Page 1:How Should You Accelerate Adobe?
Page 2:The Mercury Playback Engine
Page 3:Hacking CS5 For Cheaper Acceleration
Page 4:Test Configuration
Page 5:After Effects CS4
Page 6:After Effects CS5
Page 7:Photoshop CS5: Scaling And Rotation
Page 8:Photoshop CS5: CUDA
Page 9:Premiere Pro CS4
Page 10:Premiere Pro CS5 (Render)
Page 11:Premiere Pro CS5 (AME Export)
|Motherboard||Gigabyte X58A-UD7, LGA 1366, BIOS F7|
|Processor||Intel Core i7-980X Extreme (Gulftown), Six-Core, 32 nm, 12 MB Shared L3 Cache|
|CPU Cooler||Thermalright Ultra 120 eXtreme|
|Memory||OCZ Gold-Series DDR3-1333, Triple-Channel, 3 x 4 GB, 1333 MT/s, CAS 9-9-9-20-1T|
|Graphics||Sparkle GeForce GTX 480 (SXX4801536D5-NM), 1.5 GB GDDR5, 700/3696 MHz GPU/Memory|
|Storage||Primary: Intel X25-M (G2) 160 GB SSD|
Secondary: OCZ RevoDrive 120 GB PCIe x4
|Software and Drivers|
|Operating System||Microsoft Windows 7 Ultimate 64-bit|
|DirectX Version||DirectX 11|
|Graphics Drivers||Nvidia GeForce 258.96 WHQL (7/19/2010)|
|Adobe After Effects||CS4 and CS5; Custom Workload, SD project with three picture-in-picture frames, source video at 720p|
|Adobe Photoshop||CS5; 20K x 20K test image|
|Digital Anarchy||Beauty Box Photo|
|Adobe Premiere Pro||CS4 and CS5; Custom Workload, 1280x720p, 59.94 FPS video, Panasonic DVCPro100, HVX-200 camcorder on P2 media, Render to Work Area.|
|Adobe Media Encoder||Custom Workload, Encode Premiere Pro project to H.264 for Blu-ray|
To isolate the impact of scaling core counts, we picked the Gigabyte board for its ability to enable anywhere from one to six active cores in the BIOS. We chose to examine results using two, four, and six cores. For each of these, we tested both with and without Hyper-Threading enabled, giving us essentially six different (logical) thread count scenarios. This would also hopefully help us see how much, if any, benefit there is to having Hyper-Threading in play with CS5.
Figuring that most people running Creative Suite would be unlikely to purchase a Core i3, we left Turbo Boost enabled across all tests. Note that when we measured CPU ranges, it was done by waiting ten seconds after launching the test and observing utilization for at least two minutes total across multiple areas within the test run. The lowest and highest observed utilization values were discarded unless seen to repeat at least twice.
Having 12 GB was important for us, because in upgrading from CS4 (or earlier) to CS5, we’re assuming that one of the main draws is to also make that leap from 32-bit to 64-bit and leverage the increased memory addressing that comes with it. When testing CS4, we still have 12 GB of system memory in the configuration, but the applications only access the first 4 GB of it. In one sense, this muddies the waters a bit since we’re effectively changing the hardware resources utilized along with the applications, but ultimately we decided this best reflected a real life decision process. If you’re going to pay for a CS5 upgrade, odds are high that you’re going to increase your memory along with it, and 12 GB is a sweet spot for triple-channel memory kits in today’s high-end systems.
We also debated for a while over the storage devices to use here. Ultimately, we decided to use Intel’s widely respected G2 SSD to house our OS and applications, while all data and scratch disk targets were placed on the OCZ RevoDrive.