I've just upgraded from my old Q6600 PC to a new i7-2600K, and I wondered how much better the real world performance was. I'm an engineer and my current project has me doing a lot of performance analysis on very large scale systems, so I have a bit of a background in this - though this is very lightweight and superficial compared with my real work.
I did a series of tests, none that scientific but geared toward how I use the PC. Generally I process 1000-2000 wedding images at a time, and I may be doing exports in ACR at the same time as doing something else in Photoshop/Image Processor, using the web, and ftp'ing data around.
Here are the results and my analysis.
PC Setup
Both PCs ran Win7-64 and Photoshop CS4, the Q6600 had 4GB RAM, the i7 had 16GB RAM. The hard drive setup (which moved between machines for the tests):
- OS is 60GB OWC SSD (3Gbps), and was reinstalled on the new machine
- Bridge/ACR cache is 120GB OWC SSD (3Gbps)
- Various hard drives - 2x1GB Seagate Barracuda, 1x2TB Seagate Barracuda (main images drive), plus others not relevant
Notes
- I generally restarted the PC between runs so the memory cache wasn't making things faster, unless I used a different set of images.
- Three sets of images were used, as some tests ran things simultaneously.
- The RAW images used are mostly Nikon D700 images, but with a smattering of 5DII and 7D RAW files thrown in. The same files were used on each PC for each set of tests.
- I probably should've recorded the disk throughput, but I didn't. I had the performance monitor open at times, but didn't record anything.
Raw Data
See a spreadsheet of my timings here
.
Disk performance was interesting. When I started ACR and Image Processor at the same time, using two different spinning disks to hold the data, it was initially reading data at around 25MB/sec in total. Each disk is capable of 50-100MB/sec each, so they weren't working that hard. It was writing a total of 8-12MB/sec to the two destination disks.
During the export the disk read bandwidth used dropped off to almost zero, around 100KB/sec, but the destination disks were still going at the same speed. This tells me that the data was all read into system ram as a cache (I have 16GB RAM) before it was even needed. CPU usage was 40% for ACR/Bridge, 30% for Photoshop during this test.
When I repeated the export, the write speed stayed the same, but no data was read. This confirms that the source data files were cached.
When I tried again with a different set of images, with just Bridge going, data was read at 15MB/sec, and CPU usage was at 53%. This read performance again dropped off to near zero part way through the export.
Observations
The Q6600 seemed to be going at 100% most of the time, with the disk not really working that hard. The i7 occasionally reached 100% CPU, but more often sat at 30-80% CPU, and the disks were working harder. The disks weren't working at full speed though.
Conclusion
The i7-2600K takes between 1/3 and 1/2 the time the Q6600 takes to do real world batch image processing tasks. That's a really worthwhile speedup for people who have a large number of images to process.
Of course interactive speed is probably even more important than batch speed, but it's difficult to quantify. It definitely feels faster and more snappy, with less waiting around.
I'm not sure where the bottleneck is. The disks can read data a lot faster, and the CPU wasn't fully used, so my theory is the software is the bottleneck. Multithreaded programming for multicore machines is relatively difficult, so I expect things to get faster as technology improves. CS5 may be better than CS4 in this regard. I hear Lightroom can do simultaneous exports, which speeds things up, but I don't like the LR interface much.
The tests do show that Windows 7-64 can take advantage of lots of RAM to effectively cache files, both files already requested by a program, and reading ahead to files not yet requested. Once i've culled my photos the working images i'm left with are generally 6-12GB, so they should sit nicely cached in RAM 
