So, some of my impressions of how FG runs on the GeForce 670M (Windows). * in default terrain (I did Nevada, going with the F-16 from KNID to KLSV), all procedural effects in atmospheric light scattering on, some snow on the Sierra Nevada, 120 km max. visibility range, some Cirrus cloud development I got most of the time the framerate indicator locked at 60 fps with dead-constant frame spacing even at 30.000 ft with full visibility range. Occasionally it drops to 30 fps and stays locked there for 10 seconds or so, then jumps up. I don't know if the card needs a cooling break now and then or if it's something else, but it's not affecting the experience at all.
* going to Canaima (SVCN) with dense jungle (vegetation cover to 2) and noon convective cloud development with the A-10, low level flying over treetops and up to 15.000 ft to get the view - same story - the trees just go by at 60 fps, occasionally clocking down to 30. I would not have thought this possible at all. * really extreme testing, France custom scenery with the F14 from Lyon into the Alps, normal convective clouds drawn to 75 km, 120 km visibility range with trees to 2. Framerate stays most of the time locked at 30 fps, occasionally drops to values between 20 and 30, especially when I pass by a cloud. According to my estimates, the memory usage must have been into the 10 GB range with so many trees and terrain generated over such a huge area in custom scenery, and after 15 minutes or so I did notice the occasional freak frame of 0.3 seconds length or so - that was probably when I started digging into swap space. Custom scenery with high visibility and dense tree cover looks quite stunning - I've never been able to run that before. As Alexis said, the elephant in the room are the clouds - overcast skies render almost as slow as on my old box, which is quite surprise to me. But that implies that if I ever get the GPU to run properly under Linux, I need to focus on making cloud shaders faster, there's no problem with the terrain shaders and procedural textures which a modern GPU can't handle. I did not sufficiently appreciate this point before. Both a LOD range system to ease the vertex load and the tracers to ease the fragment shader load will help here. I did abandon tracers since my main motivation was to ease the load from terrain rendering if obscured by clouds, but if terrain isn't the real problem, then they are useful because they do work for clouds. So not all hope is lost, I just need to experiment and see what operation exactly causes the bottleneck. I've also flagged a few issues: * I've spent 5 minutes puzzling over why the terrain isn't displayed out to the horizon till I remembered that I have to adjust LOD bare. What do the LOD ranges do these days? Should we simply couple LOD bare with visibility settings, assuming that a user who sets visibility to 100 km also wants to see terrain to 100 km? I've had at least one support request per email who had the same problem a while ago. * strange bug - all generated clouds (on Linux and Wndows) used only one texture of a multi-texture sheet, which gives some odd repetitions. This may be something trivial (my binary and FGData snapshots are a few days apart) so if there's been work on clouds in the mean time it could explain it. Or something with the gl_MultiTextureCoord isn't set up properly in an architecture dependent way. Or I messed up the Nasal and didn't realize it on the old computer. Does anyone else see this? * urban shader effect slider in Lightfield is stuck at 3 - no idea why. But the detailed shader control menu has been behaving odd in a number of ways and thrown Nasal errors - I think Gijs created it (?) - if you could take a look? * Restarting Advanced Weather seems to have a timing issue for really high framerates. The problem is that Nasal storage arrays are removed with a time delay as to allow loops in the current frame to terminate properly, but then you can't restart the system before the arrays are removed, otherwise it creates arrays in the restart which are removed on cleanup later, so there's a time delay in restarting the system which should allow the system to clean before restart. Works fine with 20 fps, apparently doesn't with 60, no idea why. Anyone else seeing this? Restarting the weather in pause mode does the trick, but I'll have to adjust the time constant I guess. * Thorsten ------------------------------------------------------------------------------ Monitor your physical, virtual and cloud infrastructure from a single web console. Get in-depth insight into apps, servers, databases, vmware, SAP, cloud infrastructure, etc. Download 30-day Free Trial. Pricing starts from $795 for 25 servers or applications! http://p.sf.net/sfu/zoho_dev2dev_nov _______________________________________________ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
