I develope on a macBook now so I am using a GMA 950. I am doing a casual 2 1/2D game. To get speed up to 50fps I use ambient light only and not light sources. I may be able to oprtimize later and start using nicer lighting but that is a future consideration. I don’t know if you could run a FPS type game macBook.
I’m also developing on a intel macbook… And I’m glad I’m doing that. I’ve already made a lot of speed improvements just by limiting particle life, polygon counts and intelligent use of shaders.
I’m working on a freeride snowboard-game, so it’s kind of hard to make it fluent.
I think I’ll only change the usage of shaders if the host has a better graphical card.
We are trying to get WolfQuest to run decently on 950. OTEE fixed some render to texture problems for that card and even older Intel graphics in the 1.6.2 release.
On the Windows side, PCs with cruddy graphics chips still often have very fast (2 GHz or higher) CPUs, since Intel shipped a lot of fast P4s. So at least in our early testing, you get a decent AI framerate even though you have to cut back on the visual effects a lot.
We hope that we can do some checking in code to turn off some of the more dramatic eye candy like 3D grass, etc.
The radeon 9200s don’t have shared vram in any Macs I know of. In my experience they are as fast or faster than the GMA 950, especially in high triangle count scenes. See NCarter’s Phoenix Final for a perfect example of this. 5 FPS in a GMA 950 when looking at a lot of triangles, 20-30 (if I recall correctly) on a Radeon 9200 with a much slower CPU. They don’t support as many special effects as GMA 950, but usually you can’t use them anyways as they’re too darn slow.
We really had to make the Big Bang Brain Games run well on the GMA 950. If you are making a game for the Mac market, you need to do everything to make it run well on it. You’d be losing a large market. Too many people have them, and they’re still being sold. Freeverse spent a lot of extra time on their recent games making them okay for GMA 950, so I suspect it is basically a requirement to publish with them. I really would think it’d be similar with other publishers, unless there are exceptional circumstances.
In our case, it meant me running almost every model through a polygon reduction script inside of Blender. With the GMA 950, pretty much each triangle saved is that much higher FPS.
The most annoying thing about Intel cards are that they have a different performance profile than others: Its not that they are 4x slower than something else. It’s that vertrices are MUCH more expensive (could easily be 20x slower).
Back in the days you could safely push 60k polys per frame - we did in GooBall and it ran FINE on a Radeon9200. Haven’t tested on an Intel - and you can be damn sure I’m not gonna
Bottom line: Intel cards are VERY vertex sensitive, so you need to test on that hardware…