Premiere Pro is a suite of serious software for video content creation and is very popular in the professional market.  It is also one of the 7 high end applications certified by Nvidia by June 2012 to run Maximus.  What is Maximus and why only 7 applications in the world are certified?  This article is about the findings we obtained recently through 2 months of in-house testing.

Maximus is a scheme using a combination of CUDA enabled GPU hardware to handle visualisation and computation in the same workstation (PC).  Quadro is designed for visualisation (interactive design) as the main purpose, whereas Tesla is designed for computation as the main purpose.  Normally a 3D content creation workstation would be fitted with Quadro, and a high performance computational station would be fitted with Tesla.  A Maximus workstation has both Quadro and Tesla and employs their unique designs for their respective top performances. Instead of waiting for Quadro to do the computation parts, Tesla can do them better and in parallel.  The result is a much shorter turnaround time of design tasks. 

In order to benchmark the additional performance of Tesla to Quadro, our first target is Premiere Pro CS 5.5 due to its popularity.  We tested PPBM5 which is a standard and well documented benchmarking application to find out the time to complete selected functions of Premiere Pro by various computer and GPU hardware.  There is in fact a PPBM5 website which explains what the tests are about and it records all test results submitted by whoever care to release their results to the website.  PPBM5 is therefore a good starting point.  Nevertheless we must remember that PPBM5 is not Premiere Pro, and time to complete PPBM5 test sets does not equate to test to complete Premiere Pro data sets.

We tested Quadro 600, Quadro 2000, Tesla C2075 separately and singly, and then Q600+C2075 and Q2000+C2075 under Maximus.  The price of the GPU hardware goes up in the same sequence as listed here.  The performance can be said to follow the same sequence but not quite.  The combination of Q600+C2075 produced the same performance as Q2000+C2075.  This implies the extra CUDA cores of Q2000 over Q600 are useless as far as PPBM5 is concerned.  Another observation is that the increase in performance is not in the same ratio of the increase in price.  This reinforces the same implication as above.  If we look at the PPBM5 website for scores submitted by various people around the world, the top scores were achieved with GeForce cards which are much cheaper than Quadro.  This is another reinforcement of the same implication.

These observations have a huge limitation to be useful.  The limitation is that the PPBM5 is not Premiere Pro.   PPBM5 may not have been tuned to respond to Quadro and Maximus as Premiere Pro has.  The exercise however has indicated the “without-tuning” scenarios for running various GPU on Premiere Pro.  Nvidia website has good information on how tuned applications would perform.  The chart shown here was obtained in June 2012 from http://www.nvidia.com/object/media-entertainment-with-maximus.html

These are the other 6 applications certified for Maximus by Nvidia by June 2012

- Autodesk 3ds Max 2012
- Solidworks 2011
- PTC Pro/Engineer 5.0
- Dassault Systemes Catia V5/V6
- Simulia Abaquo 6.11-1
- ANSYS Mechanical V13SP2/V14