Computer Specs For Red Post-Production

[Bruce Allen]

I agree with Nick.

In a non-client-supervised session, it is similar too. Speed of project completion usually depends on a combination of:
1. management and understanding of client expectations
2. showing them work in progress at the right time (not too early or too late)
3. getting the work looking good as soon as possible
4. luck!

I guess having a fast computer helps with #3. But talent and listening to clients (and selecting ones with good temperaments...) is more important.

Over the past three years my rate has more than quadrupled even though I have kept the same computer (a single-core 2.8ghz Pentium 4) at home. This is for not just editing but also After Effects, Maya work, coloring, etc. I'll admit, I'm mostly working on other peoples' workstations. On the other hand I know a lot of great artists with super-expensive machines that are sitting at home losing value because the artists got jobs at a company that had workstations already. And when I do the occasional client-supervised job from home, I have never had complaints because I work quickly and know how to schedule things so renders aren't an apparent issue.

Also, not spending money on constant computer upgrades means that you can afford to make higher risk / reward job moves because your finances are in better order if everything goes wrong.

Bruce Allen
www.boacinema.com

 

[Seán_T]

Well I never said anything was "that simple".
Of course theres plenty of people out there making money with old gear. The question being asked here is, "is it worth spending more on a speed bump when buying a new machine" if its for professional work, I say yes it is(well within reason).
If you cant afford the extra cost of the quad core vs dual (yes the mac os is not optimised yet) then maybe your cutting it a bit too close, (hopefully you would have a contingency plan that would absorb the extra $2K)

Nick I never said you up your rates I said you get more work done and move on to the next job. If your not billing by the hour you deffinetly win. If you are billing by the hour you get to make 2 clients happy in the same time instead of one, also a win. (yes of course you need to be busy/good enough to have clients lining up)

I think that all of these billing models have their pluses and minuses.

I also agree that keeping your gear up to date every time an update comes out hurts and is not recommended, but thats not what I were talking about, were talking about a new machine.

It's not quite that simple though. If you charge for online editing by the hour, and there is less waiting for rendering you end up billing for less hours. It's not always easy to tell clients "I've just bought a new computer, so my hourly rate has gone up". There is an extent to which you are just expected to keep up with the bleeding edge of technology, and the prices you can charge sometimes come down as that technology becomes more commonly available.

 

[Seán_T]

I guess having a fast computer helps with #3. But talent and listening to clients (and selecting ones with good temperaments...) is more important.

Over the past three years my rate has more than quadrupled even though I have kept the same computer (a single-core 2.8ghz Pentium 4) at home.

Bruce this is exactly what I'm talking about. If you buy a fast machine today it could be still useful in 3+ years. And every render you do over the next 3 years those extra CPU cycles pay you back in saved time. And if you can turn that saved time in to revenue then the greater investment will be paid back many times over.

And even if you cant turn the saved time in to more jobs done your fast machine will be useful longer. If your 3 year old machine was a 1.6Ghz P4 you probably would have dumped it already and bought a new machine. And of course were talking about 4K files here lotsohorsepower required for fx colour etc.

 

[Antoine Baumann]

Hi,
I think it is nice to have 8 cores when rendering, but people do not forget your RAM is divided by each processors.... actually I have 4 gig of RAM in a dualcore machine, so it 2gig of RAM per processor, so when you get to 8 cores you need 16gig of RAM, which cost a lot. But as you said time is money and I totaly agree. But personnaly I think 8 cores is more for 3D stuff, than color grading; ok yes may be compositing with shake if you have really complex comp.

If you are thinking to go for 2k/4k wokflow but need to do compositing work, you will need to work on other codec than REDCODE RAW, meaning you might need really fast drives to handle that, so I would suggest getting an ATTO card, and a lot of hard drives in a nice enclosure.

About the graphic card, I wouldn't go for the quadro fx 4500, unless you are unsing a specific software that requires it.

As for color grading, I do not think shake is efficient for it, and we have to see what color will be able to do, but it will be certainly twicked to work best with prores, so again if you are looking 2k/4k road, you might need to look for better tools.


antoine.

 

[Chris Kenny]

Hi,
I think it is nice to have 8 cores when rendering, but people do not forget your RAM is divided by each processors.... actually I have 4 gig of RAM in a dualcore machine, so it 2gig of RAM per processor, so when you get to 8 cores you need 16gig of RAM, which cost a lot.

Err? In an SMP architecture (which is what every desktop multiprocessor system I've ever heard of uses), all processors have equal access to the entire memory address space.

 

[Rob Lohman]

I'm not following Antoine, Chris is correct.

 

[Jeff Kilgroe]

Err? In an SMP architecture (which is what every desktop multiprocessor system I've ever heard of uses), all processors have equal access to the entire memory address space.

Chris is mostly right as far as desktop workstations go. But moving into the server market, even in the world of x86 systems, we do see some NUMA (non-uniform memory architecture) implementations and some of AMD's Opteron configurations that bridge the gap between server and workstation also have dedicated RAM blocks per CPU. The memory controller in AMD64 CPUs is right on the CPU and they don't scale well beyond two CPU sockets / four CPU cores at the moment. In fact, for high-bandwidth applications it still makes far more sense to dedicate each CPU socket with its own RAM bank. Just about any 2-way and 4-way Opteron board is an example of this. Now all CPUs still have global access to the entire RAM heap, but the RAM that's associated with each CPU socket is available at a much higher speed to their corresponding sockets, than RAM on a bank tied to another socket. Most of these boards provide BIOS hooks and through MPS info in the firmware about which RAM locations are optimal for which CPUs. But it's up to the OS and software to make use of this information...

A couple mainboards with RAM dedicated to individual CPU sockets would be the TYAN h2000M dual socket <link> and the TYAN 4-way Opteron Thunder n4250QE <link>

 

[Bruce Allen]

I think what laboprod is trying to say might be "don't expect you can get a 4x speed increase over a 1-core sytem just by buying a 4-core system. You need to buy 4x the amount of RAM as well, if what you were doing has a RAM bottleneck."

Eg if your Mental Ray scene uses 1.5GB RAM and is taking 20 minutes per frame to render on a single core system with 2GB RAM, don't think you'll get 5 minutes per frame on a four-core system with 2GB RAM. You'll need 8GB RAM, which might get exponentially expensive because greater density chips are more highly priced.

Bruce this is exactly what I'm talking about. If you buy a fast machine today it could be still useful in 3+ years. And every render you do over the next 3 years those extra CPU cycles pay you back in saved time. And if you can turn that saved time in to revenue then the greater investment will be paid back many times over.

And even if you cant turn the saved time in to more jobs done your fast machine will be useful longer. If your 3 year old machine was a 1.6Ghz P4 you probably would have dumped it already and bought a new machine. And of course were talking about 4K files here lotsohorsepower required for fx colour etc.

Gotcha. I agree. But the P4 I got years ago wasn't top of the line when I bought it then. It was a little bit below top of the line, which is often where the best value lies. The graphics card was mid-range even then. The whole thing cost me < $1000.

If I'd bought a 4-core $5000 computer 3 years ago with top-of-the line gfx card, RAID, etc back then, it would still be out of date pretty soon after the $1000 system I bought then.

Yes, I'm glad I didn't buy a 1.6Ghz P4, but I'm also REALLY glad that I didn't buy a "top-of-the-line" computer back then either.

Bruce Allen
www.boacinema.com

 

[Antoine Baumann]

Thanks Chris and Rob for the technical precision.

And thanks Bruce, that is exactly the point I wanted to raise.

I also think that the best value is always a little below the top of the line.

antoine.

 

[Jeff Kilgroe]

Eg if your Mental Ray scene uses 1.5GB RAM and is taking 20 minutes per frame to render on a single core system with 2GB RAM, don't think you'll get 5 minutes per frame on a four-core system with 2GB RAM. You'll need 8GB RAM, which might get exponentially expensive because greater density chips are more highly priced.

Yep. I take that as a given, especially with my 3D work. So I don't really think of it. I suppose a lot of people new to the multiprocessor multi-core world may not initially realize this. I'm often setting individual render nodes per CPU/core because it usually gives better results than the brain-dead thread "scheduling" when trying to run a multithreaded render. Other apps beyond 3D render node software works this way too. Database systems are a good example.

 

[Rob Lohman]

Luckily REDCODE stuff doesn't require that much memory, so I wouldn't go overboard unless you're also using it for other stuff that benefits from it.

 

[Bruce Allen]

True, I can't imagine RedCine using too much RAM.

Maybe there should be a "REDBENCH" which we can convince all of the hardware review sites to run every time they test a new computer... kinda like CINEBENCH. It'd be much more useful for pros doing video work than the stupid "Microsoft Media Encoder / iTunes MP3 conversion" benchmarks they do now and would be one of the few benchmarks that measures a combination of CPU and GPU.

Of course, I imagine you'll be tweaking the algorithms and quality for a while so it won't be as set in stone... but the advantage would be that the graphics card manufacturers watch those benchmarks like a hawk, so if you got REDBENCH on the regular benchmark list, they'd probably become REALLY keen on making their drivers play nice with RedCine

Bruce Allen
www.boacinema.com

 

[Rob Lohman]

We're working hard at all the code we need to ship with the camera

A lot of programs (including REDCINE) can benefit from more memory to cache. However, on 32-bit systems there can be problems with going over 2, 3 or 4 GB of memory.

 

[Brook Willard]

Forgive me if this has been asked before. Rob, does REDCINE populate a system with multiple instances to take advantage of 4-8 cores? Also, does it address higher quantities of RAM in a 64-bit system? It may be complete overkill... you know its performance better than anybody here... but curiosity gets the best of me.

 

[Ketch Rossi]

Hi,
I believe that a Mac will do the work
* Two 3.0GHz Quad-Core Intel Xeon
* 8GB (4 x 2GB)
* 500GB 7200-rpm Serial ATA 3Gb/s
* NVIDIA Quadro FX 4500 512MB, Stereo 3D (2 x dual-link DVI)
* Apple Cinema HD Display (23" flat panel)
* Apple Cinema HD Display (23" flat panel)
* One 16x SuperDrive
* Quad-channel 4Gb Fibre Channel PCI Express card
* Apple Keyboard and Mighty Mouse - U.S. English
* Mac OS X - U.S. English
* Final Cut Studio 2
* Shake
Total of about $12800
and also G-SPEED S
4.5TB
WAS $7,499
NOW $5,999
or MAC RAID Storage
3500GB ADM (7x500GB Ultra ATA)
$8,199.00
and AJA's Io HD not sure for the price
I believe that will do the work, preview will be at 2k if I'm not mistaken.

Hi Panicos,

I like your option, similar to the one I intend to built but with 750GB 7200-rpm..., not sure on the screens I like better the 30inc, is there a reason why you choosed the 23inc?

I like to know, thanks.

Ciao,

KETCH ROSSI
www.KETCHFRAME.com

 

[Anders Holck]

Note that Apple highly recommends the ATI X1900 for Color, not the Quadro or any other Nvidia card.

http://www.apple.com/finalcutstudio/color/specs.html

In the Final Touch days I believe they endorsed ATI as well.

Maybe Graeme has some info about this...

 

[coboman]

G-SPEED S 4.5TB

Why the SCSI version of GSpeed? Is it good enough, or is the 4Gbit/sec Fibre Channel a better option?

I would really like to know the difference between these two, and the GSpeed eS (SATA), with 3 Gbit/sec sounds like a great choice (when it is available), costing a third of any of these other options.

Any info on the subject is greatly appreciated.

 

[Chris Kenny]

There's really not much reason to buy something like a 4.5 TB G-SPEED for working with Red footage. It's not really large or fast enough for an uncompressed 4K workflow, and the speed is overkill for a compressed workflow. You can get a similar amount of storage for 1/3 of the price by just buying commodity eSATA enclosures (that do port multiplication) and loading them up with lots of bare drives.

 

[Rob Lohman]

Forgive me if this has been asked before. Rob, does REDCINE populate a system with multiple instances to take advantage of 4-8 cores? Also, does it address higher quantities of RAM in a 64-bit system? It may be complete overkill... you know its performance better than anybody here... but curiosity gets the best of me.

The decoder will take advantage of multiple CPU's. The image processing is done on the GPU. Memory we'll have to see. Without caching we don't really need that much.

 

[sNovaK]

Note that Apple highly recommends the ATI X1900 for Color, not the Quadro or any other Nvidia card.

http://www.apple.com/finalcutstudio/color/specs.html

In the Final Touch days I believe they endorsed ATI as well.

Maybe Graeme has some info about this...

Is there any part of Color's color processing that is hardware accelerated? I mean, if the actual color grading processing or final render takes advantage of the GPU's processing acceleration, this will make a difference.

I may be wrong here, but otherwise, I don't think it would make a considerable difference as to what graphics card your using whether it's the Quadro or another NVIDIA, or ATI.... as long as the card displays the GUI correctly and efficiently, the color grading process and rendering shouldn't be improved by any large stretch.

If Color's processing pipeline is the way I think it is, you're better off investing in more RAM and CPU "ommf".

Sean
Cineworks, Miami