Premiere Pro CS4 Benchmark
 

I would like to draw your attention to Bill Gehrke's benchmark program and invite you to actually run the benchmark. The more results, the easier it is to see how your own system holds up or where improvements can be made.

Adobe Forums: PPBM is Back--Benchmarking Premiere Pro...

Thanks I'll run it tonight :)

This is awesome! I don't see a QX6700 on Bill's chart yet so I'll give it a try.

Harm, if I'm not misunderstanding the chart, it looks like Win7 actually did outperform Vista even at a lower (over)clock, but otherwise identical system?

Of course, with an "ancient" QX6700, I've been aching to build a new system. Anyone have a 55XX dual processor workstation with lots of RAM onboard to test? I was thinking about holding off until I could go crazy with a 6-core Gulftown dual processor workstation next year, but I wonder how performance vs cost will be when you get SO MUCH PPro performance out of an OC i920 for so little money? What are your thoughts, Harm?

Pete,

Currently the i7-920 overclocked is the most economical solution and the enormous benefit of a X58 mobo, that is significantly more expensive than a P55 mobo for use with an equally priced and equally performing i7-860, is that a simple BIOS update will allow a simple exchange of a Bloomfield (i7) for a hex-core Gulftown.

I currently have a Dell Poweredge T410 with dual E5520's in the hallway for testing VMware, but have not even unpacked it yet. So no results to give yet.

If you look at PassMark Software - PerformanceTest System Benchmarks - Fastest Performing PCs you will see a number of significant results.

Look for instance at BL#46692 (rank 16), which is a dual W5580, or BL#75006 (rank 20) with dual X5570's and compare these with BL#21147 (rank 15) and you will see the economic advantage of i7's versus dual Xeon's.

BTW, I'm still waiting for the Top 20 list to be updated with my latest Win7 results (BL#106949). When that happens, I will have rank #1 with Vista (score 6055), #9 with Win7 (score 4623) at much lower clock speed (3.44 GHz), and the current score of 4300, now at #15, but then dropping down to #16.

And yes, Win7 is around 10% faster than Vista in these tests.

DavidDwyer, Personal or Computer ID
HP Proliant 115GS, Computer Manufacturer
HP, Computer Model
159.4, secs Total Benchmark Time
7.1, secs AVI Encoding Time
94.3, secs MPEG Elapsed Time
58, secs Rendering Time
AMD, CPU Manufacturer
2.1GHZ, CPU Model
2.1, GHz CPU speed
1, Number of CPU chips
4, Total Number of Cores
8, GB RAM
4.1, APP Version PPBM4 DV
Win 7 64bit, OSVersion
SATA, OS Disk Interface
150, GB OS Disk Capacity
10,000, OS Disk Speed
SATA, Project Disk Interface
400x2 R0, GB Project Disk Capacity
7200, Project Disk Speed
Nvida FX3700, Graphics Board
, Comment


Tried to email but bounced back

David,

I copied your results and have mailed them to Bill. Thanks for your effort.

Your results look very good for an AMD system. Once they have been included, I'll have a second look.

Ran my system two different ways and emailed the results:

- First was default PPBM which used my D drive (older 10K Raptor) for all files
- Second I changed the location for preview and output files to a 7200 SATA RAID0 750GB x 2 drive

Text files attached. I was a little surprised that even though it is onboard RAID at 7200, it was still a little slower than writing all the files to a single 10K drive.

Ok, Harm, now set up that dual E5520 system and give it a go. Gotta know!

Pete,

I took the liberty of mailing your results to Bill.

Looking at your data, I tried to look up your mobo, but the 6xx series is no longer available, so I am merely guessing about an explanation for the differences between your results.

The differences are pretty small, but the relatively compute intensive parts of the benchmark, rendering and encoding to MPEG2-DVD seem to suffer the most from using the nVidia chipset and raid capabilities. Maybe the speed advantage of the raid is more than offset by the overhead of the nVidia chipset for the raid, having a relatively high CPU load. Also, the Raptor is by nature very quick and depending on your 750 G disks (16 or 32 G cache) may have a significant advantage for the relatively small files used in the benchmark.

From the looks of it, some further tuning of your system (services turned to manual, processes turned off) may give you some more responsiveness.

Naw, what I need is the latest Monster Computer.
;-)

Don't forget to benchmark that E5520 -- I was so sure you'd immediately unbox it, set it up, and run the PPBM before taking a bathroom break that I could hardly sleep last night. I was so very deeply disappointed, perhaps clinically depressed, to wake up this morning and not see the results posted.
:-)~

Well, maybe it isn't THAT urgent, but am interested in seeing what it'll do. If the dual processor machines don't give all that much improvement for the large increase in cost, I may break down and build an i920.

Pete,

Up to late last night my son and I were up to discuss the installation of the disks in that system (whether to install 1, 1.5 or 2.0 TB disks, how to install more than the default 6 disks, how to preserve cooling charateristics, etc.), contemplating photos of the interior, arguing about the benefits of passive CPU cooling versus active cooling, the lack of fans in the default system, the way Dell had done the SAS connectors and the like. When I finally called it quits, my son was still busy on installing the system. I will ask him to run the PassMark test but since it is his system, I don't think I can get him as far as to install PR and run Bill's benchmark. (Much of our discussion went over our VPN connection). BTW, Bill has received your results and will try to get them posted by tonight.

Another side remark, PassMark informed me that they do not accept results in the Top-20 list from the same machine, even if running different OS'es, so they will not publish my results (4623) while running at 3441 MHz clock speed, because my machine already holds the top position (6055) while running at 4195 MHz. I argued that it might be interesting to bend the rules a bit, since my system was the only one of all the single CPU systems running at far below 4000+ MHZ in the Top-20. Although they conceded that it was interesting, it would give them serious headaches to bend the rules for me. So no publication of the results. You have to do with the image in my previous post.

Bill Gehrke's results, as well as Averdahl's will give you an inkling of the performance of dual Xeon's versus a single i7, as do the PassMark results. If you really want top-notch performance, wait a little bit for SuperMicro to come out with dual Xeon hex-cores somewhere next year. However, if you want to be at the forefront of the technology curve, be prepared to pay for it. The other drawback is the limited functionality of the installed BIOS, which makes overclocking rather difficult.

I'll keep you informed on our progress with the T410 system.

Harm: I noticed that you have the 1680ix in your pc. I just added one to my pc that also has a 3ware 9650 in it and I also setup a Matrox Mini around the same time. But now my system has a serious lag when clicking on folders and opening apps. I am using a 1 month old Vista x64 installed on an Intel X-25 80GB SSD, ASUS P6T & i7 920. I have 7 x64 Ultimate and am about to install it but I would like your feedback on any problems you have had with your 1680ix in Vista and 7.

I uninstalled all Matrox software & drivers after 2 days of use so I am hoping something got screwed up from the Matrox drivers. (I didn't like the lag in Premiere while using the Mini)

One other quick question about your ASUS WS board: I didn't get it because I had read many people were having problems with Raid cards. Have you had any issues related to your board? I ask because I realized that I need more pci express slots and am considering that board.

Thanks

Steve can you tell me about these Matrox cards? What card do you have and how much does it speed up your video editing? Is it worth the money?

...and while we're picking brains, a question I've been meaning to ask the seasoned overclockers around here: do the processor and most/all of the motherboards allow the QPI on the i920 to be overclocked as well, or does that stay at 4.8 despite increasing the clock speed of the processor?

Not sure how much it matters, but I was wondering about that in terms of possible impact to AE OpenGL or other software using the graphics card to speed tasks; perhaps the slower QPI could get saturated?

Steve,

I have not experienced any problems with the Areca controller. In fact I am quite happy with it, since it gives me a sustained read transfer rate of over 850 MB/s and a sustained write transfer of around 600 MB/s, regardless of fill rate, and 10 TB of net space (12 TB gross in raid30). I have not experienced any delays under Vista 64, nor under Win7 64. Just today I installed the official Win7 drivers for the 1680iX, no hiccups, and a slight performance improvement over the Vista drivers, around +7%.

Pete,

On my system, which is somewhat overclocked to 3.6 GHz for the CPU and memory to 1440 MHz, I use a QPI link of 3240 MHz, which should be multiplied by 2 to get the effective QPI rate, so 6.3 GHz. Most mobo's give you endless possibilities to fine-tune your overclocking and voltages, but I have a simple rule: Keep voltages as low as possible and adjust the rest (BCLK, multiplier, turbo mode, etc.) to give maximum stability and longevity. Then edit along. Gamers would probably cry out loud at my simplistic approach, but it works for me. While trying out different settings, I keep a close watch on voltages and temperatures and check stability with things like Prime95 to ensure the system remains stable under heavy load. I don't want to have a system stall halfway during a long render...

Thanks, Harm. Yup, with its overclockability (is that a word?), certainly seems a better bang for buck than the i975, even though that can be overclocked a bit, too. Lots more money for a little more performance.

And I see that Bill has already updated the chart on the PPBM web site to include my system runs. You guys are on top of things!

David: using the Matrox Mini with Premiere CS4, my editing was actually slowed down quite a bit due to the lag in switching from Source monitor to Program and vice versa. The Mini took about 2-3 seconds to switch video source and then another 3-4 seconds before I could click on anything. I like to set the in/out in Source and drag to the timeline and then go back to Source. The Mini added about 15 or so seconds to this process. Plus, you must use Matrox's presets for a timeline and opening this sequence in a non-Matrox system caused Premiere and AE to crash.

Harm: thanks for the reassurance. Also, what cpu cooler are you using and what are your cpu temps during full load? My Zalman 9500 can only keep temps around 60c according to Core Temp so I haven't explored overclocking yet.

While we are on the topic of performance, I must say that once you go SSD, you will never go back (provided you are using a good SSD). I have a G1 Intel X-25 80GB for the OS+Apps, and during boot up, once you see your desktop, booting up is done. Photoshop opens in 2 seconds; I first tested opening Photoshop, Premiere and AE as quickly as I could click on their icons and they all were open in 8 seconds from the first click. I made the switch from 4 74GB Raptors(10k rpm) in Raid 10 and its night and day. The only bad part is now I can't stand using anyone else's computer because I must wait.

Cheers Steve,

I'm having slow rendering compared to most other people and I'm wondering if a SSD or even a Matrox CompressedHD card would speed things up.

SSD is the easy option but once the application is open would I see any performance gains?

Pete,

Since Bill did not have your email adress, he asked me to pass this message on to you:

Harm, I do not have a way to contact Pete Bauer, maybe you could pass this on to him.



[I]Pete, your results are posted. Thanks for being persistent enough to find Harm’s back door route.

Unfortunately your RAID 0 configuration is not as good as the non-RAID configuration, the big difference was the Render Timeline test. This is a single test and I have found has quite a bit of variation from test to test.

Did you have two RAID 0 arrays or was it one that you used for both sets of files? I would suggest (if feasible) that you put your all the benchmark files on the RAID array and retest. I have never tried using a standard disk for just the project and my RAID 0 for the preview and output. Looking at your scores and my own tests I do not see much difference between two 7200 rpm drives in RAID 0 and a 10,000 rpm Raptor.

Your scores for the CPU intensive testing (MPEG encoding) seems slightly high compared to the VRV test of a 2.4 GHz Q6600. You might see if you could remove any CPU sucking processes that would be wasting your CPU cycles. I am guessing but you might be able to knock as much as 10 seconds off your MPEG score. That would also reduce the Render Timeline somewhat.

Also going to Win 7 64 will also speed up the benchmark by up to 10%

Raid 30? Never looked at this can you help me with a few questions then?

How many disks are you using?
What controller card?

Seems I might have to go changing my editing machine to get some decent speed out of it.

Thanks, Harm. I removed Bill's personal contact info from this public forum (safety first!) and will contact him via the email you listed. I emailed my results directly to the address that PPBM lists and I didn't get a bounce message so not sure why he didn't receive my results directly.

I was just helping populate the chart since the QX6700 wasn't yet represented -- and I was curious where my current system stood. I don't plan to mod/troubleshoot it as it is long in the tooth anyway, although it still does routine editing quite nicely. My main thing is that AE tends to bog it down.

When I decide what new system I will next build, that'll be my performance boost.

I have 12 disks, 1 TB each in a raid30 on an Areca ARC-1680iX-12 controller with 2 GB cache and a BBM (battery backup module) with expansion capabilities for another 4 disks. Due to the nature of raid30 (2 raid3 arrays of 6 disks each and then striped to form a raid30) of course you lose 2 disks for parity, so the effective net space is only 10 TB.

Pete, thanks for the thoughtful removal of Bill's contact data.

To give you an idea of how screwed up my pc is:

Everything on 4-drive Raid 10 (Areca 1680ix)
Total: 73.6s
AVI: 5.4s
MPEG: 46.2s

Output now on 2 150GB Raptors(connected to 3ware Raid card)
Total: 79.5s
AVI: 11.8s
MPEG: 45.7s

During the first run, Vista had an error: a Host Process stopped working, the screen flashed and changed from Aero to regular. Then it flashed again during the end of the first run and was back to Aero.

This is with an i7 920 @ stock clock and 12GB ram.

Windows 7, here I come!

Steve,

Once you have submitted your data to Bill and they have been incorporated in the listing, it is easier to give you some feedback, but the results are not overwhelmingly good in comparison to others, even though the 22 second render time is pretty good. The AVI test on the 3Ware is a bit disappointing. This can be caused by background processes or services, or just by the 3Ware. The Areca gives much nicer results, especially with the latest drivers and firmware.

David: could you post your current computer details and I can tell you what might be slowing you down.

About the Matrox CompressHD, it only assists the encoding, not the rendering. If you are using multiple filters/effects, then the CPU is the most important for speed.

I just learned about flashing my nvidia 8800GT to function as a Quadro FX3700. With Premiere CS4, there is a plug-in available - Elemental Accelerator for $250 which uses certain Quadro video cards(incl FX3700). The plug-in is $250 and ebay has 8800GTs for $70-80. I am thinking of trying this route as I don't trust Matrox drivers anymore. There are some reviews of this accelerator including More Info - Elemental Accelerator 2.0 - Authorized Reseller - Safe Harbor 800-544-6599

To speed up AE rendering, there is Nucleo Pro. I am about to buy this program as it greatly speeds up rendering by always rendering the timeline while you work.

heh Steve I was the one who posted the 8800 to 3700's.

Hp Proliant 2.1ghz Quad Core
8GB RAM
150GB 10,000 OS
800GB Raid 0 Project
750GB Scratch Disk
FX3700/8800

"heh Steve I was the one who posted the 8800 to 3700's."

Duhhh! Dummy me, I forgot.

What processor(s) is in the HP?

Looking at benchmarks, the Intel i7 920 with 12gb ram can keep up with the prior generation dual-quad core systems.

Your hard drives certainly aren't slowing you down. I am assuming you are using Vista x64 and Premiere CS4.

Steve,

I'm using Windows 7 x64 and the HP has

AMD Opteron™ Processor Model 1352 (2.1 GHz, 75W, 2MB) Quad Core.

The machine was only purchased for $500 so its cheap editing but my benchmark times doesnt look good at all!

Yep, the i7 920 plus 12gb ram would be much faster. The i7 has been one of the largest increase in performance from a prior generation. Then, sometime next year(first half of 2010), Intel will release their hex-core i7 so increasing performance can still be done without buying a whole new system.

I'll wait to early next year and will also see Nvidias Fermi card as well.

Stephen: start a new thread and copy our last few comments to get it started.

MOD NOTE: As suggested, moved posts on networking to a new thread:
http://www.dvinfo.net/forum/non-line...p-editing.html

Your results have been included and from the looks of it, your CPU intensive testing, the MPEG encoding, is the culprit. You AVI results are quite good, but the AMD CPU has trouble keeping up with the faster Intel CPU's. Your disk setup is quite good, hence the good AVI performance. You lose on the render test and the MPEG test, which are largely CPU dependent. You may be able to gain marginally from an SSD, but IMO it would not be a wise investment. If you want to do something about responsiveness and perfomance, you have to look at a different CPU.

DavidDwyer, Personal or Computer ID
HP Proliant 115GS, Computer Manufacturer
HP, Computer Model
86, secs Total Benchmark Time
9.2, secs AVI Encoding Time
18.8, secs MPEG Elapsed Time
58, secs Rendering Time
AMD, CPU Manufacturer
2.1GHZ, CPU Model
2.1, GHz CPU speed
1, Number of CPU chips
4, Total Number of Cores
8, GB RAM
4.1, APP Version PPBM4 DV
Win 7 64bit, OSVersion
SATA, OS Disk Interface
150, GB OS Disk Capacity
10,000, OS Disk Speed
SATA, Project Disk Interface
400x2 R0, GB Project Disk Capacity
7200, Project Disk Speed
Nvida FX3700, Graphics Board
, Comment

I reran the test but changed the output disk back to the 10k rpm OS Disk - Improvement!

David,

What happened with your MPEG2-DVD results? It appears like there is some mistake going on here. 18.8 seconds is in no way a reasonable comparison to you AVI results, nor to the render results. Is there a typo? How could it be that your AVI results are around 6 times slower than the current top score, your MPEG results better by around 50% and your render results are again around 2.5 times slower than the top performance. Something does not sound right here.

Don't think so but I'll rerun it when I get home tonight to double check.

I'm trying to run this on CS5 but it doesnt like my output files directory when I run the script?

"C:\Video Capture Files"

Sorted it - Pr set the AVI files as _11, _12 _13 etc so reran it with the correct names.

Tried to send my results through but got a mail bounce back

David,

Nice improvement. Don't worry about the increased AVI times, that is something all CS5 users have seen, but the improvement in MPEG encoding and render times are quite good, don't you think?

You could try the soft-mod on your FX3700, it may bring down your render times at the detriment of your AVI scores, but better overall.

Did you send these results to Bill or do you want me to include them?

PS. Your disk setup changed in comparison to your current results. 75 instead of 150 G OS disk, 400 x 2 R1 instead of R0 for project disk. Is this correct or was it wrong to start with?

Softmod? I should of said its actually 8800 GTX that that was flashed to the FX 3700

OS is 150GB 10k RPM
Project is 400x2 R0 7200rpm
Scratch is 750GB 7200rpm

I can't send the email to Bill so if you could pass it on?

Reason why I ran the test again is I have a Intel i7 with 12GB of ram coming tomorrow!! Will be OC'd to 4GHz. Same disk setup with a GTX 275 for MPE goodness.

David,

I'll add your data to the results and send them on to Bill.

By softmod I mean a simple "hack" to make CS5 believe you have a certified MPE card. I did that with my video card and reduced my rendering time from 14 to 3 seconds. Here is a link to some instructions, although it has been posted here as well: Adobe Forums: Unsupported GPU for CS5

This may be informative as well: Adobe Forums: Overclocking the i7, a beginners guide