Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Faster FFMPEG conversion.

warrenl

warrenl

macrumors member
Original poster
Jun 23, 2015
32
4
Hi there.

I need some help or advice for a faster Mac. I have had all the Mac Pro versions from 1.1 to 5.1 then got a 5K i5 iMac. I am actually very happy with it.

I have now hit an issue. FFMPEG encoding is to slow. I need to encode 150 1080P videos per hour and convert 8000 high res photos to low res photos. I use this program, Jalbum, http://jalbum.net/en/ to make web galleries. it uses FFMPEG for video conversion.

For a sample set of 4 videos, and 280 photos, my 2014 3.5 GHZ i5 iMac takes 2 minutes 30 seconds. A 2015 4GHZ takes 2 min, 5 seconds.

I tried various PC, and a tricked out Win 10 box with a 6700K - i7 and 1080 GTX video card took 2 min 3 seconds.

Any suggestions how to get this to around 1 minute 30 seconds and still use MAC OS?
 
warrenl said:
Hi there.

I need some help or advice for a faster Mac. I have had all the Mac Pro versions from 1.1 to 5.1 then got a 5K i5 iMac. I am actually very happy with it.

I have now hit an issue. FFMPEG encoding is to slow. I need to encode 150 1080P videos per hour and convert 8000 high res photos to low res photos. I use this program, Jalbum, http://jalbum.net/en/ to make web galleries. it uses FFMPEG for video conversion.

For a sample set of 4 videos, and 280 photos, my 2014 3.5 GHZ i5 iMac takes 2 minutes 30 seconds. A 2015 4GHZ takes 2 min, 5 seconds.

I tried various PC, and a tricked out Win 10 box with a 6700K - i7 and 1080 GTX video card took 2 min 3 seconds.

Any suggestions how to get this to around 1 minute 30 seconds and still use MAC OS?
Click to expand...

Sell all of that and buy a 12-Core Mac Pro (Late 2013). I don't know how the GPUs will perform, but that CPU still rates highly on Geekbench for multi-threaded workloads.
 
warrenl said:
Any suggestions how to get this to around 1 minute 30 seconds and still use MAC OS?
Click to expand...

Do a hackintosh on one of the newer chips out there or as has already been mentioned pay some crazy money for as it now stands better than four year old tech in the trash can mac.
 
If all you need is just raw multi thread CPU power and MacOS. Mac Pro 5,1 should be the cheapest and easiest way to achieve.

A dual X5690 (or X5680 for lower cost) cMP can gives you around 30000 GB3 score. Which is way higher than the 5K iMac.
 
h9826790 said:
If all you need is just raw multi thread CPU power and MacOS. Mac Pro 5,1 should be the cheapest and easiest way to achieve.

A dual X5690 (or X5680 for lower cost) cMP can gives you around 30000 GB3 score. Which is way higher than the 5K iMac.
Click to expand...


I will look into that.
 
http://jalbum.net/help/en/Improving_performance

looks like this software takes advantage of the GPU for acceleration and CPU cores for multi-threading. so some of the above advice of a 4,1 or 5,1 Mac Pro modified to run 12 fast cores and a newer graphics card (or 2?) could be the answer. that could let you run 10 to 20 processes simultaneously. then you also need a lot of RAM and fast storage. 48 to 96GB RAM is that box is easy. I would set up separate source and destination volumes to maximize throughput with a RAID or SSD (or SSD RAID) at least on the sources side.

or just a fully maxed 12 core nMP if you have more money than time and inclination.

you are probably also well served by contacting the developer for their recommendations on the setup. please report back what you work out.
 
  • Like
Reactions: Flint Ironstag
thats all folks said:
http://jalbum.net/help/en/Improving_performance

looks like this software takes advantage of the GPU for acceleration and CPU cores for multi-threading. so some of the above advice of a 4,1 or 5,1 Mac Pro modified to run 12 fast cores and a newer graphics card (or 2?) could be the answer. that could let you run 10 to 20 processes simultaneously. then you also need a lot of RAM and fast storage. 48 to 96GB RAM is that box is easy. I would set up separate source and destination volumes to maximize throughput with a RAID or SSD (or SSD RAID) at least on the sources side.

or just a fully maxed 12 core nMP if you have more money than time and inclination.

you are probably also well served by contacting the developer for their recommendations on the setup. please report back what you work out.
Click to expand...
[doublepost=1480201538][/doublepost]"or just a fully maxed 12 core nMP if you have more money than time and inclination."

I just tested my sample set of 4 videos, and 280 photos on a fully maxed out 12 core nMP - I tested it on their computer http://www.kijiji.ca/v-desktop-comp...-2-7ghz-12-core-1-tb-ssd-64-gb-ram/1217607452 and it was 2 seconds slower than my iMac using the same files and settings. It did show the 24 cores available. So i5 iMac 2:25, maxed out nMP 2:30 and maxed out i7 iMac 2:05. My gut feeling is that the i5 and i7 where faster on the video compression and the zeon was faster on the photo compression.

Just giving my feedback and flow up.

I did speak to the developer and the program uses FFMPEG for the video compression.
[doublepost=1480202737][/doublepost]Out of curiosity my 2015 2.5GHZ 15" Macbook Pro was 2:32!. I thought it would be slower than that.
 
IHelpId10t5 said:
I understand that it's not a macOS solution, nor is it local, but if you are looking for the ultimate scriptable solution for a large conversion project, have a look at Amazon Elastic Transcoder (https://aws.amazon.com/elastictranscoder/).
Click to expand...
[doublepost=1480217273][/doublepost]Thanks @IHelpId10t5. I have to use a specific program, Jalbum, which uses FFMPEG for video encoding. I do event photography and video at different locations and most of the time have no internet facilities.
 
You did adjust jAlbum preferences to allow it to use more cores and the GPU? I would think its default setting would favor the iMac where changing the settings may favor the machine with more cores. I'm not surprised an iMac would seem faster given identical settings.

A lot of apps use FFMPEG under the hood, just slapping a GUI on top. Handbrake uses all 8 of my 3Ghz nMP cores and its about 20% faster than running on all the cores in my 2.6 i7 rMBP. Fast SSDs make a big difference, one with the source files, second as the destination.
 
The files are in a 16TB Thunderbolt2 raid5 drive. The setting on the FFMPEG is "0" which should enable all the cores.

i'm just surprised that the i7 retina iMac is way faster than the 12 core nMacpro . I need to use a mac, as I use mac os server, and I believe it does not run on a Hakintosh.
 
warrenl said:
[doublepost=1480201538][/doublepost]"or just a fully maxed 12 core nMP if you have more money than time and inclination."

I just tested my sample set of 4 videos, and 280 photos on a fully maxed out 12 core nMP - I tested it on their computer http://www.kijiji.ca/v-desktop-comp...-2-7ghz-12-core-1-tb-ssd-64-gb-ram/1217607452 and it was 2 seconds slower than my iMac using the same files and settings. It did show the 24 cores available. So i5 iMac 2:25, maxed out nMP 2:30 and maxed out i7 iMac 2:05. My gut feeling is that the i5 and i7 where faster on the video compression and the zeon was faster on the photo compression.

Just giving my feedback and flow up.

I did speak to the developer and the program uses FFMPEG for the video compression.
[doublepost=1480202737][/doublepost]Out of curiosity my 2015 2.5GHZ 15" Macbook Pro was 2:32!. I thought it would be slower than that.
Click to expand...

Did you optimize based on the details in the link? are you monitoring CPU usage? Memory usage?
 
No, I was not able to monitor CPU usage. I did monitor Activity monitor and it went up to 300% which is what I was seeing on the i5 and i7 iMac.

I'm testing with a six core nMacpro tomorrow. Any suggestions how to monitor the individual cores?
 
  • Like
Reactions: thats all folks
http://jalbum.net/help/en/Improving_performance

set hardware acceleration to : on (that uses the GPU)
set number of threads to: (the number of cores)

in Activity Monitor, under Window is CPU History. they will be a graph for every real and every hyperthreading "core". Also, in Activity Monitor, under Memory, keep an eye on the memory pressure.

There will be some balance of the number of threads (either more or less than the actual number of cores) and the maximized use of resources. if you did your test on the 12 core using only 1 thread, you left a lot of potential sitting idle.
 
Hi

I am talking with the developer. The "docs" discuss photo processing performance, which is maxed out. The 12 core flew through the photo processing, but was slower in the video processing.

According to the developer, there is a -threads option for ffmpeg, which is set by default to -threads=0, so all cores should be used.

If you look at the program, the video setting are under Settings->Video->Advanced ? That's the only settings that are used when converting videos.
 
warrenl said:
Hi

I am talking with the developer. The "docs" discuss photo processing performance, which is maxed out. The 12 core flew through the photo processing, but was slower in the video processing.

According to the developer, there is a -threads option for ffmpeg, which is set by default to -threads=0, so all cores should be used.

If you look at the program, the video setting are under Settings->Video->Advanced ? That's the only settings that are used when converting videos.
Click to expand...

I don't have the program, in fact I never even heard of it until this thread. but I was curious

if FFMPEG uses Intel Quick Sync in this situation, then encoding on an iMac/MacBook Pro will be faster. not a feature that most Xeons have so that leaves out all Mac Pros. your best bet is the fastest, newest comaptible CPU you can get.
 
Have you considered getting two or more headless render boxes onto a networked storage and running parallel renders ?
There is a performance cap when rendering video (you can see an exemple here in regards to Premiere Pro https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pro-CC-Multi-Core-Performance-698/).
You might be better served having a few relatively cheap machines running quad core i7 CPUs rather than one expensive box where performance per dollar isn't that great.
 
Interesting thread. If activity monitor shows that you're not using all available cpu resources, you may want to try setting up a way to run multiple instances, such as multiple VMs. Your single-job throughput time may take a small hit, but you'll still get more work done in the same amount of time.
 
initialsBB said:
Have you considered getting two or more headless render boxes onto a networked storage and running parallel renders?
Click to expand...

Yeah, that's what I was thinking, too. Although it's curious to me that the nMP wouldn't trounce the iMac (unless we're talking single-pass w/ QuikSync on iMac, of course) if running at full song, doubling down on hardware would seem to be an easy way to achieve the desired timeframe. If your 2014 iMac can do the test job in 2m30s, then adding a 2nd iMac would allow you to do 2 jobs in 2m30s, or effectively 1m15s per job completed. This would exceed your 1m30s target.

But as "thats all folks" mentioned, I would make sure you're getting full core utilization whatever you're using...
 
Just for the records, as this thread is old now: I think the "-threads 0" flag doesn't activate the usage of all the cores. This is from ffmpeg's help:

-threads <int> ED.VA.... set the number of threads (from 0 to INT_MAX) (default 1)

In a quick test on my machine, "-threads 0" only uses around 100% CPU, while "-threads 8" uses around 625% CPU (which is 82% of my quad-core processor). So, in my tests the speed is about:
-threads 0 => speed=0.266x
-threads 8 => speed=0.66x

Just make some tests and monitorize using Activity Monitor.app
 
julifos said:
Just for the records, as this thread is old now: I think the "-threads 0" flag doesn't activate the usage of all the cores. This is from ffmpeg's help:

-threads <int> ED.VA.... set the number of threads (from 0 to INT_MAX) (default 1)

In a quick test on my machine, "-threads 0" only uses around 100% CPU, while "-threads 8" uses around 625% CPU (which is 82% of my quad-core processor). So, in my tests the speed is about:
-threads 0 => speed=0.266x
-threads 8 => speed=0.66x

Just make some tests and monitorize using Activity Monitor.app
Click to expand...

Interesting finding. I never really think about of using threads=0. In general, if want all cores to work, just don't add that parameter.

And even more interesting. 525% more CPU usage only make the process 150% faster, very non-linear.
 
Badly explained on my side. My local copy of the ffmpeg binary defaults to 1 thread, as stated on the docs. Maybe that one thread uses a tiny amount of all the cores (1/8 of each one) :)
 
Anyway, just in case someone care (and missed) this news. cMP can now use GPU hardware encoding in FFMpeg.

https://forums.macrumors.com/threads/activate-amd-hardware-acceleration.2180095/

HEVC to H264 hwaccel.png
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back