My DaVinci WorkStation Configuration .

[Vigen Vartanov]

Good day . I am planing to Update my Workstation with all accessories that i will need to work with DaVinci . Please comment on my configuration .

Processor :2x 3 GHZ Quad-Core Intel Mac - 3.1
Memory : 32 GB

Slots :
-------------------------------------------------------------
Slot 1 - Force 8800 GT (Only for DaVinci Interface )
Slot 2 - Cubix 16 CH 4 Slot - GPU-Xpander Desktop 4 (want to buy)
Slot 3 - Decklink HD Extreme 3D + (want to buy )
Sot 4 - Apple 4 Port Fiber Card
-------------------------------------------------------------

In Cubix :
----------------------
1x RedRocket Card (want to buy)
2x Quadro FX4800 (want to buy)
3x Quadro FX4800
-----------------------

Raid :
----------------------------------------------------------------
2x Easy Raid 5TB - Read Speed 352 / Write Speed : 318
----------------------------------------------------------------

Monitoring :

------------------------------------
Cinema Display 30 ' for Interface
Cinema Display 23' for WaveForms
Sony LMD 2050W - any comments for monitor ? is it normal for cheap configuration ? I own this monitor and want to use it and don not buy another monitor .
-------------------------------------

I want to get real time performance with 10-15 nodes and R3D footage , will i get good result with my future configuration ?
All parts that i need to add will cost me 8K $ .

Thanks .

[Jeff Kilgroe]

Don't use the Quadro FX4800. The GTX285 and Quadro 4000 are much faster. GTX285 is out of production, unfortunately. I don't know about where you can buy to get the parts in Georgia, but here in the USA, the Quadro 4000 now sells for about $765. The Quadro FX4800 is an older generation card and is also out of production. But there seems to be ample supply out there. They sell for around $1300 because people haven't figured out that it's actually older and slower than the new 4000.

You may run into memory bandwidth issues on this '08 Mac Pro. And 32GB is kinda overkill, but hey, if you got it. :)

Why the fiber card? Unless you already have fiber devices to connect to. I suggest an ATTO SAS card so you can connect more mainstream eSATA/SAS RAIDs and devices as well as an LTO tape if needed.

With two quadro cards and a rocket, you will get your real-time performance at 10-15 nodes. That shouldn't be a problem there. It will come back to two factors though. Once again, memory bandwidth in this system is quite a bit lower than on newer Mac Pro systems. Your RAIDs may end up being a bottleneck. I'm running 4x2TB in a software RAID-0 inside my DaVinci system right now until I buy an expander. It sustains 380MB/s read, 320MB/s write and it's a bit on the slow side for when I really try to get working.

Those 3GHz CPUs are actually 15% slower than the 2.66GHz Westmere CPUs in the newer Macs, before memory speed is factored in.

This system should work very well, I'm just saying you might find a few bottlenecks. I would only go with the FX4800 cards if you already own one and can easily buy a second one to match.

[Vigen Vartanov]

Don't use the Quadro FX4800. The GTX285 and Quadro 4000 are much faster. GTX285 is out of production, unfortunately. I don't know about where you can buy to get the parts in Georgia, but here in the USA, the Quadro 4000 now sells for about $765. The Quadro FX4800 is an older generation card and is also out of production. But there seems to be ample supply out there. They sell for around $1300 because people haven't figured out that it's actually older and slower than the new 4000.

Thanks , will get 4000 in USA :)

[Vigen Vartanov]

Why the fiber card? Unless you already have fiber devices to connect to. I suggest an ATTO SAS card so you can connect more mainstream eSATA/SAS RAIDs and devices as well as an LTO tape if needed.

.

I have fiber card installed in system , did not want to spend more money on ATTO .

[luigivaltulini]

Jeff is simply perfect.
I agree with him using a nvidia 4000 chipset GF100 and much better than 4800.
congratulations Vigen nice Workstation .

Luigi

[Vigen Vartanov]

Thanks Jeff .
Any comments about Monitoring with my Sony ? this monitor is not 4:4:4 it is only 4:2:2 will i have problems with monitoring ?

[luigivaltulini]

Vigen,
see this.
http://reduser.net/forum/showthread.php?t=44849
***this monitor is not 4:4:4 it is only 4:2:2 will i have problems with monitoring***
no you not have a problem.
i have a Panasonic 2550 ,I really like this in bt709 . but it costs too much.

[Vigen Vartanov]

I also have BT-LH1710W , that i am using on set . May be better to change my Sony with panasonic ?

[luigivaltulini]

Yes ,
is it better to have all the same, so you can be sure about the colors.
Vigen this is my opinion.

[Alexander Christ]

Vigen, for the GUI you have to replace the 8800 GT, these cards work with DaVinci: NVIDIA GeForce GT 120, NVIDIA Quadro 4000 or ATI Radeon HD 5770.

[Vigen Vartanov]

Just put RedRocket in My Workstation .
Works fine with 8 nodes , 2 of them are blur .
15FPS with Half Good debayer . Looks like my configuration will work in realtime in 1/2 deb ? or not ?

[Peter Chamberlain]

HI. Just to clarify, the Red Rocket will accelerate the debayer and decompress process and is not related to the node count or type of grade. With a Rocket, I expect you can select half res premium debayer, set your timeline resolution to HD and depending on number and type of GPUs get between 5 and 15 nodes in realtime. Adding additional GPU processing will increase the number of nodes.

Peter

[Andrae Palmer]

Peter,

Can you comment on how the RED Rocket affects render out to compressed file formats. Currently I have a 12 core 2.66GHz without RED Rocket and I'm getting realtime debayer at half res. Will getting a RED Rocket speed up my workflow in anyway? I'm outputting to HD compressed formats like Apple ProRes 4444.

[Peter Chamberlain]

DaVinci Resolve uses GPU for all image grade processing, then completes the render by the CPU's if the user sets the Render screen to a use particular codec. i.e. not dpx. So the CPUs will share the available CPU time between encode and decode. I have not compared the times with a 12 core but I imagine having a Rocket may reduce the 444 rendering time. Be mindful however, all we are doing is moving speed bumps. Is it the CPU, Rocket, GPU, Motherboard, shared RAM, RAID card, disk I/O, number of spindles, etc etc. I have seen render speeds significantly increase if the source files are from one raid array and the render to another as the speed bump can sometimes be the disk and raid changing from supplier to destination of files.
Peter

[Andrae Palmer]

I have not compared the times with a 12 core but I imagine having a Rocket may reduce the 444 rendering time.

Is this pretty much the process when doing a render out to compressed file formats?

1. Debayer with RED Rocket or CPU
2. GPU effects
3. CPU transcode to compressed format.

I'm interested in knowing if the RED Rocket is utilized in any other stage. Really would love someone on a 12 core with a RED rocket post their times with and without the card. My question is if in step one I'm getting realtime debayer with the CPU then how does having a RED Rocket improve performance?

[jake blackstone]

My question is if in step one I'm getting realtime debayer with the CPU then how does having a RED Rocket improve performance?

You can't possibly get a real time full premium HD debayer even with 12 cores. I'm not even talking about 4k or 5k debayer in real time. But if you're happy with present non real time full premium debayer speed, then, obviously, having a Red Rocket would not be beneficial.
BTW, you may not readily see the difference between 1/2 and full premium debayer with 4k shot material, but you will see the difference very quickly with 2k and 3k shot material.

[Andrae Palmer]

22-24 FPS playback for 2048x1152 Resolution RED footage at "Full Resolution Premium" processed for a 2K 2048x1556 timeline.... CPU load at 70% on a 12 core Mac Pro 2.66GHz. I suspect with a 12 core 2.93 the playback would be a solid 24 fPS.

Unfortunately I haven't shot any material in 3K to do a test. Currently observing resolution differences on 2048x1152 material on a 2K timeline. This may sound crazy... but "Half. Res. Good" looks better than "Ful Res. Premium"....

Half Res. Premium is absolutely horrible (blurry).

I'll post some pics comparing Half. Res Premium to Full Res. Premium. The blurriness of Full Res. Premium is observable. Half Res. Good shows more detail. These observations are related to playback in Davinci Resolve... I don't know if the final render would have the same characteristics. Jose has made the same observation:


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?.

http://farm6.static.flickr.com/5053/5430309711_85b02eff23_o.jpg

I'm finding out some information that is jsut blowing my mind away:
http://reduser.net/forum/showthread.php?t=52985

The RED Rocket is not suitable for finishing and mastering?

[Rob Ruffo]

22-24 FPS playback for 2048x1152 Resolution RED footage at "Full Resolution Premium" processed for a 2K 2048x1556 timeline.... CPU load at 70% on a 12 core Mac Pro 2.66GHz. I suspect with a 12 core 2.93 the playback would be a solid 24 fPS.

Unfortunately I haven't shot any material in 3K to do a test. Currently observing resolution differences on 2048x1152 material on a 2K timeline. This may sound crazy... but "Half. Res. Good" looks better than "Ful Res. Premium"....

Half Res. Premium is absolutely horrible (blurry).

I'll post some pics comparing Half. Res Premium to Full Res. Premium. The blurriness of Full Res. Premium is observable. Half Res. Good shows more detail. These observations are related to playback in Davinci Resolve... I don't know if the final render would have the same characteristics. Jose has made the same observation:



http://farm6.static.flickr.com/5053/5430309711_85b02eff23_o.jpg

I'm finding out some information that is jsut blowing my mind away:
http://reduser.net/forum/showthread.php?t=52985

The RED Rocket is not suitable for finishing and mastering?

The line-skipping is simply causing false detail to appear, creating the impression of "sharpness" but it's just aliasing. I also find "good" color decoding is a wee bit off, creating false regions of color that should be smooth and other artifacts. You really don't know this stuff? No offense but that's a wee bit shocking.