My experience with 12-core 2.66GHz Mac pro & Quadro 4000

[Andrae Palmer]

System configuration:
Davinci Resolve v7.0.3
Mac OSX 10.6.6
2 x 2.66 GHz 6-core Intel Xeon
24GB RAM (6x4GB ECC in triple channel config)
ATI Radeon HD 5770 Slot 1
NVIDIA Quadro 4000 Slot 2
G-RAID 2TB media drive connected via FW800
Dual 30" Apple Cinema Display

Before running Davinci Resolve, didn't realize that CPU debayering of RED footage was possible at 1/2 Res and realtime. Wasn't even aware that this would be the method that Resolve uses. Was thinking debayering would be GPU based with the Quadro 4000. Terrible experiences in Redcine-X, Final Cut Pro and all apps that are not multicore optimized made me ignorant to the possibilities. Then again I've only had a 12-core for the past month and wasn't totally aware of what it's real capabilities are regarding debayering RED footage. Was pleasantly surprised to see this:

click for larger version


CPU load 1920x1080 timeline at Half Res. Good (add 4-5% more for 2K film timelines):

It's a pleasure to see all 24 virtual cores being utilized and the UI being very responsive even with all 4 scopes loaded on dual 30" Apple Cinema Display setup. Since then I've added all kinds of secondary nodes to push the GPU. Found out the only thing that gives the system a hiccup is 3+ nodes of blur. With realtime playback at 1920x1080 possible on a 12-core at "Half Res. Good". I would recommend going this route instead of a RED Rocket (especially if financially challenged). I've seen many on here tout their experiences with a hackintosh and GTX 480, etc. I don't really quite understand using playback settings at a 1/4 res or 1/8 res on a desktop computer. The main holy grail of Davinci Resolve is realtime performance. If you are grading RED 4k RAW... you should at least be able to preview it at HD resolution decently in realtime on a professional monitor.

My opinion is that if someone is using a hackintosh system then it should supersede the capabilities of a certified system. The pro's would be more expansion slots and access to faster GPUs. The con is that to get a professional hackintosh system you are looking at costs that start to enter the realm of certified systems. I just don't see the point of cost cutting just to get a 1/4 res or 1/8 res realtime playback with a superfast GPU.

I'm really excited at now having the possibility of being able to grade 4k footage on a 4k res timeline. Wow! Look how fast the world of tech is moving. Thank you Blackmagic! What your company is doing in the world of post is just as important as what RED s doing with the camera.

 

[Aleksandar Colancevski]

Maybe my eyes are not OK, but I find very difficult to make difference between 1/2 and 1/4 debayer on JVC HD projector.
Strange that my 285gtx is little faster than quadro4000. I can do 4 nodes with blur.

 

[Andrae Palmer]

That doesn't surprise me... some people can't see the difference between HD & SD. I on the other hand can clearly & easily see a difference between 1/4 res and 1/2 res. The difference is even more acute in formats where the viewers eyes will be right next to the broadcast medium. The effect is akin to having a blur node on the entire clip especially with up close shots.

 

[Aleksandar Colancevski]

I was refering to significant difference that could affect your grading.

 

[Christoffer Glans]

Maybe my eyes are not OK, but I find very difficult to make difference between 1/2 and 1/4 debayer on JVC HD projector.
Strange that my 285gtx is little faster than quadro4000. I can do 4 nodes with blur.

Not strange, the geforce cards are more powerful then the quadro, it's just that they are tweaked and optimized for Open GL, which the geforce isn't.

As for the 1/2 and 1/4 I clearly see a difference, sharp edges are jittery.

 

[Andrae Palmer]

Notice the loss of detail in the headboard texture and how it affects perception of brightness... same goes for the eyes. The jaggedness on the fingers and lines on the neck.

Well any colorist can grade in 1/16th res and it wouldn't affect the grading (especially for primary grades)... the situation switches slightly when a client is wondering why their very sharp footage filmed on a 4k RED camera, etc... now looks so verry blurred. The situation is even more disastrous when the footage has to be output to tape realtime.

 

[Aleksandar Colancevski]

Very nice example. It was my eyes.
I render it after with full debayer, but never noticed that I mess it up with brightnes.
With BT Rocket Hack comes 1000$ cheaper than MacPro 12core and you don't have to buy expansion box.
Asus P6T7 Supercomputer 460$
i7 930 250$
Memory 12GB 300$
Cooler 50$
Case 100$
285gtx 300$
GUI card 50$
HD 50$
BT Rocket 3000$
Total cca 4700$ and you will still have an extra PCIe to put another 285gtx or whatever.

 

[Jeetu Chawla]

Andrae, are you using the Apple Cinema Display as your Grading monitor? How would you rate it - precision wise?

Thanks

 

[Jose Lomeña]

Do you render at 1/2 or Full?.

In my opinion 1/2 Standard get more "sharpnest" than full debayer. Can you make the test with the same image?.

Notice the loss of detail in the headboard texture and how it affects perception of brightness... same goes for the eyes. The jaggedness on the fingers and lines on the neck.

Well any colorist can grade in 1/16th res and it wouldn't affect the grading (especially for primary grades)... the situation switches slightly when a client is wondering why their very sharp footage filmed on a 4k RED camera, etc... now looks so verry blurred. The situation is even more disastrous when the footage has to be output to tape realtime.

 

[Andrae Palmer]

Very nice example. It was my eyes.
I render it after with full debayer, but never noticed that I mess it up with brightnes.
With BT Rocket Hack comes 1000$ cheaper than MacPro 12core and you don't have to buy expansion box.
Asus P6T7 Supercomputer 460$
i7 930 250$
Memory 12GB 300$
Cooler 50$
Case 100$
285gtx 300$
GUI card 50$
HD 50$
BT Rocket 3000$
Total cca 4700$ and you will still have an extra PCIe to put another 285gtx or whatever.

The problem with the above configuration is rendering out to compressed file formats.

"Compressed formats may or may not be processed in realtime depending upon the codec used. Compressed file formats place higher demands on the CPU than uncompressed formats. Of course uncompressed file formats require a fast disk array to cope with the high read and write data rates.

If you are regularly using compressed file formats, a certified 2009 or newer series Mac Pro is recommended for realtime performance."

http://forums.creativecow.net/thread/277/1739

Andrae, are you using the Apple Cinema Display as your Grading monitor? How would you rate it - precision wise?

Thanks

Just using the ACD to play around currently while studyung the manual and researching the software online. Today ordering the BM Decklink 3D+ and will then try to find a proper grading monitor. I'll have to do even more research in that quest... price being the main issue. Would love to get a 10 bit monitor as soon as possible.

Do you render at 1/2 or Full?.

In my opinion 1/2 Standard get more "sharpnest" than full debayer. Can you make the test with the same image?.

I render out to disk at 1/2 Standard, can't see a difference between the two... your perception could be right, let me look into it today.

 

[Eric Santiago]

Thanks!!

Thanks!!

The SuperBowl aftermath has not been good to my stomach.
However, reading this post picked me up again BIG TIME!!
Thanks Andrae for that
Quick question, do you have a RED Rocket at all and if you do, does it add to the experience of the 12Core at all?
Running a MP 2.97 12Core with dual ACD 30's, Quadro 4000 and RED Rocket.
Its our REDCineX box but researching if were good enough for Davinci (without anymore additions).

Thanks again!

 

[Andrae Palmer]

Sold RED Rocket card a few months back... wondering the same thing you are. If rendering out involves this process: First Debayer then apply GPU effects and finally compressing to codec via CPU. Then having the RR of course would eliminate process one. But lets look at it this way... this is a serial cascading process... the program has to wait on the RR to finish the debayer first. So in that regards having the RED Rocket card may not make a difference at all since the Mac Pro can also process the debayer in realtime at 1/2 Res Good. So Mac Pro 12-core process without RED Rocket card would be... CPU debayer in realtime at 1/2 Res. Good, GPU effects and then CPU compressing out to codec. It looks like the same process to me... but I'm not fully aware of any potential downsides. Visually I cannot perceive any difference between 1/2 Res. Good and a full debayer. Of course it would make a world of difference in a 3K or 4K render out environment... the CPU would not be able to handle the debayer in realtime. Unfortunately my workflow is not so high end as of yet to need a 3K or 4K disk out.

Currently getting around 18fps out to a 2TB G-RAID with Apple ProRes 4444. Let me do some testing with ProRes 422 today. Perhaps others with a similar config as mine plus a RED Rocket card can post their compressed file formats render out results.

------

Apple ProRes 422 HQ... 18 FPS
Apple ProRes 422... 22 FPS

 

[Aleksandar Colancevski]

I don't understand what is the problem if I wait for Resolve to render.
Ofcorse if I could afford HDCAM or D5 I wouldn't be dicussing Hacks.

 

[Andrae Palmer]

We are comparing fast configs... a 12 core Mac Pro can CPU handle realtime debayer playback and significantly boost your render out FPS. The config you posted with a RED Rocket would be very limited in render outs. I think you would agree that realtime playback and realtime render out to disk would be ideal.

Very nice example. It was my eyes.
I render it after with full debayer, but never noticed that I mess it up with brightnes.
With BT Rocket Hack comes 1000$ cheaper than MacPro 12core and you don't have to buy expansion box.
Asus P6T7 Supercomputer 460$
i7 930 250$
Memory 12GB 300$
Cooler 50$
Case 100$
285gtx 300$
GUI card 50$
HD 50$
BT Rocket 3000$
Total cca 4700$ and you will still have an extra PCIe to put another 285gtx or whatever.

A 12-core Mac Pro is on B&H Photo for $4799, free shipping and no tax. That's how I got my system. The mobo you listed are single slots? Two GTX 285's would take up four slots... the BT rocket would take up the 5th slot, and the GUi card the 6th slot. I could do the same by getting a GT 120 and installing two GTX 285's. The RED Rocket would be eliminated because of the realtime CPU debayer at 1/2 res. good. I would have the benefit of being able to do realtime renders out to compressed file formats while enjoying the same realtime playback as the hack. My system would be certified and have much more of a higher resale value.

 

[Christopher Barrett]

Very interesting, Andrae. I have the same 12 core with 48gb ram and internal raid, but haven't been able to see anything at higher res because of the lack of multi-core apps. Do you know what, besides Resolve, makes good use of the power?

I was thrilled when i got my system and started processing stills. Capture One Pro fully utilized 24 threads, processing 60megapixel raws into 335mb tiffs wicked fast. Good stuff!

Ironically, I just ordered a Rocket this weekend.

CB

 

[Christopher Barrett]

Edit : post seemed to clone itself.

 

[Eric Santiago]

Very interesting, Andrae. I have the same 12 core with 48gb ram and internal raid..

Care to tell me which Internal RAID and HD set-up you have? Ive been researching for days now and still havent pulled the trigger.

 

[Andrae Palmer]

Chris,

Premiere Pro makes use of the multicore power for debayering realtime. After Effects saturates the cores for rendering. I'm on the hunt for other apps that both utilize GPU power and multicores. What I found out is that Mercury Playback Engine is not just GPU based... it's a technology that utilizes full system resources. It really made me feel happy in my purchasing decision of getting the 12-core 2.66 GHz system.

 

[Christopher Barrett]

Eric, just 4 drives in the Tower slots configured Raid 0 via system software. Not sure if a Raid card is worth giving up a slot, I'm getting about 275mb/sec on that... backed up to an eSATA box... backed up to a Drobo Pro. I've got two OWC SSD's in the optical bay in Raid 0 for system. Getting > 500mb/sec on those lil buggers.

Andrae, awesome. I've just recently switched from FCP to Premier and was happy just to be able to make use of the RAM, will play with debayer settings some more for viewing.

Coolio.

 

[Aleksandar Colancevski]

A 12-core Mac Pro is on B&H Photo for $4799, free shipping and no tax. That's how I got my system. The mobo you listed are single slots? Two GTX 285's would take up four slots... the BT rocket would take up the 5th slot, and the GUi card the 6th slot. I could do the same by getting a GT 120 and installing two GTX 285's. The RED Rocket would be eliminated because of the realtime CPU debayer at 1/2 res. good. I would have the benefit of being able to do realtime renders out to compressed file formats while enjoying the same realtime playback as the hack. My system would be certified and have much more of a higher resale value.

With 24GB ram for 4799$? Wow that is a steal.
Luigi managed to put 3x 470gtx and a Blackmagic card and a raid card.
I didn't try it, but hey there is always a water cooling solution which can make 285s occupy single slot.
Currently I am on 150$ Gigabyte x58 USB3 mobo with the two gtx card and one Blackmagic, rendering on softraid 0 with 4 drives. Doing realtime only in 1/4 debayer.
Waiting for Jose Lomena to report on his i7 980x 6 core prosessor if it is capable to do 1/2res debayer.