> > One possibility is if the Lindens set up an island, with an agent limit > of *1*. Region forces AV to face a particular direction. Avatar drops > in, and waits 5 mins. During that time, the system records frame rates, > CPU, and vid card. AV reports graphics settings (unless that can be > pulled from the client automatically). AVs do it with client set to the > absolute defaults for low, mid, high, and ultra, turning off anistropic > filtering and antialiasing. A few weeks of that, and we should have > some good data - would have to be the stock client, too, though - but > could record stock 1.23 and stock 2.1, and get good data on whether the > changes to the rendering engine helped. > > I've found that graphics cards and machines will reverse performance ranking sometimes based on the scene they're presented with. The load of having other avatars around is substantial and sometimes different from card to card (and on lower end machines occasionally it's seemed that turning off avatar nametags can make a noticeable difference). Some "less capable" cards will hold up better in some circumstances, in terms of FPS degradation percent, than "better" ones, and it's hard to guess based on looking at the scene. A single scenario will tell something, but it's hard to extrapolate from it and generalize over the range of scenes found on the grid.
Joel
_______________________________________________ Policies and (un)subscribe information available here: http://wiki.secondlife.com/wiki/OpenSource-Dev Please read the policies before posting to keep unmoderated posting privileges
