FASTEST 4K rendering system for CS6 - Dual XEON or GPU (NVIDIA Cuda Cores) ?

[Guest]

Hi, I need to build a new POWERFUL video editing machine for After Effects/Premiere and could need some input as I am low on time.

The machine must be really fast in video editing/rendering, and I will be working both 1920x1080, 3840x2160 (24p) and 4K x 4K resolution projects. I work with a lot of timelapse, meaning a lot of large frames, and also computer generated frames inbetween key frames, eg. to slow down footage (Using both Adobes own, and different 3rd party engines for this).

First, I work in Adobe Creative Suite CS6 (After Effect and Premiere).

1. Should I choose a dual XEON R5 system (2630/2640), or is a fast i7 3960/3930K sufficient?

Higher end XEON E5 cpu's are really expensive, and as I understand from some sources, current models are "old" and should be replaced anytime soon, but I don't think I have time to wait. I know EVGA/ASUS have dual i7 motherboards, but are there any software (Adobe CS6) able to utilize dual i7 cpus as of today? I am unsure if there is any point in buying such a motherboard for my purpose.

2. I know some processes can be done by GPU. And I am tempted to buy the new GTX690, (and mod/hack Adobe info files so it becomes compatible with AE/PR), even if I know only a single GPU is currently supported in Adobe CS6 (giving me 1536 cuda cores), due to the possible use of the new NVIDIA hardware encoding system on newer NVIDIA cards, that may be possible (hopefully) to utilize in Adobe CS6 in the future. Or at least stand-alone.

3. What would be the overall fastest in rendering with Adobe CS6 (primarily After Effects)? Dual XEON E5-2630/2640 or a GTX680/690 card with 1536 cuda cores? (The higher XEONS above 2630/2640 are way too expensive I think, as well as NVIDIA QUADRO 5000/6000 Cards)

4. When I search about hardware/GPU rendering I now get a lot of threads about the "new raytracing engine" for rendering 3D. But I am not looking to render 3D, but 2D (What happened to "good old" mercury engine? ), and I need to do 2D rendering as FAST as possible!! I will be producing a 24 minute 4K x 4K (circular) movie for a client, as well as several "regular" 4K projects (3840x2160). I am confused why 2D rendering is not mentioned in raytracing rendering enginge.

5. Would 64GB DDR3 ram be sufficient? I am worried if I decide to overclock a little I may run into problems if going for too much RAM. Then again I may not need to overclock if the system is fast as it is.

6. Regarding Storage I could go for eiter a RAID workstatoin (RADI 5 or 6? Or other?), or a fast rendering workstation, with a slower and larger server for backup. I suppose I could use some increased write speed on the workstation as I will be rendering 4K x 4K footage in high bitrates, but I don't think the hard-drives will be the bottleneck here, but the rendering capabilites!?. Although I know choice of harddisk/storage is important in a video editing system, the other issues as with choosing Dual XEONS or i7 3930K + GTX690 (or both XEON + GTX690) for rendering is more urgent do decide upon right now.

Any input will be greatly appreciated!!

Best,

Ola

[Web Magi]

In my experience, Precision workstations are some of the most robust and reliable production machines made. I've been using a tower since 2008, and it's a beast, plus I still have a working 2005 Precision laptop.

I'm currently looking at this for upgrade purposes: http://www.dell.com/us/business/p/precision-t7600/pd?refid=precision-t7600

I built my own computers too, right up until 2005 when I got too busy to fix my main PC after it failed. If you go for the Precision, I would call to order. When I purchased the tower in 2008, I went through the motions of ordering online while I was ordering over the phone. I wanted to see if they were giving me the best deal possible. The machine I built online was over $5,000. The total cost of that same computer over the phone was less than $3,000. Not sure how they did that, but I didn't complain.

I honestly can't remember how many days it was between the order, and delivery, I just remember it being very few. I'm sure I paid a little extra for fast delivery.

Good luck!

[Guest]

Yes I am also considering Dell. But I think they charge a little too much. They charge 3 x the price of a hard-drive than what it cost elsewhere..

Will see what I decide, maybe I will call them as well.

Thanks.

[Mylenium]

4. When I search about hardware/GPU rendering I now get a lot of threads about the "new raytracing engine" for rendering 3D. But I am not looking to render 3D, but 2D (What happened to "good old" mercury engine? ), and I need to do 2D rendering as FAST as possible!!

The Mercury engine only exists in Premiere Pro at this point, but even there it is dependent on feeding it with formats it actualyl supports (which are a lot, but not all). Any acceleration in AE is dependent on third-party plug-ins using OpenGL/ CUDA, mostly, which will do zero if you don't use any of them. The rest is not really of any consequence. With AE as it is now, it is always an "either....,or..." situation. You cannot use GPU acceleration with MP rendering, so if you were to buy a dual Xeon setup to use MP, any  of those super-expensive Quadros wouldn't do much. And in reverse, plug-ins/ features that use GPU don't give a flyin' frakkk about MP. It's simply inherent in how most of that works. That aside, even if you limit yourself to CPU-only rendering, there are enough scenarios where MP rendering will just not work, including time-remapping or other temporal effects. So to cut a long story short: If interactive performance is crucial, a single core7i is better and it will also much better handle all that time mangling stuff. Everything else would be a waste.

Mylenium

[Guest]

Hi Mylenium and thank you for your reply.

Maybe I was a bit unclear, and most likely my knowledge on the area is outdated/limited, but I do not entirely understand your answer (at least not all of it).

I do not really care what the technologies for fast hardware accelereated rendering is/was called, I just seem to recall from a few years back, that hardware video (2D) rendering had something to do with "mercury engine" with Adobe/NVIDIA, hence I get confused with this new "raytraced" 3D rendering engine beinge mentioned everywhere, but no 2D hardware rendering being mentioned anymore. I am not necessarily looking to use "Mercure engine", I am simply looking for the fastest and best hardware way to go for working/editing/rendering high resolution high quality 2D video, dual Xeon E5, or a GTX680/690 with it's 1536 available CUDA cores for Adobe CS6.

I will try to rephrase my main challenge down to one question:

What CPU(s) would I need to beat video (1080/4K) rendering performance from a 1536 CUDA cores GTX680/690.

Or do I misunderstand? Are you saying that a CUDA based GPU simply can only do hardware encoding in/with specialized filters where CUDA is enabled, and cannot do the same core video number crunching rendering tasks as eg. a dual Xeon setup?

My After Effects CS6 task will consist of a lot of trial-and-error with rendering preview, using frame blending, different time-streching, warping, remapping, twixtoring, noise reduction filer (NEAT), and finally rendering it all to disk as fast as possible. So I will need a lot of fast RAM, but what else is best for achieving best performance here.. The xeons or the Cudas?

I was hoping to be able get away with a cheaper solution using an i7 3930K ir 3960K + GTX680/690, but if I really have a huge advantage of going dual Xeon E5, then maybe I will have to take the cost. And then the question is, would a dual E5 2630 6 core be good enough, or do I have to sell my left kidney and go for a dual E5 2660 with 8 cores.

Thanks for any help,

O.

[Mylenium]

Yes, CUDA is dependent on plug-ins/ programs actually using the pertinent nVidia libraries. It's not a generalized processing framework that is integrated in the operating system. Again, the rest is not relevant. As soon as you do any time-remapping/ time-stretching, the GPU is struck from the equation. GPU stuff in AE only works with "dumb" linear/ progressive processing tasks and loses its magic as soon as it cannot process all required commands during a clock cycle and it can't do that when it needs to look up multiple frames in your footage at different times. The same is true for parallel processing multiple frames on multiprocessor/ multicore machines - there are simply limits how much "out of order" processing you can do. And even if that would be possible processor wise, you'd still have to append frames in the correct order in a clip file, so that would be a bottleneck as would be file I/O in general, if 20 separate threads/ cores try to access the same clip files or write to them. Sorry, but you're looking for a simple answer that doesn't exist here and having the beefiest graphics card will be of no use to you nor will having multiple processors. Do as I said: go with a single core7i and rely on its TurboBoost feature. It's a lot more efficient than even the best Xeon for the kind of work you want to do.

Mylenium

[Guest]

If you look at the videos Web Magi is referring to below, they (Adobe) are using dual XEON processors for those hard core workstations. This is also the case for the HP 820 workstations for editing RED footage.

So I am a little confused how you can say an i7 with max 6 cores (12 with hyperthreading) will be better than eg. a dual XEON E5 2680 setup with 16 cores (32 with hyperthreading enabled) and lots of RAM (min.64GB) in editing and rendering high resolution footage.

I asked the same question at CreativeCOW, and there a forum user claims a dual XEON will "massively outperform an i7". I am not out to start a war here, and I really do appreciate your opinion, but it also sounds a little strange to me that an i7 will be better for my tasks than a dual 8-core XEON setup.

http://forums.creativecow.net/thread/2/1025615

I am now considering a DELL T7600 with a dual E5 2680 setup with 128GB RAM. I allready ordered the NVIDIA GTX 690 since I decided I will need it regardless. It will give med 1536 CUDA cores to do some aditional works, but now I just need the fastest hardware for the video crunching..

I will also be using some RAIDED SSD disks for boot/cache/programs, and a RAID5 SAS setup for storage.

[Mylenium]

Sorry, I give up. This is like talking to my mom about computers. You're missing the point - both here and on the COW. Buy whatever you want and burn your money, if you have too much to spare.

Mylenium

[Guest]

Mylenium, I am sorry if I offended you, not my intention. BUT, yes this will be a big investment for me regardless what system I choose. And yes the XEON system as the T7600 is insanely expensive, but what could be even more expensive for me is to buy a system which is underpowered, making me have to switch again, loosing even more money, and more important loosing time, which is a factor I currently do not have.

Now, if I sum up, there seem to be consensus about video rendering using GPU/CUDA cores will not gain much unless there are some special effects involved which takes advantage of the CUDA/GPU technology. OR?? I have seen many videos showing how CUDA rendering of regular video goes much faster than CPU rendering, at least on "normal" computers.

And this article from Toms Hardware may somewhat seem to support your view about i7 vs E5 for After Effects working with lower res footage at least. I wonder how this test would fare with 4K footage:

http://www.tomshardware.com/reviews/xeon-e5-2687w-benchmark-review,3149-6.html

But not for rendering from Premiere, which I also will be using. But then they say; "If you’re using a desktop card like Nvidia’s GeForce GTX 580, even a Phenom II X6 1100T can get this job finished in half the time of two pricey Xeon E5s. I’m no fan of locking out the competition, but when there’s money on the line, professionals working in CS5 simply owe it themselves to use a CUDA-enabled card."

This is confusing me even more, some say CUDA does not help on rendering, and then some say it really does.

But IF CUDA rendering has little or no effect on rendering speed, then we are down to CPU / memory power. And this is where have trouble understanding how an i7 system could be a better choice than a dual Xeon system with Quad channel memory (money put aside). Is it because After Effects is not able to fully utilize  such a heavy XEON system?

I am not rich by any means, and I certainly do not want to waste my money, but as I say, I am low on time and I cannot afford to fail as I have 4K commercial projects in the pipeline. These will mostly be done ine After Effects, but also some work in Premiere.

Thank you for your input Mylenium. I have still not decided 100% what to do, and your points does make my a little nervous for wasting my money.

[Web Magi]

From the After Effects Help section: http://helpx.adobe.com/after-effects/using/memory-storage.html

After Effects can start additional processes of the After Effects application to run in the background to assist the main foreground application with the rendering of frames for RAM previews or final output. These background processes have the name AfterFX.exe (Windows) or aeselflink (Mac OS).

In this form of multiprocessing, each background process renders its own frame and runs on a separate processor core (CPU). The number of processes used to render multiple frames simultaneously is never more than the number of processors.

More info here as well: http://helpx.adobe.com/content/help/en/after-effects/using/improve-performance.html

Even though it is 2 years old, Adobe has some great input on this issue as well: http://forums.adobe.com/thread/634192

Intel lists i7 is a desktop/gaming class CPU, and Xeon as a server/workstation class CPU. Dell, HP, and Apple (Mac Pro) all use Xeon processors for their high-end workstations. There is also the question of ECC memory, which helps to keep systems stable, and comes standard with most workstations.

I guess the question is, do you need a workstation, or a super-powered desktop?

[fredsky2003]

hi guys!

i've building machine for after effects since years, for a creative artist.

he know loves the new 3D renderer

there are different tehcniques in CS6 to "hide" the render process (background render), but this in not what you want. you want a VERY responsive system for testing / trying a effect and idea. this is what Millenium has tried to explain.

in  this case, XEON (like quadro) are only overpriced marketing fluf from intel/nvidia/HP/DEll/Apple (name it).... they just want to sell you hardware that you will not use...

so what you need is :

- the most high clocked processor with max cores... this mean desktop CPU, single socket. take a i7 with 6 cores, and "K" suffix so you can overclock about 4Ghz.

- max RAM, you need about 2GB per core (counting hyperthread). so 24GB + some for the OS. the bare minimum is 32GB.

- you want a responsive system. so you need a SSD for the OS + SSDs for cache, which is a killer time saver in CS6

- you need also a fast and secure storage for your footage. use a RAID hardware card. LSI or Areca, RAID5 + spare.

- as Millenium explain, the current "wow" effect in CS6 is when you use something that CUDA and Optix technologies can handle, using a Geforce card.

-> best performance for the price is GTX570

-> best performance single GPU is GTX680 (since CUDA 5 apparently... otherwise GTX580)

-> best performance at all should be with GTX690 dual GPU... but for what a read this far away from doubling the speed vs 680. you'll have to ask Tod from AE what it means by "support of GTX 690" in 11.0.2 update...

- in premiere, even when you export a file, you will use the nvida card for acceleration... but this accel is ONLY used for resizing at best quality and rendering GPU effects you used on your timeline. that mean that a very basic timeline with no effects is NOT accelrated; it use the CPU. so in any worklow, it will use CPU + GPU

- contrary to you title name, you will use CPU AND GPU (not Xeon or nvidia, but both)

- i read a article about high end xeons with lot of cores (32) and after effects CS5.5/CS6

     * CS5.5 doesn't hanle all those cores correctly at all

     * CS6 does much better, but too muche cores simply... slow down the rendering.

(see the link from ola_1974 as example). as Millenium said, a LOT of render in AE are still linear, and OoO, so you can not fead 32 cores correctly.

this emphasys the point that, like a lot of computer stuff, the hardware is far ahead software implementation.

to summup, go for a high clocked i7 + lots of RAM + GTX680...

you will have the most efficient system for the money, that make Apple Mac fans crying for the money they waste in their unused and not upgradable hardware.

fred

[Guest]

fredsky,

thank you for your input.

Indeed I think you are spot on, althoug rendering speed IS important, what is even more important, the more I think abuot it, is that I have a responsive system, fast in previewing my wanted effects and how they look before I render.

So what you're saying, even for 4K footage, or even 4K x 4K footage, the i7 + GTX680 would be faster previewing than a dual XEON+GTX680, since most previewing is done by the cudas anyway?

I was considering the DELL Aurora R4, as it came pre-overclocked, but it only has 32GB of RAM, and I think I would like to go for 64GB.

Still haven't decided 100%, and are getting some more input here on the forum. But a high overclocked i7 39XX sounds reasonable..

Thanks,

Ole

[fredsky2003]

hello Ole!

my friend does huge rendering and complex jobs in after effects, and link them with premiere.

so far, he's got 32GB ram, SSD, i7 3920, raid5 array and gtx470. very very happy.

32GB is the minimum but 64Gb, i'm not sure it will be faster... it will help for long AE ram preview; and perhaps you can use it as a swap ramdisk.

so for a prebuilt machine DELL Aurora R4  seem to be good, support 4 HDDs.

otherwise, built your own.

[Stone_Swan]

Just thought I'd chime in here and confirm what Ola_1974 wrote in his last post.

Last fall, I upgraded from an i7 2600K Dell XPS to a dual Xeon 2630 HP Z820 workstation with a better graphics card.  The only "downgrade" was my CPU clock speed.  I found the responsiveness of AE while editing to be slowed down, while actual rendering time decreased by 2.5x.  Thus, don't ignore CPU clock speed when building/buying a system.  More cores will benefit you for operations which are obviously multi-core (like rendering), but not much else.

[ECBowen]

The overall performance of AE  has several factors involved. Codecs decide much of the threading to clock speed performance comparison. The amount of ram assigned per thread has a factor on the Ram preview performance. Many FX especially 3rd party have a significant effect on the performance due to lack of multi-threading ability.  The overall user experience is decided by the editors workflow with the configuration when running Dual Xeons. Many workflows would work far better on a X79 workstation due to clock speed while others require the Dual Xeon for an excellent experience. Just remember that sacrificing a significant clock speed for the extra threads rarely results in far better return especially under 2.6GHz. That is where the performance really drops with most editing applications.

Eric

ADK

[VIRU9970]

hi,

I am new on the forum..

I have following Config. Machin.

Asus Mobo KGPE D16 ( single cpu )

Amd 8 core 6218

G skill 8 gb

Segate 1 tb

Lg dvd rw

ZB 100 EATX

750 watt PSU

now i want to config.  Asus GTX 670 DC II 2gb

for video editing , with 2 Full hd monitor set up ..

i want good deft in video with high bit,

and fully high professional video quality out.

with hi speed rendering......

How can i update my set up .....?

[Bill Gehrke]

You have a terrible configuration for Premiere editing.  Everything is wrong:

1. AMD processor, you really need an Intel Quad core with hyperthreading
2. Only one disk drive
3. Only 8 GB of RAM
4. No CUDA nVidia graphics card

[Guest]

We have a hp z820 with dual 2687w and 128gb ram. We also have an overclocked i7 3930k 4.3ghz I can conviently tel u the hp runs circles around the i7. After effects needs a lot of ram wen u put say 32gb ram for 16 processing core and 16 threads the program gets confused and slow down. But our system has 8gig per coer and 8gig per thread. To sum it up if u have the clients to pay for the system go for it cos dats one machine. It uses a evga classifed gtx 680 4gb, a lsi 9286cv conected to 16bay raid from raid machine. The lsi 2308 in the machine uses the western digital 900gb sas on raid 5 and the intel c602 chipset hosts two samsung 840 pro ssds its scremeanily fast. From cinema 4d to aftereffects to reslove to premier it eats anything u throw at it. So if u have the client get a beffy system. Espically wen u have to deliver in hours.

[Alex_-_DV411]

We have a hp z820 with dual 2687w and 128gb ram. ...  the intel c602 chipset hosts two samsung 840 pro ssds its scremeanily fast.

What's the boot drive - one of the Samsung SSD drives?

If you don't mind me asking, what are your boot (cold to login prompt) and shutdown times?  The fastest I could get a Z820 to boot was 1:01 (shutdown at 0:05).  I did tweak BIOS settings but didn't do anything in Windows yet.  Time to "Starting Windows" is 15-17 seconds - so BIOS init isn't all that slow - it's Windows starting that is.  My feeling is, Windows goes through a lot more on a Z-series vs. your regular desktop - something to look into.

[Guest]

Ours is within that range. U know 2 pro cessors comunicating. Many pci devices the gtx 680, a red rocket, blackmagic extreme 3d, lsi 9286cv all has to be registerd, fact is the more components u have on a system the slower it takes it to boot up. Dats a bbrand new macbook air with the highest processor and ram option would always be faster than a macbook pro with the highest processor and ram option. Dat typa analogy. The hp has to register so many components dats why it boots up slow.

[LeoGrey]

A couple of quick google searches will quickly debunk the "xeons run circles around the single cpu" myth.. In fact, the single cpu outperforms the pair of xeon's in multiple areas within CC. You could spend 40% more on a dual xenon workstation and only ever see a 2-5% increase in performance within Adobe's software suite. High end Autocad / 3D / Float point calculation rigs dealing with large math and complex geometry is where workstations shine.

Put simply, the amount of processing power is RARELY the bottleneck of performance. Storage is a far more important, albeit often overlooked area of the system.

[MarkWeiss]

Even throwing $30K at a dual Xeon machine doesn't seem to solve the 4K playback problem.

How can I get Premiere to use more than 3% of the CPU when playing 4K material? My CPUs are idle during playback and the clocks are at resting speed most of the time (1200MHz) when playing 4K video in Premiere. This same machine tears through Maya BiFrost simulations and Mentalray renderings with breathtaking speed and good CPU utilization.

In Premiere, Taskmanager is showing just a blip across 32 processors occasionally, not fully saturated, yet Premiere is jerky and stuttery on playback, where as my 8 year old NLE with a Core2Quad plays smoothly with no dropped frames the same piece of footage. I shut off Mercury Playback and it got 20X worse. There must be something else that needs to be tweaked in order to make Premiere get the CPUs attention.

[Web Magi]

Additional info: http://tv.adobe.com/search/?q=High+Performance+Workstation

[Guest]

Yes and they (Adobe) recommend dual XEON processors for those hard core workstations. I am not sure how Mylenium means an i7 with max 6 cores (12 with hyperthreading) will be better than eg. a dual XEON E5 2680 setup with 16 cores (32 with hyperthreading enabled).

But obviously this is more complex than I thought..

[MarkWeiss]

After having built a dual Xeon workstation, and discovering that Premiere CS6 runs worse on it than it does on the Core2Quad that it replaced, I'm not sure that Adobe is all that compatible with multiple core / CPU machines. I get erratic playback, dropped frames, playback speed going to 8X by itself after five minutes of playing at normal speed, then stopping. Sometimes playback starts by itself with no user input. I've tried seven different nVidia Titan X drivers, but none give me the performance I enjoyed on my 8 year old workstation that this replaces. Sometimes too many cores spells havoc with otherwise reliable programs.