> Alexis, i thought we resolved this yesterday. but since you ask
> again. the folks who write the code i use for compositing tell me
> that if two images can fit into the texture ram, i can expect orders
> of magnitude better performance than if it must be paged from
> ordinary memory.
Depends. If he's used to an Indigo2, the `order of magnitude' is going to be much smaller with Octane [*obviously* if you can all have it in texture memory, it's even faster]. I've tried to explain it to you several times, but I'll quit and leave you to beat this one to death on your own from now on. Flog your favourite salesperson to death with it if you insist, but leave Silicon Investor readers out of it.
There's no easy way to add more TRAM memory on Impact gfx. If you select the same technology, you can have the same excellent fill rates, but if you want more, you'll have a bigger board (-> deskside). If you select a technology with better density, you can have more, but your textured fill rate drops like a stone (see O2).
Texture RAM is *not* DRAM, not VRAM, not anything like it (except on O2, where it's SDRAM). Any card using that for TRAM is going to be incapable of the fill rates achieved by an IMPACT gfx board.
>The O2 dies as far as texturing after a few spheres.
Watch GLquake run on an O2 and tell me that again ;). There's more in these scenes than `a few spheres'.
>this is what you get working on an Onyx
If you want large textures and have RE2-gfx targeted apps, the RealityStation (aka Onyx RE2 monoprocessor) is just the thing for you. Not even much more expensive than an Octane MXI, at that...or even an iStation, with InfiniteReality gfx at deskside Reality gfx prices. |
