Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
CPU Intel P4 840 D P4 820 D P4 630 P4 640 P4 650 P4 660 P4 670 AMD Athlon64 3500+ 3700+ 3800+ 4000+ X2-3800+ X2-4200+ X2-4400+ X2-4600+ X2-4800+ 1-Way Opteron Opteron 144 Opteron 146 Opteron 148 Opteron 150 Opteron 152 2-Way Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Opteron 248 Opteron 250 Opteron 252 2-Way Dual Core Opteron Opteron 270 Opteron 275 nVidia GF 7800GT GF 6800GT GF 6600GT ATI R X850 XT PE R X850 XT R X800 XT PE R X800 XT R X800 XL Memory Corsair Crucial Kingston Mushkin OCZ What are you shopping for?

Adobe PhotoShop 7.0

Adobe Photoshop has a built-in timing tool that can be activated by clicking on the arrow in the footer / margin, and which is used for a number of benchmarks including the various iterations of PSBench. There is only one problem: we manually timed the execution of commands and the numbers [in seconds] generated by the timing routine had very little in common with the real time it took to generate the results. Often times, the numbers were too short, which can be explained, at least in theory by the fact that there were disk access times that were not taken into account by the counter. Plausible enough.

However, this is not the end of the story since, depending on the workload, Photoshop generated numbers that were way in excess of the real time interval it took to run one or the other filter. For example, running a smart blur on the picture shown on the first page in its original format of 7936 x 5952 pixels took on average 15 seconds as stopped with a stop watch. The rendering time displayed by Photoshop was correct about 50% of all runs (14.5 sec), the average of several runs, however, was some 23 seconds with results of 38.5 sec or even to up to 61 seconds. It is important to note that there was NO difference in the manual timing of the execution of the same filter, that is, the real time was always around 14.5-15 sec, irrespective of what the Photoshop timer showed.

Suffice it to say that these findings completely invalidate results from PSBench (which we found rather unreproducible anyway) unless the timings were obtained manually rather than relying on the Photoshop output. In any case, we were running some of these benchmarks with 2 x 256 MB or 2 x 512 MB or else with 4 x 512 MB and there was no difference between the manually stopped times for filtering. This is in stark contrast to the numbers recorded and output by Photoshop.

Conclusion

The reason for going through all these benchmarks and writing this article is not to point a finger at anybody. Rather, it is to dam spreading of wrong information caused by some software glitches or else, being typical only for one particular hardware configuration but which, by no means, can be generalized. In not a single instance did we incur any performance hit greater than 6%, and those can be accounted for by some stabilizing measures, that is, more relaxed timings on the chipset level.

As mentioned earlier, we don't have any reason to question the accuracy in reporting of the results that led to the conclusion of a dramatic penalty for high system memory density configurations. However, if the generation of the results is flawed a priori, the most careful reporting will not be able to change the inherent faults of the results.

The one fact remains, more memory is better but only if the operating system and the application are capable of using it. Adobe Photoshop is not really operating within the Windows environment. It uses is own scratch file, which is not part of the Windows virtual memory but actually competes with the latter for HDD space. Moreover, the entire history of any filter is stored resident within the system memory unless purged after every transaction, which also invalidates the history of the image. This is yet another reason why e.g. PSBench generates somewhat unreproducible results. It only takes a few filters on a large file to fill up the entire system memory, after which the scratch pad on the HDD has to be accessed and again, keep in mind that that one and the Windows swapfile are mutually exclusive.

In the end, the take home message is that on a correctly configured high quality i875 / i865-based mainboard, there is no performance hit per se associated with increase in memory density or filling up all memory ranks. The situation may be different with OEM boards depending on whoever got their hands on the BIOS to FUBAR it for the sake of OEM stability and maybe the memory itself also plays a role in this scenario for the reasons outlined on the second page of this review but memory density and performance do not mutually exclude each other.

next page:    => More =>