e
  Premiere Pro Benchmark--PPBM4
   Conclusions
  

    Updated 12/28/2009





Home
Hardware Characteristics
Conclusions
BENCHMARK RESULTS

Mac problems

My conclusions are based on extensive benchmark testing over now a four year period with quite a few of my own personal systems and results from over a hundred of you that have submitted PPBM result


THE PURPOSE OF PPBM4 IS A HARDWARE EVALUATION TEST, WITH IT YOU AND I CAN FINETUNE YOUR HARDWARE TO BEST RUN
PREMIER PRO.

Background

T
here are three tests in PPBM4 identical to the earlier versions.
  1. Render the Timeline to create Preview files (Pressing Enter).
  2. Export the Timeline with the Adobe Media Encoder to a single Microsoft DV-AVI file.
  3. Export the Timeline to a MPEG2 DVD file.
These tests were selected because they fully test the performance of the disk subsystem and the CPU system. 
  1. The Export to DV-AVI is very highly a disk intensive test. 
  2. The Export to MPEG2 DVD is very highly CPU intensive test.
  3. The Rendering of the Timeline is a combination of the CPU and the disk subsystem.

Methodology

The Total Time score is the sum of all individual tests, but testing has shown that ranking of results, based solely on time, puts a heavy impact on MPEG encoding and far less on I/O related activities. To get a feel for real-life performance a new performance index has been added, which will show you your system's performance relative to the top-ranking machine in daily life.

To avoid CPU centric or disk centric systems, the top-10 systems have been used to calculate a 'normalized' PPBM score, that takes away the impact of heavy CPU or disk centric systems. This normalized score is the foundation of the Relative Performance Index. Using weights for the three different tests, normalized scores for each machine are calculated and indexed to the top-ranking system to show the difference in performance.

I
In addition, the results are now presented in an easier to read format. For each test the time result is now preceded by an indicator representing the relative position of an individual score in comparison to all other scores. When all results are ordered from high to low, the top 10 percent, upper decile, are indicated by D9. The upper quartile are indicated by Q3 (between 10 and 25%), the results between the median and upper quartile (25 and 50%) are shown as Med, the results between the median and the lower quartile (50 and 75%) are shown as Q1 and finally all results below the lower quartile (75 to 100%) are shown as D1.  This is all explained in the legends on the results page.


In this way it is easier to see where room for improvement can be found.

How To Interpret The Results

If your Total Time Score is relatively better than your Relative Performance Index that means your I/O system is lacking, either in disk setup or memory, if your MPEG encoding is better than the rest of your scores, it is not your CPU, if your AVI scores are lower than the other scores, your disk setup is the culprit and the other way around.

If you never render a timeline for preview purposes attach more weight to the Total Time Score than the RPI. If your activities include dynamic linking to other Adobe applications, RPI is more important. If you are a heavy user of dynamic linking, render scores are the overriding factor.


My Hardware Design Conclusions
  1. Absolutely most important is the CPU, right now that is the Intel i7 processor or the dual processor the Xeon E55xx series.  As you can see from the performance data the speed of the processor is also very significant.
  2. The Adobe minimum basic disk system is a absolute minimum of two 7200 rpm disk drives..  My personal preference is for a 10,000 rpm drive for the Operating System/Applications disk and a RAID array for the project files.  Specific functions like a separate dedicated drive for writing Output files or Preview files are of less value as it may just slow things down compared to a high performance RAID, the only way to tell is run the PPBM4 benchmark.
  3. With CS4 4 GB of RAM is minimum and optimum for single tasking operation, for multitasking more memory and a 64-bit Operating System are essential.
  4. Just for Premiere there is no reason to spend extra money on special graphics cards, this is not true if you intend to use After Effects which uses OpenGL and there it is better to have a high performing OpenGL is significant.  If you have unlimited financial resources and want to spend $1700 there is the nVidia Quadro CX with a software bundle from Elemental Technologies that may speed up h.264 encoding. I have yet seen any completely specified benchmark that could be tested against the current CPU performance.  Note: This conclusion will change drastically when Adobe releases CS5
  5. From my identical hardware platform tests it is apparent that Vista 64 is faster than XP Pro (32) and that Win7 64-bit is faster (overall) than Vista 64.  .


Because of new data this area below is under major revision


CPU Testing Results
From the data on the Benchmark Results page is it easy to see that unless you are willing to spend thousands of dollars more for the Dual CPU Xeon systems that currently the i7-9xx family it the ideal CPU for non-overclocked Premiere CS4.  It also offers the possibility of overclocking as Harm and others have demonstrated.

Disk Testing Results
I have run many many different combinations of  disk locations and permutations.  My most recent runs were with a SSD (Patriot Torx M28 128 GB)  I was hoping for some breakthough numbers because of the SSD's 0.1 millisecond access times.  I was terribly disappointed.  While the HDtach performace was about 250 MB per second average read rate, the results were not as good in the single disk trials as my 600 MB per second SAS RAID 0 array. Also I am terribly disapointed that Patriot Memory has no support with this SSD for Windows 7 TRIM function and no seperate utilities for cleaning the drive and they discourge using Diskeeper HyperFast utility.
 
Memory Testing Results  TBA
Conclusions  TBA



Old Disk Testing Results
All data below is the result of early CS3 testing does not represent the much better results from current benchmark data but can be used for relative comparisons.

Preface
My benchmarking system is a little exotic.  Ten removable disk drives with a library of over 20 drives, an Areca ARC-1680i controller and 8 cores. Here are the HDtach Average Read Transfer Rates of some of the disks and arrays in quite a few different configurations.
Max Transfer Rate
What I did next is to measure the PPBM disk intensive test (Export AVI) under most of the configurations above.where the entire Project is on a single disk or array.  Here are the results:

Table-1


Plotting the results of HDtach Average Read Rate versus Export AVI encoding rate gives this curve.

Plot

As you can see anything with a transfer rate of over 200-300 MB/s will only provide diminishing improved results on this aspect of the system

There may be other factors involved and I will have to investigate those further before finishing this final posting. 
I will also be exploring more disk configurations
 
On this computer with this version of the DV-AVI disk intensive benchmark I have recorded scores ranging from 41 seconds down to 3.5 seconds with all Project files going to a single Project disk or RAID array, no change in the OS disk or any other parameters. You can see my results in the Table above.  Now most people will not be creating an array of 5 SAS disks but similar results have been achieved with a larger array of much lower cost SATA disks.

Notice that there is essentially no performance difference between RAID 3 and RAID 5 that my benchmarks can determine.

The next Table below shows what happens when a separate disk or array is used for the Output files.  With the 5 disk RAID 0 SAS array or a 5 disk RAID 0 SATA array for all the Project files except the Output files you can see that there is no significant difference in performance, there might be a slight improvement for another fast array.  There are three tests of each SAS configuration (first three colors) to show the consistency, normally I just average a series of tests and present the averages.  This testing shows that with a very high performance RAID array (average read > 500MB/s) there is no advantage to a separate Output file RAID array, and further a single disk for Output files (for the SAS test colored green and the second line of the violet SATA test) actually slows performance significantly.  On the bottom results a slightly slower Project array RAID 5 configuration of the same SAS disks with a two disk RAID 0 Raptor for the Output files shows a slight improvement, but hardly worth the price of the two disks.

Table-2

My conclusion, on a medium high to high performance RAID array for the Project files it is a better utilization to use another disk or two on the RAID array instead of providing a separate Output files drive or array.