Skip navigation
Currently Being Moderated

Now that CS5 has arrived, what does it mean for your system setup?

May 7, 2010 2:03 AM


Apart from nice added features, improved stability and increased speed, there are two major factors that impact on your system setup:


1. 64 bit


2. Mercury Playback Engine


64 Bit only


This means a couple of things:

 

You need a 64 bit OS, you need 64 bit plug-ins and you can use much more memory than the 4 GB limitation that applied to 32 bit OS'es (with effective memory in the 2 - 3 GB range, depending on the Boot.ini switch).


In practice, to benefit from the move from 32 bit to 64 bit, the recommended memory needs to be around 8 or 12 GB, depending on your mobo and chipset and the number of DIMM slots available. With current prices (May 2010) it is not yet economical to opt for 4 GB sticks.


If you have a dual socket board with 12 or more DIMM slots, 24 GB is optimal.


Mercury Playback Engine


This is where it gets interesting.


In the past, up to and including CS4 4.2.1, performance was largely impacted by:


1. CPU


2. Memory


3. Disk setup


4. OS & Tuning


5. ...


and finally by the video card. It did not really matter what video card you had installed. There never was any discernable performance gain from expensive video cards.


The tables have turned.


With CS5, installing a CUDA enabled video card has a large impact on performance and lessens the burden on the CPU. The GPU does a lot of the work.


However, since the CPU gets a lot more breathing space, the CPU will no longer be the primary bottelneck, as was often the case in the past.  And since all DIMM slots in the average machine will be occupied, it is a costly exercise to exchange the DIMM sticks with larger capacity ones. People who have only 3 slots in use on a X58 motherboard can directly benefit from adding three more sticks, but this does not happen very often.


So with the CPU running with a relatively low load and memory more or less a given, the only thing that may be a bottleneck to be improved upon is the disk setup. This can be seen easily by scrubbing fast through the time line with the popular AVCHD material. Due to the MPEG nature, while scrubbing one needs to look backwards and forwards to create the image under the CTI and that means a lot of disk activity. CPU is not a bottleneck, memory is not a bottleneck and the GPU is not the bottleneck, it is the disk setup.


Do not be fooled by the claims that SATA2 or SATA3 have enough bandwidth to support the data rate of AVCHD or other MPEG streams. Tests have shown that significant performance gains can be achieved by using a large number of disks, preferably in a raid. This is especially true when one has multiple tracks.


SSD's are not yet an economical alternative for conventional hard disks, with prices that are a factor 30 - 50 higher per gigabyte than conventional disks.


Conclusion


To fully benefit from the performance gains that CS5 allows, it may be that using a MPE supported card will lead to the disk setup being the new bottleneck. Be aware of that fact.

 
Replies
  • Currently Being Moderated
    May 7, 2010 2:18 AM   in reply to Harm Millaard

    Another great summary!

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 7:24 AM   in reply to Harm Millaard

    Harm --

     

    Very useful piece. Would also be good on the FAQ list. I can add it there, but it would be better if you did it, so it has your name up front.

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 8:53 AM   in reply to Harm Millaard

    i have to disagree. the bottle neck (once using 2 drive raid 0 (2 sets) )

    is not the drives. at least untill you get to uncompressed HD.

    even with Red2k we did not see an improvement going from 2 to 4 drive raid 0

     

    in fact it seems to be CS5 itself.

    Scott

    ADK

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 9:13 AM   in reply to Harm Millaard

    If it's the disks causing the bottle neck then it's either the Random Access Time(latency) or the controller(Raid or AHCI). During the benchmarking so far on AVCHD or even Red 4K, the disk usage has been allot lower than what it can handle. AVCHD whether on playback or export to H264 averaged less than 10MB/s. For the Export to H264, the usage stays under 10MB/s until the very end when it actually writes the file. At that point I have seen the usage fluctuate between 90MB/s to 150BM/s although it normally stays around 50MB/s. This was the case on both the 980X system and the Dual Xeon 5660's I just finished the AVCHD benchmarks on. I am currently benchmarking the R3D on the Dual Xeon but so far, I have not seen a difference in the disk usage from the R3D on the 980X. This means that if the bottle neck is at the drives then it has to be the RAT or Controller. I doubt it's at the controller since the onboard controller has not  shown any difference in performance from the Intel SAS 600 that was used in the 980X system benchmarks. We have some new raptor drives that I will test asap to see if the RAT is causing this bottle neck. Let me know if you have any questions.

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 11:00 AM   in reply to ECBowen

    I need some help or an explaination.

     

    I now have a good number of results running the PPBM4 benchmark on CS5 software (a few published and more unpublished) and as anyone can see exporting a DV test file file to a single DV-AVI file using the preview files is primarily very disk intensive.  Why in CS5 does it take longer to export the benchmark than it did with CS4?

     

    Message was edited by: Bill Gehrke Actually in most cases it doubles the time especially when MPE is enabled

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 11:47 AM   in reply to Bill Gehrke

    at this point we are still trying to figure out whats up.

    bear in mind DV VS AVCHD or Red is a completely different animal. there is a lot to process on encoding vs DV

     

    we think the processors and HDDs are sitting there waiting on CS5

     

    even with Dual Xeon 2.8GHz 12 cores its no faster than the 980x niether are CPU pegging, HDD pegging or ram..

    interesting is the Dual Xeons will use about 10G (with 24G installed) ram vs the 980x using 6 (with 12 installed)

     

    dual Xeons have dual memoery pipe lines 1 for each cpu so we think thats where its coming from

    but we are still scratching our heads with a good deal of what we are finding in these benchmarks.

    most is pointing to CS5 coding.

     

    i am sure Eric will poke his head in here...

     

    Scott

    ADK

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 1:39 PM   in reply to Bill Gehrke

    Until I can test the new raptor drives, I can't rule out the disks as the bottle neck but I don't really believe they are after what I have seen in testing. The CPU load for AVCHD on the Dual Xeon 5660 system averaged around 25% on 3 layer export and 45% on 4 layer export with effects. The ram usage was almost double what the 980X system's 6.5 GB sitting at 10.5 to 11.5GB most of the time. The Disk usage never rose above 10MB/s until the very end where it actually writes the file. Then it went as high as 150MB/s but averaged around 50MB/s a sec. If the disks are the problem, then it has to be the Random Access Time. If not then it has to be the decoding/encoding engine. If I had to guess, I would say the CPU's are constantly caching threads and starting new ones waiting on the GPU to send effects data back in order to actually process the encoding. This would cause allot of thread overhead by the CPU Prefetch engine as it's constantly revising the priority and queue. Something like that would definitely slow the entire encoding down. This is definitely leaning that way at this point.

     
    |
    Mark as:
  • Currently Being Moderated
    May 7, 2010 3:50 PM   in reply to ECBowen

    Scott and Eric, I do not believe it is a disk problem, My theory is it is a bus "hogging" problem and I do not know if it is the drivers or the Adobe software. My GTX 285 arrives next week and I will have some controlled repeatable numbers.  It is not a Random Access Time problem as I think I will see the same problem with my SSD arrays

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 8:56 AM   in reply to Harm Millaard

    Harm,

    I currently have 6GB of RAM installed in my system, would it greatly benefit me if I were to up that to 12GB's?

    Thanks!

     

    SIDE NOTE: I have a MPE supported GPU.

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 4:42 PM   in reply to Harm Millaard

    Greetings all,

     

    I've been reading the threads &  forums regularly since the announced release of CS5 and didn't realize until this thread that ECBowen is Eric of ADK Video Editing, the uber-computer brainiac savant who is building my new system!

     

    Hooray! Lucky me!

     

    i7 980x, 24 GB RAM, GTX 285

     

    I tremble with anticipation of its arrival like a school boy with a crush on his teacher.

     

    Best,

     

    Dailey Pike

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 6:55 PM   in reply to Dailey Pike

    24gig of ram...nice!  What did that run you?  I wish I could've done that.

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 7:35 PM   in reply to tclark513

    24 GB RAM was about $1200 which is very reasonable IMHO.I  hope no one thinks I'm "bragging" about the 24 gigs, it wasn't my intention. I'm just excited about the new system.

     

    You have to understand I'm moving up from a 4GB 32 bit Windows XP.

     

    The beautiful thing about this new system is my CS5 Production Premium software is already going to be installed and tested by ADK who were CS5 beta testers. I'm getting one hard drive for the OS and two Raid 0s since I edit video. In addition it comes with a Paragon hard disk manager to back everything up every night.

     

    I'm jazzed and I apologize for my overt enthusiasm.

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 7:57 PM   in reply to Dailey Pike

    Sounds like  a great system. Let us know how it runs.

     

    I have been editing HD (mxf P2, DVCPROHD,  HD mov, HD AVI, ProRes) for a couple of years now and have never seen the need for RAID in my workflow.  

     

    With my new CS5 system , I7 - 930 , 12GB , Quadro FGX3800...blah, blah, blah...etc...   I see even less need for RAID.

     

    A benefit of CS5 in a sytem like mine is less render files so less disk accessing ( I assume)

     

    Just interested in the big push for  RAID systems.

     

    I guess its driven by AVCHD which ironically is a "consumer" format.  The non pros are buying the grunty RAID systems so they can play and edit their home movies.

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 10:09 PM   in reply to shooternz

    Hey shooternz,

     

    You might be interested in this thread about RAID.

     

    http://forums.adobe.com/message/2795413#2795413

     

    My understanding from what I've read, and what people I trust to know tell me, is having 2 RAID 0s (Capture/Output) speeds up video transcoding.

     

    As far as AVCHD, some people think it's the future.

     

    I shoot on a Canon XHA1 which allows for some fiddling with the Gamma, Knee, Black, etc... and really like it.

     

    Also, Sony is supposed to introduce an "affordable" 35mm video cam next year.

     

    I wonder what they consider affordable?

     
    |
    Mark as:
  • Currently Being Moderated
    May 8, 2010 10:20 PM   in reply to Dailey Pike
    My understanding from what I've read, and what people I trust to know tell me, is having 2 RAID 0s (Capture/Output) speeds up video transcoding.

     

    That supports my point......

     

    I dont VIDEO TRANSCODE.  (I edit NATIVE in a professional environment without issue and export simply and quickly)

     

    So...No need for RAID for me.   But... it seems that those using consumer formats have the need for heavy weight hardware to enable them.

     

    BTW - I had followed that thread.  One of a few that had made me post my thoughts and experience in a world without RAID.

     
    |
    Mark as:
  • Currently Being Moderated
    May 10, 2010 6:47 AM   in reply to shooternz

    In ours and others testing and benchmarks, a 2 drive raid 0 gave the ideal performance for most codecs. You gain a performance boost up to about 150MB/s a sec or a little higher and then little to no returns above that accept for uncompressed codecs or DPX format. AVCIntra will likely have a higher Ceiling as well since it has the highest bitrate out of the native Adobe codecs. I will be getting some raw AVCIntra to test from a client of ours soon. R3D material definitely makes use of the 2 drive raid 0 and would need higher dpending on how many layers of video you might be dealing with. For 3 or 4 layers, the 2 drive raid 0 was enough. Right now that is the breakdown for most of the codecs. Uncompressed of course requires atleast 4 drive raid 0 or 6+drive raid 5 or 6. 8 Drive raid 5 or 6 would be ideal though.

     
    |
    Mark as:
  • Currently Being Moderated
    May 10, 2010 6:49 AM   in reply to Dailey Pike

    Thanks for the comments. Please let me know if you have any questions about your configuration.

     
    |
    Mark as:

More Like This

  • Retrieving data ...

Bookmarked By (0)