|Encoding Speed Tests with Elgato Turbo.264 Hardware Encoder
First Posted: 6/13/2007
Updated 6/14/2007 for Apple TV setting tests w/CPU usage
Updated 6/18/2007 for iMac Core Duo 1.83GHz Tests
Updated 12/19/2007 for FYI on v1.2 Software Update
Updated 12/19/2007 for Apple TV conversion tests w/1.1 software
Updated 5/22/2008 for FYI on v1.3 Software Update
Updated 1/2/2009 for iMac 3GHz/Turbo.264 v1.3/EyeTV 3.0.4/OS X 10.5.6 tests
Updated 6/2/2009 for FYI on v1.4 Software Update
Turbo.264 v1.4 Software Update: ElGato has posted a v1.4 software update for their original Turbo.264. Their page's change list initially (fixed now) had the exact same features mentioned in the May 2008 v1.3 release (YouTube Upload, DVD Chapter Markers, Dolby Digital AC-3 Audio Passthrough for Apple TV Exports, QuickTime Broadcaster Support and bug fixes) - so I downloaded the ReadMe file for v1.4, which shows these changes:
"What's new in Turbo.264 1.4?
Improved Picture Quality: In many cases, encoded files are sharper due to improved deinterlacing. This reduces encoding speed somewhat, especially on slower Macs.
Advanced Options: Three advanced custom options have been added.
The "Dolby Digital" option preserves Dolby Digital AC-3 audio when encoding to Apple TV.
The "Chapter Markers" option creates QuickTime chapter markers from sources that contain chapter markers, e.g. DVD.
The "Closed Captions" option creates a QuickTime closed caption track from NTSC sources with closed captions.
Bug Fixes:Files encoded to iPod/iPhone formats no longer cause the 2nd-gen iPod touch to crash.
A problem where files encoded from EyeTV 3.1.1 to Apple TV, iPhone or iPod did not contain the sound track has been resolved.
QuickTime Player no longer plays back files with an incorrect gamma value.
iMovie projects that contained a title at the beginning no longer export at 1 frame per second.
Files exported from Final Cut Pro are no longer cropped.
Improved recovery after encountering a communication problem on the USB bus.
Exports to Apple TV now play on the PS3.
Audio/Video synchronization issues with certain source files have been resolved.
I welcome other reader feedback on Turbo.264 (include Mac Model Info/OS X version/Turbo.264 software version used) - let me know the results for posting here.
OS X 10.5.6/Turbo.264 v1.3/EyeTV 3.0.4/iMac C2D 3.06GHz:
(from 1/2/2009 mail)
"... After discovering that my new Turbo.264 was slowing down my exports, I ran some very quick tests using the EyeTV software on my iMac 3.06GHz Intel Core Duo (with 4MB DDR2 RAM). I was using EyeTV version 3.0.4 (4279) and the Turbo.264 software is version 1.3 (294) and all of this was running on OS X Leopard 10.5.6.
I recorded two programs in EyeTV - one of exactly 60 seconds length (53.6MB) and another of exactly 5 minutes (263.2MB). (I asked for details on those)
Source was 576i (Standard Def. PAL Widescreen) which is 720x576. The full stream info is as follows:
- Video: MPEG-2, 720x576, 4.5Mbps, 50/25fps
- Audio: MPEG-1, 48kHz, 256kbps
I then clicked on the "Prepare for WiFi access again" link for each and timed how long the export took with a stopwatch. I did this both with and without the Turbo.264 plugged in. The results were as follows:
EyeTV WiFi Export without Turbo.264: 0:24.5 (53.6MB File), 1:49.2 (263.2MB File)
EyeTV WiFi Export w/Turbo.264 Hardware: 0:43.0 (53.6MB File), 3:03.5 (263.2MB File)
As you can see, on the current top-of-the-range iMac the Turbo.264 actually slows down exports by around 40%.
(He later wrote)
I decided to compare times for the WiFi export first, as this is the setting in EyeTV preferences where they claim using the Turbo.264 would speed up export times. In fact, I only learned about the Turbo.264 when I came across this setting!
I have just run a test for the AppleTV export on my iMac (specs as described in previous email) for a 46.8MB file (video source 576i, as above) and got 1:06.8 without the Turbo.264 and 1:45.1 with the device, a slowdown of about 36%.
Regards, Richard C."
I've not compared times under 10.5.6, but IIRC my initial results (almost a year ago now) with EyeTV 3.0x converting EyeTV recordings (1080i OTA recordings) to AppleTV (default one-click conversion in EyeTV) were very similar in times (not much difference) on this 2007 AL iMac 2.4GHz. (Some earlier reports below included results from G5 towers that were faster without it also - biggest benefit is from lower-end machines, despite claims.) I still keep it connected though just for the lower CPU usage (as I generally do other things while converting in the background). But again not sure if 10.5.6 has made a significant difference in performance. Next time I record a show in EyeTV I'll do some comparisons with and without Turbo.264 connected.
(Earlier reports/results follow - using different machines/earlier software and OS X versions - biggest gains are from slower CPU systems. Also some notes on lower CPU usage for those multi-tasking (running other apps/tasks while converting video). Also keep in mind the 800x600 limit for Turbo.264.)
(First user report from 2007, using early v1.0.1 Turbo.264 software)
Elgato Turbo.264 User Review - real world numbers
Just got the Turbo.264 encoder, ran some basic tests... Here are the results, some interesting numbers! No hard science here, just a stopwatch.
All latest updates applied on all machines. (10.4.9 etc)
Turbo.264 Program version was 1.0.1 (Build 99), Runs were done twice, and most
of the time came out identical unless noted below.
Tests show machine, time to crunch, and final file size.
Quicktime crunch was done with pulldown "Movie to iPod" stock setting, or the "Movie to iPod (Elgato Turbo.264)" setting. This last one is added when you install the software that comes with the encoder. Crunch with the Turbo.264 program was done using it's "iPod High
Quality" setting. (Is that a higher quality setting than the std QT "Movie to iPod"?-Mike)
The files created with exports from quicktime, or the Turbo.264 program are both the same resolution, etc. You can downsize video using the Turbo.264 program, which is the iPod Low setting. Sets output to 320x240 I think. I just tested iPod High, which outputs 640X480. Again, doesn't matter if you use quicktime or Turbo.264, the files are the same rez, framerate, etc... Don't know if it uses 2 pass encoding... I hope so.
(I was wondering about image quality/bitrate settings, I assumed rez was the same. But the file sizes were not larger than the standard QT files, so I guess they use similar settings.-Mike)
Export to AppleTV or PSP settings were not tested as our needs are for simple podcasting.
The end file was equal in quality to the eye, and according to the "Get Info" option in quicktime, all final files were the same specs. (These machines below are the only ones I have handy)
Test File was a 2 Min Quicktime Movie, DV Format (MiniDV capture from FCP) 436.4 MB File with mix of action shots, static scenes and flashy graphics and titles.
Quad G5, 2.5Ghz, 8GB RAM, Saving to 50% full RAID 0. I tested also on the normal internal (boot) drive, and numbers were nearly identical.
Quad G5 2.5GHz:
- Stock Quicktime Export: 1:55 (23.5MB File)
- 1st run QT Export accelerated w/Turbo.264 Hardware: 2:49 (slower) (20.5MB File)
- 2nd run QT Export accelerated w/Turbo.264 Hardware: 2:10 (20.4MB File)
(unplugged all USB devices except for keyboard/mouse)
- Turbo.264 Program w/Turbo.264 Hardware: 2:49 (20.5MB File)
(did not run second test without other USB devices like the test before)
(On 2nd thought maybe it's really not surprising that a Quad Core 2.5GHz (although only 200% CPU usage per Kevin) beats a USB 2.0 interface hardware encoder, but the Core2 Duo results were not nearly as good as ElGato's example.-Mike)
iBook G4 1.2Ghz, 1.25GB RAM (Has USB 2.0 Port)
- Stock Quicktime Export: 10:58 (23.5MB File)
- QT Export Accelerated w/Turbo.264 Hardware: 2:17 (20.4MB File) (Money!)
- Turbo.264 Program w/Turbo.264 Hardware: 3:05 (20.5MB File)
(Slower than the quicktime hardware accelerated export?)
Core 2 Duo MacBook Pro, 2.16GHz, 2GB RAM, 70% Full HD.
- Stock Quicktime Export: 3:15 (23.6MB File)
- QT Export Accelerated w/Turbo.264 Hardware: 2:11 (20.4MB File)
- Turbo.264 Program w/Turbo.264 Hardware: 2:09 (20.4MB File)
I work for a TV station (Park City Telvision), and we got this in the hopes of speeding up some encodes so we could start podcasting some stories and shows.
First off, the time savings are nice, but not huge. I was a little surprised that my Quad G5 was faster then the hardware encoder. (Elgato's product page claims that Intel Core CPUs show up to 4x improvement, but that PPC G4 or G5s can show "exponential" improvements) Just the stock quad only used 200% of an available 400% CPU, so in theory, in quicktime, I could export 2 movies using the CPU, and a 3rd using the hardware encoder, and not take a big performance hit. But how often am I encoding 3 things at once?
Second, and more important. Lack of options for the export is a major bummer. I don't know if the encoder is hardwired to only export with its (limited) options, or if it's a driver issue, but file size is important, and the ability to tweak settings is crucial for us. Some things we would like to podcast work better @ 640x480 but 15fps, but others lower picture quality and size, but 30 fps is more important (like a sit down interview vs. an action sports clip). I hope that Elgato will release a better driver with the ability to use custom settings.
Another thing we would have liked was just to compress stock DV footage to equal quality and size, but h.264 for archiving segments and/or shows on the stations RAID. Saves a lot of space, and that
way we can dump that footage back into FCP if we ever needed it again quickly, and didn't want to re-digitize it off a tape. That is not possible at this time using the current drivers.
When you export using quicktime's stock "Movie to iPod" setting, the extension is .m4v, when you use the "Movie to iPod (Elgato Turbo.264)" it's .mp4, and using the Turbo.264 Program that comes with the encoder it's also .mp4.
When you use the included Turbo.264 Program program it adds whatever you just crunched automatically to your iTunes.
Also, it should be noted that even when using the hardware encoder, CPU usage is still around 20-50% of one CPU. (my quad showed 20-30% of a possible 400%, the MacBook Pro 50% of a possible 200%, and the iBook 30% of a possible 100%).
I did not test using the export to AppleTV or PSP settings, since the file sizes at the end are too large for our needs (simple podcasting).
If you have an older machine, this thing turns it into a speed demon. I wish I still had my old G4 350 with PCI USB 2 card to test how it would run on that. I suspect if you have old useless machines laying around, they can become instant encoding stations with this little device! Also, I would like to see how fast a quad core or 8 core intel MacPro can crunch my sample, for comparison.
So in summary, this is great if you have a original model G5, G4 or older machine, and as long as tweaking settings are not needed. Speed increases are not that significant on G5 or newer intel machines.
Great site by the way, been a loyal reader for years!
Oh one final note I forgot... we bought 2 of these, it would be nice to be able to "Shotgun" them, i.e. use both at once kinda like a multiprocessor for 2x performance.
(Mike's comments) Interesting - I asked Kevin if he had contacted ElGato support to see what they would say about his results. (A reader asked if Kevin could also run some comparisons to ffmpegX.) ElGato's Turbo.264 product page had an example comparison using a MacBook Core2 Duo system and mentions PPC CPUs (including G5s) should show higher gains:
Encoding time in minutes. Test conducted on a MacBook 2GHz Core 2 Duo with
QuickTime Player Pro 7.1.5. Test file: 10 minute DV 16:9 clip. Export
setting: Movie to Apple TV.
With Turbo.264: 11:18
(Here's a copy of the product page note on general performance:)
"Turbo.264 accelerates video encoding up to four times faster on Macs with
Intel Core processors. On an older Mac like a PowerPC G4 or PowerPC G5, the
acceleration is exponentially faster."
Their FAQ includes an item on slow encoding, but that item mentions things like using USB 1.1 ports (vs USB 2.0) for example.
If you've bought a Turbo.264 and have run some comparison tests, send them in for adding here. Thanks.
Updates: (after original posting above)
(A reader had asked Kevin to compare ffmpegx and after a Core Solo Mini/Turbo.264 owner's comments on image quality and playback issues (not smooth) with Apple TV settings I asked Kevin if he could compare them and he's included CPU Usage notes-Mike)
The ffmpegX tests will take a few days, kinda busy here. I did test the AppleTV export using the same machines and same file as
yesterday. Here are the results on that.
------ Apple TV Tests ------
- Stock QT - 3:10 (35.5MB file, 236% cpu avg)
- QT Accelerated - 3:30 (38.9MB file, 30% cpu avg)
- Turbo.264 program - 3:19 (38.9MB file, 30% cpu avg)
- Turbo.264 program, Different USB port - 3:27 (38.9MB, 30% cpu avg)
(His previous Macbook Pro tests w/iPod export showed only about 1/3 gain but the Apple TV export results show more delta)
2.16GHz Core 2 Duo MacBook Pro, 2GB RAM, 70% Full HD
- Stock QT - 5:36 (35.4MB file, 150-170% cpu)
- QT Accelerated - 2:13 (38.9MB file 30% cpu)
- Turbo.264 program - 2:11 (38.9MB 20% cpu)
iBook G4 1.2Ghz, 1.25GB RAM (w/USB 2.0 port)
- Stock QT - 19:47 (Turned my iBook into EZ bake oven.)
- QT Accelerated - 2:35 (30% CPU)
- Turbo.264 program - 2:39 (25% CPU)
Final file sizes on the iBook were inline with the others above, but I didn't write them down.... Again, the iBook loves this thing.
About Bitrates: The Stock movie is 30.33 mbits/sec, the export from the Turbo.264 program is 2701.70 kbits/sec, and the quicktime export (accelerate) is 2701.70 kbits/sec (same number) and the export from the stock quicktime encoder was 2445.30 kbits/sec. The hardware encoder was the same, give or take 10kbits across all machines, the quicktime encode was always around 2400, but was more like +/- 50-75 kbits. Again, to the eye they look the same.
Notes about Playback: The stock movie plays fine in everything I have to play it back, except VLC (.0.8.6b). In VLC the stock movie is all studdery, but VLC will play back the final encoded files fine and smooth.
I'll hop on the ffMpegX tests this weekend, but I'll only be able to test on the Quad G5 and the iBook, as my mate who owns the MacBook Pro is wary of installing it. (You're right, it's a little complicated.)
(After seeing the results of Kevin's tests on the Quad G5, I asked Nick of ElGato support if a Dual 2+GHz G5 owner for instance should expect any benefit from the Turbo.264-Mike)
"Question - is it reasonable to expect Turbo.264 to outperform QT encoding on a say a dual 2+GHz G5?
Yes - for example on the Dual 2.3GHz G5 that I'm using right now, there is clear acceleration.
So, a typical 3 minute 720 by 480 MPEG-2 music video might take 9-10 minutes when using EyeTV alone, but only 5-6 minutes with Turbo.264, on my particular Mac. Other Macs might have more of a difference in times. (some variation due to source content is expected, but the results of Kevin's tests on both iPod and Apple TV exports showed his Quad G5 was faster using standard Quicktime exports than using the Turbo.264.-Mike)
Times will vary slightly depending on the source material, the export preset, and the particular Mac being used. No absolute acceleration figure, like 3X, is possible, due to all of the variables.
(he later wrote)
My export comparison was with the Apple TV export.
The Turbo.264 will offer clear acceleration for most Macs, particularly those in the low and mid tiers.
However, the fastest Macs available, like a Quad G5, will most like match or beat the acceleration the Turbo.264 can provide. Those 4 processors add up to more power than the chips in the Turbo.264 have available.
That said, when you export using the Quad G5, your processors will be highly used - making it hard to do other things at the same time. When you use Turbo.264, your processors won't be taxed, and therefore you can do other things on your Mac while the export is passed to the Turbo.264.
In the end, the Turbo.264 offers acceleration compared to all but the fastest Macs, and offers little to no processor overhead while exporting on all Macs.
Technical Support Manager
Elgato Systems LLC"
iMac Core Duo 1.83GHz Tests:
"Just ran a test on the Turbo h.264 accelerator.
iMac Core Duo 1.83GHz (ATi X1600/128MB), 1GB ram, 10.4.9
EyeTV 1.4GB 720x576 mpeg-2 file
- iMac with Turbo h.264: 7min 36sec
- iMac without Turbo h.264: 34min
Since i mainly use the Turbo h.264 to convert tv shows from the Eyetv format to my iPod this accelerator has saved me alot of time. A great bargain for $99.
FYI - as of mid-December 2007, the xlr8yourmac site specials page has the Turbo.264 for under $80, but check link for current pricing/availability.
AppleTV Conversion w/1.1 Software Update (G5 owner)
(I sent him a note that a v1.2 Turbo.264 software update is available now.)
(from Nov. 9th, 2007 mail - updated w/Dec. 19th comments)
ElGato Turbo264 @ 30FPS!
Hi Mike, I've been using ElGato Turbo.264 v1.1 software/hardware encoder to make all my Apple TV movies for awhile, and I thought I'd share with you the No. 1 reason I'm doing it. The encoder operates in higher than real time! Enclosed find a screen shot showing it at work on The Last Boy Scout @ 30fps! (System specs: Dual G5 2.3GHz, 8GB RAM, OS X Leopard, 1.8TB RAID boot volume, Nvidia GeForce 6600.) That means, for an average 4-5GB movie, that the encoder converts to Apple TV format in just over an hour, which is a tremendous time saver over HandBrake or any other method that I know of.
Plus, as you know, the encoder itself offloads the CPU load itself and frees up my G5's CPU for normal business. I'm not trying to sell anybody on the Turbo.264, just reporting to you what I consider to be very excellent performance from the combo of the hardware encoder and the newest software v1.1 from ElGato. I've never seen anything like this from any other encoder system except when you get up into the Avid Cinema etc type of setup.
(he later wrote)
Since I sent you that screen shot, I've taken a couple others, but can't seem to find them in this mess of a desktop right now except for the one enclosed. I have seen the decoder working with various films @ 36fps! That's the highest so far.
But isn't that a bit daffy for this thing? I find it's very strange, and it doesn't happen every time. There are times when it's stuck @ 22-24fps, but mostly it's right about 30fps these days. It wasn't always like this, by the way. I have no explaining that I can do, as it's nothing I've done to my system or the decoder itself.
One day it just started working at these crazy speeds, and I just thanked my lucky stars and have been enjoying it, and using it exclusively for Apple TV encoding. Why not? It does a 100 minute movie in about an hour, an epic 2-hour film in 1.5 hours, and 180+ minute films in about 2 hours. HandBrake would be labouring for many hours to do such feats, and its work is never as perfect as the ElGato turbo.264's work seems to be. I've yet to have it screw up a Video TS file that's properly prepared.
I use the latest version of MTR, R-14j Beta (have to be a donator to get the Beta, and it's now registered to your Mac and uses the CSS Libraries on top of that, which the developer provides with MTR R-14j, + your license, which is only good for 1 Mac!) and extract a Movie Only Video TS file from a DVD, which eliminates all 0-duration cells, encryptions of all sorts (ARccOS, RipGuard, whatever...), and makes it Region Free, in case you want to make a backup copy DVD also. That's the key to getting a perfect Apple TV Video TS file for the turbo.264 to work with...it can only make a copy of what it gets, ie if the Video TS file is still full of crapola you'll get the same back in your Apple TV movie, and it won't play worth a darn.
Gosh Mike, my Apple TV movie collection runneth over! I've got a dedicated 750GB external HD for my 450+movie iTunes movie library, it's just too big to have sitting on my RAID 0 volume, and I've found that FireWire works fine in terms of speed transfer when I want to watch a movie on my wireless network. There's about a 5-second delay while the Apple TV finds my library, then another 5-seconds or so to locate the right movie, and after that it's seamless play without lapses, pauses etc. It's the best thing since sliced bread, Apple TV that is, for movie watching, since it totally eliminates the traditional DVD player and all the hardware we used to need for movie watching.
That's the best formula for Apple TV encoding I know of, the latest MTR and the ElGato turbo.264 encoder, as it's simply foolproof and will make the perfect Apple TV movie every time. I hope those tips help out your readers, if they're having any issues using HandBrake or other software for that matter: get the turbo.264 as it's very nicely priced right now, and use the latest MTR by going to rip.different.com, and get hooked up with R-14j.
My system is still the same it's been since I got the FirmTek setup last year - ref:
Seritek 2eEN4 Enclosure/Seritek 2SE2-2 PCI-e Controller RAID Tests (G5 PCI-e system) article here.
Regards, David C."
OWC has the Turbo.264 for under $80 now (refurbs under $60) - I'm tempted to get one for use with this refurb. iMac (24in/2.4GHz) for converting EyeTV recordings for my AppleTV.