Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

LOSTCIRCUITS	SHORTCUTS: New Performance Panoramas A Memory Address Space Crash Course System Configuration Disk Caching? System Memory Density Impact WoW, WinXP and WinXP-64 Conclusion Discuss this article here:
Panorama Factory and the $64-bit Question What's in a Benchmark...?
(Review by MS, September 2, 2004)

Conclusion

New Panoramas for 64-bit Computing

There is no doubt that the 64-bit version is not just a bit but substantially faster in stitching pictures together than the 32-bit version. Of course, based on a single benchmark, it is not possible to judge the overall performance gain to be expected from 64-bit porting, especially not if the application in question encompasses "uncharted" (for us) benchmarks that may slow down the 32-bit version. Overall, however, the results are in line with what we would have expected and the above statement is in reality nothing but one of our usual caveats and disclaimers not to take anything at all at face value without more supportive evidence.

One thing that stands out from the benchmark results is that the total amount of memory in the system is the limiting factor for overall performance, even in the reduced workload. This is not surprising at all, since every access of the page file will consume orders of magnitude more time than pulling the data from the system memory. Likewise, this is the main reason why running with lower amounts of memory somewhat dilutes the performance increase achieved with the migration to the 64-bit -ported version of Panorama Factory.

Along the same lines, running the benchmark with a fresh reboot for each run may add more control over the environment the application is running in, however, in reality, the main thing that is added is I/O time in system configurations using lower memory density. By extension, this dilutes the actual performance gain seen with the 64-bit application. For all these reasons, we skipped the 1GB system memory configuration for a 64-bit vs. 32-bit comparison, after all, we are interested in CPU performance and not I/O limitations. It appears, though, that with 2 GB of memory, at least the shorter benchmark runs almost exclusively from memory since we did not observe any disk caching.

What is really impressive, though, is the amount of acceleration we see in some of the tasks, as loong as they are not I/O limited. Running 3 times as fast in the 64-bit ported version is not a small accomplishment and the results shown here may hopefully set the pace for other software manufacturers to take the plunge and release equivalent portings of their applications. Needless to say that a 64-bit version of Excel will allow faster scrolling .....reminiscent of OfficeBench running on a Rambus system .... but that is not what we are talking about. Adobe, 3ds max. Lightwave and others should take a long good look at Panorama Factory and draw the consequences.

One bit of a sore spot in this benchmark suite was the mixing up of CPU limitations with the limitations posed by the memory management within the operating system. That is, without the artificial 50% reduction of usable memory space in WindowsXP, the 2GB memory limitation would not exist. And that was really what we encountered here, rather than a 4 GB limitation. In so far, the first part of the benchmark is not really valid, a move to Windows Enterprise Server2000 with a 32-bit processor would yield similar results as an upgrade to the 64-bit version. Needless to say that this particular benchmark was designed to just exceed the workable memory space of Win 2000 even with the 3GB switch but also keep in mind that removing a single picture from the stitching ensemble would have solved the problem for Win2000. Benchmarks are benchmarks and marketing is marketing and it is always dangerous to mix the two.

next page:      => More =>