AMD Dual-Core Optimization Utility Available

http://www.dailytech.com/article.aspx?newsid=3152

I downloaded and installed it on my x2 system, does not seem to do any harm, but have not had a chance to do definitive tests regarding its effect on SL.

This what the article says:

AMD has silently released a utility that fixes time stamp counter (TSC) issues with dual-core AMD processors such as the Athlon 64 X2 and Opteron. The AMD Dual-Core Optimizer utility is a driver that synchronizes the time stamp counter on both processor cores and improves game compatibility and performance. Time stamp counter issues occur in older games that are incompatible with multithreading and would run too fast or out of sync.

Issues also occur in newer games and applications that have compatibility issues with processor affinity like Need For Speed: Most Wanted. Affinity issues typically required users to bring up the task manager and set the processor affinity manually or performance would suffer dramatically.

The AMD Dual-Core Optimizer utility could possibly be the rumored “Reverse Hyper-Threading” patch that would improve single thread application and games performance by having dual-core processors show up as a single core processor in applications. AMD has declined to comment on "Reverse Hyper-Threading."

I gave it a testdrive and want to report the results: My testsystem is an opteron based system with two cores, a gefore 7600 GT and enough memory. To come to the point: The use of multiple cores made things worse with my system.
I report the stats here:
I use a resolution of 1280x1024, draw distance of 160m, LOD sliders in the upper third, everything on except anisotropic-filtering.

Waterhead (61,104,25)
With two cores (and the tool of course) : 6 fps, drawing time at around 170ms with most of it geometry (130ms)
Core one was fully used and core two at about 50percent.

Only core two: 9 fps, drawing time at around 112ms with geometry at about 70 ms

At home at a location more primmy and less avatari ... the results have been similar. The fps have been less with two cores.

From a computing point of view... though the client uses several threads... the overhead introduced by thread synchronisation bitween *two cores* seems to eat it all up and make things even worse. Perhaps performance could be better if all the threads that depend most on each other would run on one core and all others with less dependency on the other... but that's just a guess at this point.

And also not to forget the 7600 GT likely already runs at its limits.

Perhaps someone else comes to different results.
A sidenote is that after installing this tool something freezed the machine during start of all the resident tools I use and I had to reboot. After removing a few out of the autostart list it worked fine. But thats just my specific setup here.

Leffard,

Same results but mine were more drastic, 13FPS on both cores, moved it to CPU 0 and FPS jumped to 60!, I also tried switching over to 1 core then back to two and I noticed it was acting like I never changed anything, very odd.

I played a bit further with it... and the things got even stranger.
I use the same opteron two core system as described before.

If I *startup* with two cores... the results are worse than with one core. Switching over to one core only restores the normal performance after a few moments and adding the second core back again gives me a gain in fps of 2fps resulting in 14fps in total.

The way was then: 8fps (two cores) -> 12 fps (one core) -> 14 fps (two cores again). Perhaps the AMD opteron driver has some intelligence about distributing the threads as long as they have been already running. The system then burns both cores at 100%.

At Waterhead the perfomance was also about 2fps better with this procedure.
As the improvement seems only minimal (as the 7600GT likely limits everything) my second core is better invested in something else.

It would certainly be interesting to find (perhaps with the Linden's assistance) a public site with fixed prims in a given direction, where anyone could test definitively . . .

Then it would be possible to have people test & list their systems/components/drivers/settings. I would think it would be in the Linden's interest to help facilitate a mechanism to determine what systems and configurations work better.

A Database (similar to SiSandra or PcWizard) could be hosted on the SL website. I believe this could be a worthwhile project . . .

Thx for the news about this, Delta. I have an Opteron 170. I'll give it a go after work this evening.