These are not proper benchmarks, but this morning i ran some comparisons between D201GLY2 and D945GCLF2. They are interesting, and i think reinforce the german review posted earlier.
Test descriptions:
test 0
Play dvd from optical drive and observe cpu use. Software was PowerDVD 4.
test 1
Reencode first three chapters of /the worlds fastest indian/, and measure time taken. 338M file read from optical drive, transcoded and saved to hdd. Bit rate was to be reduced by 30%, using DVDShrink.
test 2
copy 156MB 2 layer psd from usb 2.0 flash stick, open the file in Photoshop CS, resample image to 6/7 smaller, resave file and close CS. Measure time taken.
test 3
The final test comprised test 1, while simultaneously carrying out test 2, twice, and also making and saving a screenshot using CS. CS was closed and reopened between each task.
Test rig A
D201GLY2, 1GB ram, new SATA 3.5" HDD, IDE 5" optical, recent XPSP2 install
CD scratch disk on D drive.
Stable OS install in use.
Page file 512MB
Test rig B
D945GCLF2, 1GB ram, new IDE 2.5" HDD, IDE slim optical, recent XPSP2 install
Both drives on same IDE cable (!?)
CD scratch disk on D drive.
Brand new OS install with a few quirks still.
Page file 1500MB
Now note that these arent strictly identical setups. I dont have the gear to do this any better. Of particular note is the slower drives in the atom rig.
Results
Test 0
The d201 board used about 22% cpu usage. (little more than my earlier observation owing to the use on that occasion of a lower bitrate file located on harddrive)
The atom board used average of 8%, but this is actually around 25% on one core-thread and not much else. Take what you want from that--in hindsight you probably cant call that 8% cpu usage, as the application appears single core optimised.
Test 1
The D201 rig completed the transcode in 2 minutes and 50 seconds. Using about 77% CPU.
The atom board completed in 1 minute 52 seconds, a good improvement , 35% faster in fact.
Test 2
D201 completed the photoshop resample exercise in 1 minute 0 seconds.
D945 completed in 1 minute 5 seconds. It appeared to have trouble using the CPU fully, showing low CPU use figures, averaging say 10% over 4 "cpus".
Test 3
D201 completed the combined test 3 minutes and 30 seconds. As you can see CPU is pretty much maxing out. Both transcode and photoshop tasks proceeded smoothly, if slightly slower than done seperately, but it felt as they were sharing CPU evenly.
D945 completed in quite a different manner. The transcode finished early, leaving the Photoshop tasks to kind of muddle along afterwards. The Photoshop tasks appeared to quite compromised by the transcode task, even though reported CPU usage was not at all high. Photoshop just seemed to be struggling. It took an age just open photoshop, and the transcode has completed before the first of the 2 resamples had even completed.
However total time was 3 minutes and 47 seconds, not a great deal more than the D201 took. To my mind there seemed to be some problem utelising the cpus power fully, perhaps averaging 45% over 4 "cpus".
Comment
While newer more multicore optimised software may have made a better job of these tests, Im wondering how well baked the 945GC chipset and the dual core atom combo actually is. My understanding is that for the better part of the 945's life it was single core only. Dual core having been hacked on at a late stage. Then you have in order execution as opposed to the out of order used in netburst and core architecture, and im left with questions, and the sense that the atom needs an optimised chipset to match both its ability's and power profile. Id be interested in anyones ideas about why photoshop struggled as it did in these tests. Will try a sata 2.5" drive when it arrives, and a bigger PSU in case its a power issue.
Linear Mode
