Hi, On Fri, 2010-11-05 at 09:46 +0000, Peter Clifton wrote: > 1. My VBOs and arrays are pretty large buffers, to fit lots of geometry. ... > So.. we have a large buffer. That is a large chunk of graphics memory to > be requesting all the time. I think that contributes if the card doesn't > have a decent chunk free yet. I did a little bit of investigation and I suspect this is the culprit to everyone's poor performance. I ran the power-hw board on my machine and achieved around 85FPS. Felipe ran the same board on his machine and achieved around 18FPS. My graphics card has 1GB of VRAM, Felipe's has 256MB (I couldn't easily compare this to other people's results as not many people said what card/driver they were using). I have tried quadrupling the size of the VBO and now get around 17FPS, which is considerably less than a quarter of my previous result and almost exactly the same as what Felipe got. I would say that this indicates that the large size of the buffer is to blame for the speed problems if everyone else has cards with ~256MB of VRAM and therefore the driver is having to swap out part of the buffer into system memory. NB: I ran the above test with the larger buffer on code before commit 0fa243b4bf which seems to change how the buffer allocation is performed. I have tried running it with the quadrupled buffer on the head of that branch and see no drop in performance any more. I still think my analysis stands though if people don't have enough VRAM to hold the size of the buffer used by the power-hw board. Richard
Attachment:
signature.asc
Description: This is a digitally signed message part
_______________________________________________ geda-user mailing list geda-user@xxxxxxxxxxxxxx http://www.seul.org/cgi-bin/mailman/listinfo/geda-user