My new 12-core Mac Pro is almost ready (it’s been copying files from my other machine for nearly 24 hours now). I re-ran a MPEG Streamclip vs. Compressor test and was pleased to discover Compressor is now insanely fast because it can be configured to use all of the cores. MPEG Streamclip doesn’t seem to multithread properly to use all 24 virtual cores when only batching up a few clips. I set it to 4 instances, and combined processor usage was around 270% (out of 2400%). Running the Compressor job against a local QuickCluster resulted in CPU usage illustrated in the screenshot above.
My new 12-core Mac Pro is almost ready (it’s been copying files from my other machine for nearly 24 hours now). I re-ran a MPEG Streamclip vs. Compressor test and was pleased to discover Compressor is now insanely fast because it can be configured to use all of the cores. MPEG Streamclip doesn’t seem to multithread properly to use all 24 virtual cores when only batching up a few clips. I set it to 4 instances, and combined processor usage was around 270% (out of 2400%). Running the Compressor job against a local QuickCluster resulted in CPU usage illustrated in the screenshot above.
Didn't I ask you not to post benchmarks so I would not be tempted? Thanks alot Eric.
Congrats on the new machine and glad the cores are working...
I recommend one instance for each pair of cores. If you have 12 cores, then 6 instances should maximize your results.
Tea, milk and 2 sugars please.
I recommend one instance for each pair of cores. If you have 12 cores, then 6 instances should maximize your results.
That's what I have it set at (6). Although... each core is virtualized into 2, so the OS things I have 24. Should I set it for 6 or 12?
I am a bit confused, maybe it is the semantics....what do you mean by instances. I installed the 12 core with the Nvidia 4800x card this weekend along with 20 gigs of ram. It is a general rule of thumb to have at least one gig per core. But I am unsure what you guys are meaning since I am not a computer tech. Could you explain what and how?
Much thanks,
Steve
ps. Apple did something in their OS and NVidia should be coming out with new drivers next week, till then don't expect the blazing speed the card is supposed to deliver with CS5....they told me so.
I am a bit confused, maybe it is the semantics....what do you mean by instances. I installed the 12 core with the Nvidia 4800x card this weekend along with 20 gigs of ram. It is a general rule of thumb to have at least one gig per core. But I am unsure what you guys are meaning since I am not a computer tech. Could you explain what and how?
Much thanks,
Steve
ps. Apple did something in their OS and NVidia should be coming out with new drivers next week, till then don't expect the blazing speed the card is supposed to deliver with CS5....they told me so.
In QMaster Preferences you can set the number of "instances" that are used. Each Core can set up two instances, so my Octo allows me to go up to 16 instances that can be used for multi processing of encodes. The 12 Core can go to 24. Cranking them up to the highest number (16 for the Octo and 24 for the 12 core) usually hits the point of diminishing returns where the trade-off is not worth it and/or declines. Seems to be about 1/2 is the sweet spot.
So the question is.. is it 1/2 the actual number of physical cores, or 1/2 the number of physical + virtual cores for hyperthreaded processors?
Depends on the job, Eric. With the Nehalem processors, I actually find treating the virtual cores as physical cores can steal a couple of % of performance. However, it slows down once you use all cores. If you watch how compressor segments the work, it throws out double the number of threads per instance. SO long story short, 1/2-3/4 of all cores (physical and virtual) is the sweet spot.
Then it depends on the i/o speed of your volumes used. Compressor slows down when it segments the job and has to reassemble the segments. However Mpeg StreamClip still does 1:1 clip speed and Compressor (because of the segmenting) can't keep up. Have you found Compressor is faster for batch processing multiple files overall? Have you tried open -n in terminal for Mpeg Stream Clip or 5D2RGB? You can use all your cores, although it'll be a little more time consuming due to having to open each file manually etc. No free lunch.
Eric, How much ram have you installed in your 12 core and what graphics card are you using?
Steve
20GB RAM and ATI Radeon HD 5870.
Same here, 20 gigs and the Nvidia FX4800 card. Waiting for the new Nvidia drivers to come out next week.
Steve
Eric/Steve
Have not looked into the video cards for the new machine yet. Someone is about to buy one and was leaning to the 5870. Figured it is good for the upgrade price. They are not heavy into 3D or gaming, but figured the boost for things that are and will come down the road for card processing makes it worthwhile. Mostly running Adobe print programs (Photoshop, Illustrator, etc.), Final Cut Studio (mainly the NLE) and Aperture. Any thoughts or specs you have on that?
Have you found Compressor is faster for batch processing multiple files overall? Have you tried open -n in terminal for Mpeg Stream Clip or 5D2RGB? You can use all your cores, although it'll be a little more time consuming due to having to open each file manually etc. No free lunch.
I've attached AppleScripts to open new instances of MPEG Streamclip and 5DtoRGB (assumes 5DtoRGB.app is located at /Applications/5DtoRGB/5DtoRGB.app).
Oh, boy. On my 12-core machine, A single instance of 5DtoRGB uses less than 2 CPUs (virtual -- out of 24) worth of processor. Converting H264 from a 5D Mark II to ProRes 422 progresses at around 5 frames per second, or 16.7% real time. Even if they are committed to quality, speed matters. You can speed up a program without changing the quality of its output.
Eric/Steve
Have not looked into the video cards for the new machine yet. Someone is about to buy one and was leaning to the 5870. Figured it is good for the upgrade price. They are not heavy into 3D or gaming, but figured the boost for things that are and will come down the road for card processing makes it worthwhile. Mostly running Adobe print programs (Photoshop, Illustrator, etc.), Final Cut Studio (mainly the NLE) and Aperture. Any thoughts or specs you have on that?
For Macs there is the FX4800 or the 5800 which is a lot more expensive for diminishing returns. They also have the FX38 and 58 but you don't want them. I am happy with the FX4800 which is somewhat better than the graphics card that came with 12 core. Never the less, the one shipping with the 12 core is pretty darn good. I took it out and put in the 4800 only because Nvidia gave it to me to be part of a review so the original card is just sitting lonely in its box. Remember that FC is still 32 bit and can only access 2.5 gigs of ram. The real benefit for all this is for those who use Premier Pro (which I don't) and the other 64bit apps that Adobe came out with in CS5 and the Mercury engine. I use After Effects more than any of the Adobe apps though I am no expert on it.
There are rumors of FC's next version being 64 bit but they will have to write the app from the ground up since it still has the same 32 bit foundation that it came with when Apple bought the original company out and renamed it. I won't hold my breathe on this one. I think it will be at least 2 years before we see a 64bit FCP but the new 12 cores will be able to work with it just fine once it does make its appearance.
Steve
Great info, Steve. I may learn After Effects just to take advantage of the new machine. I am no Motion expert, so switching over would probably be painless, and I don't really use FCP/Motion round trip, anyway. A new 64 bit FCS is sure to be buggy when / if it comes out!
Oh, boy. On my 12-core machine, A single instance of 5DtoRGB uses less than 2 CPUs (virtual -- out of 24) worth of processor. Converting H264 from a 5D Mark II to ProRes 422 progresses at around 5 frames per second, or 16.7% real time. Even if they are committed to quality, speed matters. You can speed up a program without changing the quality of its output.
Well it is a beta and they did say they are throwing out speed to get quality. I like the output, esp the ability to play with YCbCr values and the use ITU 601 or 709. The processing does give very high quality files. The chroma correction is really good. I'm thinking Rarevision is going after the feature or print to film guys.
A new 64 bit FCS is sure to be buggy when / if it comes out!
NDAs suck!
Doesn't this make your cylinders hot?
Alex -- I've been really impressed with this new machine. The temperatures inside the box really do not go up much with the CPUs are all maxed out. I haven't taxed the GPU much yet, however. I'm sure I'll hear the video card fan when it goes...
