CS5: CUDA, CPU and other hardware questions

[Geoff Vane]

CS5 can gain much speed with CUDA technology by Nvidia; I assume everybody who researched a bit knows this.

However, I like very exact information to get the best decision for a new investment.

I need to get two identical new CS5 systems, that are much faster and not as buggy as my last CS4 system.

I have about 30.000 euro for 2 machines.

Our studio demands SDI connections, so I got a Matrox Axio.

Currently my system has lots of problems with Premiere and After Effects:

- Field swaps

- Inconsistent effect parameters

- Very slow rendering

- Very slow app startups

- Multi task slowdowns

- Crashes of apps

- Very slow interface operations

1. Does CS5 only use the CUDA of supported CUDA Cards?

2. Does SLI or multi graphic card tech, help CS5 speed at all?

3. Is it useful to get a Tesla next to a Quadro 5800 or is it not seen by CS5?

4. What Intel processor is best suited for After Effects rendering: iX or Xeon architecture?

5. Does Adobe really use only 3 Gb per core, meaning you should not buy more than 3Gb per CPU core? Or is it per THREAD and should I get more RAM?

6. Speed or number: Is it better to focus on very high total core speed, or to optimal RAM distribution by getting as many cores as possible?

7. Can a Matrox MXO2 Rack unit, cooperate with the Nvidia CUDA power or could both accelerations/technologies hinder eachother?

8. Is it best to get Nvidia SDI connectivity or is there a save alternative?

9. Can iX CPU's be dual or is that a Xeon option only?

10. What is the fastest 6 core Xeon that can be dual?

11. Does CS5 need a good scratch disk: is it wise to get a SSD just for scratch or is it overkill?

12. Some say the Nvidia 5800 is not the fastest card around; it has 240 cuda cores and some gtx cards have far more than that. What is the story?

I probably forgot about 324 other questions, but for now these will do.

If any of you can give objective and confirmed information, please do!

[Mylenium]

1. Does CS5 only use the CUDA of supported CUDA Cards?

Only Premiere. CUDA support in AE is dependent on specific plug-ins using these  features. AE itself relies on conventional OpenGL features, not CUDA.

2. Does SLI or multi graphic card tech, help CS5 speed at all?

No. SLI is more or less a gaming feature utterly irrelevant for most professional apps. Even most 3D programs, which are much more dependent on OpenGL than AE ever is, do not  support it.

3. Is it useful to get a Tesla next to a Quadro 5800 or is it not seen by CS5?

You can only have one primary graphics card for acceleration features. Tesla GPGPUs are handled completrely different and are not seen by normal apps unless you compile them with the respective NVidia libraries.

4. What Intel processor is best suited for After Effects rendering: iX or Xeon architecture?

Depends on what you prefer: Quicker interactive updates whiel working or faster final rendering using multiprocessing. Though, to be fair, the difference is not that great. Xeons work just as well ffor editing (pick the W series types) as do coreXi for multiprocessing renders.

5. Does Adobe really use only 3 Gb per core, meaning you should not buy more than 3Gb per CPU core? Or is it per THREAD and should I get more RAM?

You are operating on outdated info. That only applies to 32bit versions of AE, not the 64bit CS5.

6. Speed or number: Is it better to focus on very high total core speed, or to optimal RAM distribution by getting as many cores as possible?

Speed! Several features are not MP enabled and the efficiency of multithreading in plug-ins varies. You could have a ton of course and then use your favorite, single-threaded plug-in and all that poewer goes to waste.

7. Can a Matrox MXO2 Rack unit, cooperate with the Nvidia CUDA power or could both accelerations/technologies hinder eachother?

Most likely not. Same problem as with your Tesla proposoal: Only one can rule. That aside, simple bandwidth considerations with your PCI bus would probably make it pointless, even if it worked.

8. Is it best to get Nvidia SDI connectivity or is there a save alternative?

I doubt it would be the "best", as it is merely meant for previewing, not signal processing, playout and capture as is required in a studio. It may deliver the picture, but no TC control and all that. figure a BMD or AJA card into your equation.

9. Can iX CPU's be dual

No. As designed by intel... Guess why there is a considerable price difference between coreXi and Xeons...

10. What is the fastest 6 core Xeon that can be dual?

Ask that your hardware reseller. They werre just released/ announced a few days ago, so they may not even be available yet.

11. Does CS5 need a good scratch disk: is it wise to get a SSD just for scratch or is it overkill?

SSDs as scratch disks are just plain stupid. unless you get the super-expensive models, write speed is awful. They only make good boot disks, if that is relevant. AE uses a dedicated disk cache which best is handled with fast SATA disks or even a small RAID setup.

12. Some say the Nvidia 5800 is not the fastest card around; it has 240 cuda cores and some gtx cards have far more than that. What is the story?

And, does it matter? Unless you have program code that exceeds the number of CUDA commands that can be processed in one clock cycle, you'll never know the difference. Which is a way of saying that it will be pretty much irrelevant in most cases. For premiere and AE, anyway. Just like with CPUs, more cores does not automatically mean more processing power in a given app.

Mylenium

[yenaphe]

ANswers to some of your question:

1) Only Premiere Pro is really accelerated by CUDA thanks to it's Mercury Engine. And yes, only the supported cards can enable Cuda acceleration.

5) Now that AE CS5 is 64bit you can use more than 3gb/core. You can use all the ram you want per core.

6) It really depends on your compositions (number of layers, effects applied, blending modes...).

11) Yes it's better if you have a dedicated scratch disk. My config is like this: 1 HDD for Windows, 1 HDD for swap/temp/scratch, 1 HDD for sources and 1 HDD for renders. Like that you avoid cross reading/writing on disks, and keep the pace. SSD is overkill IMHO.

12) The 280 GTX card will allow CUDA acceleration up to 3 tracks in Ppro, whereas the Quadros supports 10 tracks and more. For AE, it won't probably make any difference.

Hope that helps (and I let the others reply the other questions)

[Todd_Kopriva · Accepted solution]

I'm just repeating what Mylenium and Sebastien said, but I wanted to make sure that their answers were considered as fact (which they are).

CUDA has nothing to do with After Effects. Among the Adobe video applicaitons, only Premiere Pro CS5 uses CUDA.Also, The Premiere Pro Mercury engine doesn't support/use SLI. (Please keep the Premiere Pro, CUDA, and Mercury questions on the Premiere Pro forum.)

There is not a 3GB/core limit in After Effects CS5. It is true that the largest number in the preset values in the Memory & Multiprocessing preferences is 3GB, but you can change the presets. See the comment at the bottom of this page for details on changing the presets.

Also, consider that After Effects isn't the only software running on your computer. At the very least, you also have an OS running. If you want to dedicate 4GB per process to After Effects (which is a reasonable number for digital cinema), and you're using 4 processor cores for Render Multiple Frames Simultaneously multiprocessing, then you don't just need 16GB; you need more like 20GB, so that the OS and other software have what they need. See this post, which was written for CS4---but the principles are still relevant for CS5: "Performance tip: Don't starve your software of RAM"

[Crist_OC_PC]

Mainly I apologize by my BAD English and let know these post does not have any lucrative intention but is informative.

If you Like to have, the best of the best you have to folow theses roule:

1- O.S. 64bits, processor minimun 4 core, and RAM minimun 12.

Thent the Secrect. as rendering on CS5 use all core, the key is fast processor and fast ram, Offcourse  you can tellme that nothing new for you. BUUUT. Have you ever try a machine running @ 4.2 GHZ, Whit Ram speed off 1800mhz whit latenci 7-7-20.

Shure not, Other whise you shoul not be looking these forum.

You want to know THE TRUEE about the best PC'S for cs5. Letme knowt. krostyforever@gmail.com or ask on the forum so I can share these info.

I hav´t see no one doing OC. for cs5. I allready built  5 pc's. For shure none pc on the market is faster for use CS5, no even MAC'S WHIT multi XEON processors. whict cost over 12000 usd.

The pc's I built are under 3000USD. and they are:

*Your sheapers choise.

*The best quality

*Upgrateable if  you feel you need it. "if you like to make the edition of MATRIX 4"

*100%Compatible whit all adobe sof.

*100% stable under FULL strees test for mor that 2 days WHIT PERFECTS TEMP. for people who knowt.

I'm a harware pc expert and if there something to say to ADOBE people is Congrats for a sof. who let use the maximun potential of a xtreme pc

whict nobody could imagine.

[Geoff Vane]

I have bought a new system now.

At first I had a 6 core machine with 12 Gb memory.

Forget about that if you can afford better stuff. It's only a minimal improvement; don't let anyone else tell you otherwise!

I still got resource problems when opening AE and PP simultaneously.

I bought a 12 core Xeon with 48 GB RAM, Nvidia Quadro 4000.

That will give you some real notable improvement.

AE and PP can be open at the same time with much less problems.

Speed improves in the range of 3 times at minimum and 60 times at maximum.

You still have to adjust CPU and memory sharing between AE and PP if you want to maximize rendering speed for either one at certain moments.

I wish I could flick a switch and focus resources on a specific app if I'd want to, or choose for an even spread of resources as much as possibble.

That would make memory and CPU sharing a lot simpler to adjust.

Now I must dig into the preferences all the time and not forget to put things back again afterwards.

Matrox does work fine with CS5 but realtime gives you 95% quality in some specific cases. (frame blending while doing slomo or precomping)

Forced rendering will take you to 100% quality.

Up to ten layers with evil effects could be played at 95% in realtime so you'll absolutely see what you're making and waiting or stuttering is over.

All kinds of unpleasant files can be played completely mixed on one timeline.

Only few problems arise:

• Quicktime incompatibility with 64 bit, so things can crash (f.e. when you use TGA sequences). I had to deinstall quicktime which is a big handicap when using modern camera files.
• Premiere CS5 always uses Encore for DVDs and it forces the Matrox upper field to lower automatically which results in 50% loss of sharpness.
• Preview PC windows do not show a 100% quality image; I have to check my TV monitor to see 100% quality.
• The value sliders in PP don't react fast enough; I'm clicking and shoving about three times before anything happens.
• I can't write to XDCAM decks (PDW 1500)

[Crist_OC_PC]

Why dont you go to the http://ppbm5.com and show how god and fast is your machine. That can help you to improvise a little and surely you will Learn much to obtain the maximum performance in your pc.

You dont even mention a Raid CONFIG or O.C. or Type of ram....

To  know about the comunication betwen all the pieces of your system is the key of a faster PC.

Look Harm's PC. http://ppbm5.com/Benchmark5.htmlWhit the money you used man can buy 2 of harm pc´s and just try to make the benchmark and see that what I tellyou is true.

B.R.

[Geoff Vane]

Hello Crist OC/PC - B.R.

You are 100% right about computer speed.

>>> Most important for a computer's performance is how individual components interact <<<

-----------

Please try and see it from my perspective and read my approach:

1. First I try to find top nerds. Only then, you can get a computer that is fast.
   I posted on several forums: the Adobe forum, computer nerd forums, Nvida forum, etc. etc. I even wrote to employees of Adobe and Matrox and Nvidia I met at the IBC. This all lead to no useable results; you just get the advice to buy a decent PC. (Where were you when I needed you?)
   
2. I found a company who makes custom PC's for research labs where they design molecules and do mechanical stress tests and so on. It seemed to me they could build me a machine.
   
3. I called the computer company and they build me a test machine which had 6 cores and 12 Gb memory. My old system was a quadcore which took about 50 minutes for complex tasks. This newer test system could do this in 20 minutes.
   
4. The computer company payed me a visit, looked at the results and we decided a 12 core with 48 Gb would be better.
   
5. The new system was brought in. It took 10 minutes for the same complex task. Good enough.

-----------

• I do not think my new system is "god or fast". It's acceptable to wait 10 minutes in stead of 50 minutes.
  Some tasks are 3 times faster than on my old quadcore, some are 60 times faster.
  
• Due to company rules, it's not allowed to build your own machine; we depend on service and guarantees constructors give.

-----------

But now comes the fun part, the incompatibility issues between the hardware parts and software quircks.

There is no such thing as a "minor quirck" in your system or a "work around"; all such nonsense adds up and become dangerous hurdles in the workflow. Here we go:

• The DVD encoding by Encore is lightning fast; 50% random chance the fields are messed with and cause horrible video.
  
• Not any instance of Dynamic Link can be used because it seriously destabilizes my computer.
  
• I can't use AE and PP together since it increases chances for a crash dramatically.
  
• AE is instable anyway; even standalone it invokes DL errors.
  
• It took me 48Gb to finally have a computer that does not eat up AE RAM previews from the start when the end is almost rendered. Praise the Lord.
  
• My MXO2 board is needed to connect to the studio with SDI. Hardly any computer nerd or Adobe user can tell you anything about this card's influence on system performance. I'd rather not have such a card because I do not trust them, but I must use one to get professional connections.
  
• What I do know is that I can't write to tape anymore. Adobe thinks it's old school; come tell my clients that. Matrox tries to solve that task with their own weaved in engine but they can't seem to get it working yet.
  
• My video signal can't be synced to our house sync generator; my image bounces random.
  
• The view in all Adobe computer screen windows, is very rough in it's pixels and can't serve as a reliable reference.
  
• The Matrox and the Nvidia both have accelerating capabilities. It's very difficult to activate these engines or totaly impossible to tell if one or the other card is performing a certain task, or if ANY of those cards is performing such tasks at all. I fear those cards may collide somewhere inside the workflow.
  
• The Matrox card can not be used as preview in Encore; it crashes Encore. I must perform a certain reinstall trick.
  
• Quicktime seriously corrupts PP. If quicktimes are mixed with 32bit tga and/or psd, results are unpredictable and PP may hang. 32 Quicktimes are even worse. I can't do without Quicktime, since it is the only known way to render 32 bit footage that Avid can read.
  
• Disabling hyperthreading in the bios improved stability by 25%.
  
• I can't read SDHC cards, not even with a SDHC reader. Such a pity clients bring their material in on such cards.

-----------

You may be able to improve my computer somewhat.
I hope it's not just for speed, because lack of speed is not the greatest problem I have at the moment.
If this system does not get stable within two years I won't use a PC again but Mac.

Any help is appreciated very much and I'll try to do the test.

[Crist_OC_PC]

Well to use short words.

Your pc Is grate, but think what will happens if you have 1 pc for the importing of all your data. SDI, TAPE ETC... where all these data is goint to be storage on a external SSD. NOTE: " on these pc you can have matrox, blackmagic, whatever.... input-card". Then your  pc 2 is your Best,  perfect OC for ram and cpu, no hyperT. Totaly ready for the perfect interac CS5/PC. you just need to conect your SSD and guess what, These choice still been much better that your pc.

BTW, I posted before you got the pc.

Be my guess for any other cuestion.