about a $150 cheaper solution give or take. Less than that in my case.
But the Intel i7 offer pretty hot integrated graphics, with QuickSync video encoding technology. So if you don’t need a video card, the slightly faster 3770K solution is essentially a similar price as AMD with a decent video card.
The 4 core Intel also beats the 8 core AMD (just, again) with the MainConcept encoder:
The higher end Intels didn’t really offer any huge bang-per-buck improvement, and crucially don’t have the fast integrated graphics or quicksync encoding like the i7. So the i7 was the obvious choice in the Intel range.
The 3770K has a PassMark of 9638, almost double my current notebook.
The Intel 3770K beats the AMD 8350 in terms of power consumption, 13W less in idle mode in fact, and a massive 82W when running full tilt.
That’s a lot! And more heat means more cooling required, and faster fans, and more noise.
So the Intel beats the pants off the AMD in terms of video rendering / watt.
My current notebook is real pain, the fan screams at the top of it’s lungs when I’m rendering, and for some time after, and it’s just 50cm away from my ears. Very unpleasant. So I decided my new PC must be designed with sound reduction in mind, even if that means over-engineering a bit in terms of cooling and PSU power.
Modern PCs can be very noisy. Stock CPU fans are notoriously noisy, GPU are very noisy, and cheap arse cases can be noisy in several ways. Cheap fans, hard drives can resonate, and poorly airflow ducting can have an impact too. It all adds up to a headache.
So I decided I wanted a name brand PC cased engineered properly to be Silent. It came down to a Fractal R4 or Cooler Master Silencio 550.
More fanboy wars here! It must have Japanese gold blah blah caps. Yeah, great, I agree, but not at double or triple the price in my case. I’m a tight arse, and my system is probably going to draw 15oW max or something from it’s 600W PSU. I’ll go for a relative cheapie and take my chances.
So what did I end up going with?
- Intel i7 3770K (K is the unlocked one you can overclock, about $50 more)
- ASUS P8Z77-V-LX Motherboard (cheapest one with the top line Z77 chipset)
- 16GB Corsair DDR3 1600 memory. 2 x 8GB, so room for extra if needed.
- Corsair VS650 650W PSU
- Corsair H55 liquid cooler.
- CoolerMaster Silencio 550 case
- Seagate 7200rpm 2TB drive
The case was decided because the Coolermaster had a very convenient top mounted SD card slot, and an integrated HDD dock installed. And it was $50 cheaper than the Fractal R4, just a smidge over $100. Both have very good “silent” reviews. The coolermaster has sound proofing on the side walls and front panel, air filters, and “silent” low RPM fans. Should be good enough.
I got upsold on the CPIU cooler. I only wanted a passive one with fan, but the water cooled Corsair H55 was only $10 more, and the box said 6dB better than the stock Intel cooler. Sounds good, and the reviews are good. Overkill maybe, but that’s a good thing. Less heat ultimately means less fan speed and less noise. It should also allow me to overclock the CPU. The H55 fan replaces the fan in the back of the case, so that’s actually one less fan in my system, that’s not a bad thing.
It was either an ASUS or ASROCK motherboard. I wanted the top line Z77 chipset, and the ASUS was cheaper, and had all sorts of wizz-bang stuff, so meah, ASUS it is. It has 4 DDR3 slots, PCI-E 3.0, plenty of expansion, and integrated HDMI and DVI video outputs.
Seagate was the cheapest 7200rpm 2TB drive, that is all. It’s the only thing I got from ebay, everything else was from my local computer store. I already had a 128MB Samsung SSD. I may even end up integrating my external RAID and network media hard drive into this thing (software RAID?). Comments?
All up the whole system cost me under AU$1K, and considering I’m a tight arse and have never paid over $1K for a computer in the last 25 years, I’m happy to continue to trend.
So how will it perform?
I don’t really know for sure, I’ll find out soon. But from all the research I’ve done, I think this is probably the best value video rendering machine possible. You can get maybe 25% faster at best, but it’ll cost you 4 times the price.
I’ll benchmark it without the video card first, and then out of curiosity will try a higher end GPU enabled video card and see if it makes any difference. I’m thinking it likely won’t. Unless you have gaming needs, or very serious multi channel wizz-bang effect video editing, then the integrated graphics will do a treat.
I expect it to be near silent, even when rendering for an hour. But we’ll see. It will certainly be a huge improvement on the laptop.
Power consumption should be <100W in idle, and maybe over that when rendering. We’ll see.
And the rendering speed? I’m expecting at least double the rendering speed in both Sony and Handbrake based on the PassMark CPU results. Likely more. If it’s less than double overall for the my workflow overall then I will be disappointed. I expect 2-2.5 improvement though, and that will be very significant.
Yes, I’ll do a build video and show the results, stay tuned.
People often ask how long it takes to produce a blog post like this. Well, I timed it! The post you just read took me 2 hours and 15 minutes to write and publish. That includes the gathering of links, thinking, etc.
And I don’t consider myself particularly slow at this kind of thing, but I’m not the most efficient either. Text blogging takes time, and writing takes time.
Don’t get me started on how long video blogging takes!