I have run PPBM5 44 times for various combinations of CPU speed, hard disk speed, and RAM on the same CPU/motherboard and will report specific conclusions in an effort to help fellow users build cost effective PCs to run CS5. I have gained a lot from this forum, but have posted less than 20 times. I guess one could say this is my feeble attempt to pay it forward.
I'd like to thank first and foremost Bill and Harm for providing and supporting the PPBM5 benchmark. This provides an excellent common reference for everyone that measures different aspects of Premiere Pro CS5 performance. Next I'd like to thank Eric Bowen with ADK whose guidance has helped me a lot; many companies that build PCs for a living do not participate at all on forums. Finally, thanks to the many other contributors to this forum - there are too many to name!
So what have I learned and how can it be communicated here is a way that is useful?
I'm going to report specific PPBM5 results that reflect various hardware changes to a common motherboard and CPU and will focus on those results that significantly changed performance. Harm preaches the truth that a fast system requires a full complement of sub-systems; this is indeed what I've found.
When I comment on:
CPU speed - RAM size and drive performance has already been maximized,
RAM size - CPU and drive performance has already been maximized,
Drives - CPU speed and RAM size has already been maximized.
Base system used for testing:
Gigabyte GA-P55-UD4P, Intel i7-860 CPU, MSI Hawk GTX 460 1GB video @ 850MHz, OS/boot array is 2x80GB Intel X25-M SSD in RAID 0 (Win7 64-bit OS, CS5 v5.0.3)
1) CPU speed; important and surprisingly affects all 4 PPBM5 tests, including the GPU MPE test; I saw very linear improvements going from 2.8GHz (stock) up to 4.0GHz (OC) [PPBM5 composite score: 268 seconds stock, 195 seconds at OC]
2) RAM size: increasing from 4GB to 8GB doubled the MPE GPU performance [11 sec. vs. 6 sec.] and improved the MPEG2-DVD performance [122 sec. vs. 91 sec.]; the other tests only improved marginally
Increasing from 8GB to 16GB made significant additional improvements to the MPEG2-DVD time [91 sec. vs. 37 sec.]; all other tests hardly changed at all
3) Drives: Slowest tested was project, cache, and cache DB all sharing a single 1TB WD 7200rpm drive [PPBM5 Disk I/O test = 138 sec.] Changing to two of the same drives in a RAID 0 configuration was a massive improvement [PPBM5 Disk I/O test = 82 sec.] Bleeding edge drive arrays running on Areca controllers produced the fastest result, but were only 7 seconds faster than the 2x1TB RAID 0 array [PPBM5 Disk I/O test = 75 sec.]; a 5x64GB Crucial SSD Raid 0 array performing similar to a 8x1TB WD 7200rpm RAID 5 array.
SSD sidebar: Using a 2xSSD (Intel X25-M) RAID 0 array for the PPBM5 project was slower than the 2x1TB 7200rpm setup [93 sec. vs. 82 sec.], however I will be using a SSD array in my upcoming build for the OS and programs. They just make the system feel more eager and responsive, similar to the way a good handling car feels more responsive than a luxury car or a truck. Note too that all SSDs are NOT created equal. My small (64GB and 80GB) SSDs have lightning quick reads and slow writes (250MB/s reads, 70MB/s writes), whereas the 256GB EXPENSIVE models often write 4 times the speed that mine do. Good 1TB 7200 hard drives tend to read and write at around 130MB/s.