Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Best Mac Pro graphics card for RED work

S

Scott Simmons

Well-known member
Joined
Sep 1, 2007
Messages
243
Reaction score
0
Points
0
Website
www.scottsimmons.tv
Any consensus on what the best graphics card is for RED work (and good RedCine karma) on a new Mac Pro?

These are the options from a build to order system:

  • ATI Radeon HD 2600 XT 256MB
  • NVIDIA Quadro FX 5600 1.5GB
  • NVIDIA GeForce 8800 GT 512MB

That 5600 is one pricey option!
 
redcine or redalert

redcine or redalert

editblog said:
Any consensus on what the best graphics card is for RED work (and good RedCine karma) on a new Mac Pro?
Click to expand...

another way to look at it is also using redcine vs. redlert. i was under the impression that redcine relied heavily on the cpu while redalert relies on gpu mostly. i'm running the stock ati card on my mac pro and redcine runs fine. redalert, however, is too slow to be of any use in gauging motion or even resolution as it scales very slowly. maybe i need a gpu update myself.
 
Got it backwards, wigby. REDCINE is GPU accelerated, RED Alert is not. The actual performance boost from the GPU appears to be very minimal and I think I read somewhere at one point that the only operations being handled via GPU were some color functions.
 
Anything else? I'm with this problem too...which one??
 
GoeroeMedia said:
Go for the Nvidia cards.
ATI has some known problems with RedCine.
Render errors etc...
Click to expand...

That might be true but ATI is way better for Apple Pro Apps. Especially for Color en Motion there are lots of issues with Nvidea.

For example, ATI supports all bit depths in Color. Nvidea only 8 bit and float.
Also the performance of ATI2600 in Motion is way better compared to Nvidea which has way better specs.
Although i heard that 10.5.3 had some performance improvements for Nvidea cards.

I know it's not about RED but it's certainly something to think about.

:angry2:
 
And all of this makes me wonder why Apple elected to use Nvidia chips in the MacBook Pros...

But this all seems to be driver immaturity more than anything else.
 
What are people currently getting in terms of render times? I'm seeing 1fps with this configuration.

Mac Pro 2.8ghz
16GB RAM
ATI 2600
Firewire 800 drives

I'm rendering out to ProRes HQ 1080P. Footage was shot 4K 2:1 and I am scaling to width. I'm using the Sinc scaling algorithm. Any thoughts? Is this what most are seeing in terms of render time?

I have over 2 hours off footage that needs to be converted to the ProRes and it's saying it will take 45 hours. This is the first project that I needed to spit out the rendered files first. usually I use proxies and then just transcode what I need via Crimson and Redcine.
 
Steve, while REDCINE is no speed demon, 1fps seems a bit slow.

I'm going to guess that there are a few things slowing you down -- one is the Firewire 800 drives, the other could be display settings and sizes within RC. Also which generation Mac Pro is this?

For comparision I just ran a 1000 frame sequence from two clips in RC, shot 4K 2:1 23.98, out to 1080p, scaled to width ProRes HQ with Sinc filter, max quality. Time according to my iPhone stop watch was 9min 26.2sec. Or 0.566sec per frame = 1.766 FPS.

System specs:
8-Core Mac Pro 3.2GHz
16GB 800MHz RAM
nVidia 8800GT
Data was loaded and saved to internal hardware RAID-0 that can sustain 220MB/s.
OSX 10.5.3

RC displayed full-screen on display #2 (30" DELL LCD 2560x1600), display settings were 1/2 res medium, and fit 1:1; so I see the 1920x1080 raster with matted 2:1 RED footage. Yes, display in RC can affect performance.
 
Jeff Kilgroe said:
Steve, while REDCINE is no speed demon, 1fps seems a bit slow.

I'm going to guess that there are a few things slowing you down -- one is the Firewire 800 drives, the other could be display settings and sizes within RC. Also which generation Mac Pro is this?

For comparision I just ran a 1000 frame sequence from two clips in RC, shot 4K 2:1 23.98, out to 1080p, scaled to width ProRes HQ with Sinc filter, max quality. Time according to my iPhone stop watch was 9min 26.2sec. Or 0.566 FPS.

System specs:
8-Core Mac Pro 3.2GHz
16GB 800MHz RAM
nVidia 8800GT
Data was loaded and saved to internal hardware RAID-0 that can sustain 220MB/s.
OSX 10.5.3

RC displayed full-screen on display #2 (30" DELL LCD 2560x1600), display settings were 1/2 res medium, and fit 1:1; so I see the 1920x1080 raster with matted 2:1 RED footage. Yes, display in RC can affect performance.
Click to expand...

Current generation mac pro 2.8ghz, Dell 24" display
 
redcine gpu

redcine gpu

Jeff Kilgroe said:
Got it backwards, wigby. REDCINE is GPU accelerated, RED Alert is not. The actual performance boost from the GPU appears to be very minimal and I think I read somewhere at one point that the only operations being handled via GPU were some color functions.
Click to expand...

ok. i don't use redalert but i do use redcine. so are you saying if i swap out my ati for a faster nvidia card i should see a big, small or no improvement in redcine?

thanks
 
wigby said:
ok. i don't use redalert but i do use redcine. so are you saying if i swap out my ati for a faster nvidia card i should see a big, small or no improvement in redcine?
Click to expand...

Small improvement, basically you won't notice much for display performance in Redcine and I'm not sure if it impacts render output or not. I haven't done a direct comparison myself, but the times I'm seeing reported from people indicate there isn't much (if any) speed difference in Redcine. Someone on the forums swapped their ATI for an 8800GT and reported very little speed increase in Redcine. ...Which seems to indicate that actual performance boost from GPU functions is very minimal. Hopefully that will change in the future and we'll start seeing better performance with better GPUs in Redcine. But there may also be other things standing in the way... Redcine and Red Alert are horribly inefficient when they scale beyond 4 cores. Actually Redcine is pretty messy even on 4 cores. ...Same thing with the QT plug-in. My dual core Macbook Pro can play back H proxies just as well as my 8-core Mac Pro can, but the Mac Pro has all 8 CPUs working at 80% or better. I can restrict Redcine to only run on 4 CPUs instead of 8 and I essentially still get the same render times, so the thread scheduling and multithreading code is in dire need of some attention. ...That's just one example. The software is still very much in beta form. Hopefully we'll see an improved version soon with Build 16.
 
so does 45 hours to render 2.5 hours of material seem about right or does it sound like there is a bottleneck that could be removed?
 
45 hours?!?!

My longest render from RedCine into UC10bit was 8 hours for 2 hours 7 min of footage.
I really think the key to this lies in RAM. I've read the posts concerning the various graphics cards and how they perform against one another and I don't know, but I doubt my Kona 3 has much to do with anything until I get into FCS.

But 45 hours?!?! Has to be a joke...

EDIT: everyone knows about and uses Shadowkiller right? I know this is a small haxie but it frees up processor by killing the drop shadow on all windows in OSX. Try it out..
 
Shawn Booth said:
45 hours?!?!

My longest render from RedCine into UC10bit was 8 hours for 2 hours 7 min of footage.
I really think the key to this lies in RAM. I've read the posts concerning the various graphics cards and how they perform against one another and I don't know, but I doubt my Kona 3 has much to do with anything until I get into FCS.

But 45 hours?!?! Has to be a joke...

EDIT: everyone knows about and uses Shadowkiller right? I know this is a small haxie but it frees up processor by killing the drop shadow on all windows in OSX. Try it out..
Click to expand...

Sean-
I'm a camera geek and not a computer guru, but it would seem to me that the window shadows would be just a drop in the bucket with newer quad and 8 core systems compared to older/slower macs. In reference to Shadowkiller the unsanity site states "This changes the look of MacOS X completely, and speeds up its operation for slower Macs (such as G3)."
Any comments?
 
Steve, You may have a slight bottleneck, most likely the firewire drives.

For your quality / format options, your speed looks pretty typical. Assuming 23.98 FPS footage @ my 1.766FPS processed in this manner, that means 172,656 frames. And 1.76FPS would take 27.25 hours. Assuming you have 800MHz RAM, the 2.8GHz CPUs should still handle better than 1.5fps, we'll use that as a worse case scenario, and with that, my calculations say 39.9 hours to render out your footage. But even off my faster RAID with faster CPUs, I'm not a whole lot faster here. One big bottleneck in this method seems to be the ProRes HQ codec / encoder. The Sinc filter is slowing the process down too, but not much.

The biggest culprit is setting the Process to Full. This kills render times. If you leave it on Standard, you'll get much better results.

Going from 4K 2:1 23.98 to ProRes HQ 1080 with the 2:1 frame scaled and matted, I can get about 1/6th real-time using the Sinc filter, High Quality and Process set to Standard.

The same settings, but going to DPX sequences instead, I can go about twice that fast.

It all depends on your needs, but IMO the results are pretty damn good with the Standard process when you're effectively scaling to 1/4 res 1080p. If you really need the full process, I guess it's time to buck up and bring in more systems and use REDLINE.

FWIW, I just did a speed test, same 1000 frames, but instead of ProRes, I went to DPX sequences. Standard Process, Sinc filter. It took 2min 17sec, that's essentially 1/3 real-time.
 
You must log in or register to reply here.
Back
Top