If a person might happen to have a system with more than one graphics device in, why type of tests would we like to see?
[Edit for trinibwoy] Apart from the blatantly obvious
[Edit for trinibwoy] Apart from the blatantly obvious
Test Suggestions
- Thread starter Dave Baumann
- Start date
HL2 @ 1600x1200 max everything w/8xS AA & 16xAF......
Baseline fill rate tests, then exploration of differing performance characteristics in pixel and multitexturing fill rate, as well as varying pixel shading loads, possibly including usage of super sampling with each. Z and stencil characteristics would be interesting as well, probably.
Perhaps, along with this, investigation of render target usage and multipassing, trying for an overall impression of how the implementation handles these complexities of rendering workload, to see how much flexibility/robustness has been accomplished in the implementation, and how efficiency varies with them.
...
If there are happen to be options to control certain mechanics of how the devices work together, contrasting what is exposed about each by the above would also seem warranted.
Also, an investigation of any changes in geometry throughput due to overhead of the implementations would seem to be interesting as a datapoint, and perhaps a reference for some evaluations, especially ones involving multipassing.
...
I'd also be curious about how traffic over the system bus plays a role in implementation, but I don't know of a way to measure it or restrict bus speeds to then measure the impact of gradated amounts of restriction.
Perhaps, along with this, investigation of render target usage and multipassing, trying for an overall impression of how the implementation handles these complexities of rendering workload, to see how much flexibility/robustness has been accomplished in the implementation, and how efficiency varies with them.
...
If there are happen to be options to control certain mechanics of how the devices work together, contrasting what is exposed about each by the above would also seem warranted.
Also, an investigation of any changes in geometry throughput due to overhead of the implementations would seem to be interesting as a datapoint, and perhaps a reference for some evaluations, especially ones involving multipassing.
...
I'd also be curious about how traffic over the system bus plays a role in implementation, but I don't know of a way to measure it or restrict bus speeds to then measure the impact of gradated amounts of restriction.
Dynamic textures/vertex data used on both devices. Also maybe computing something on one device and using it on the other (render-to-texture pulled on the other device).
Unless the system is real smart these should hurt like crazy.
Way back when, I did 4 player/4 screen stuff. We had to turn animating textures off because of the total saturation of the buses (AGPx2 and PCI back then). That was 1 AGP and 2 PCI devices (G400 + 2 Voodoo3 IIRC), so thing should have changed.
Unless the system is real smart these should hurt like crazy.
Way back when, I did 4 player/4 screen stuff. We had to turn animating textures off because of the total saturation of the buses (AGPx2 and PCI back then). That was 1 AGP and 2 PCI devices (G400 + 2 Voodoo3 IIRC), so thing should have changed.
DeanoC said:
Provide me with some test material very fast and I'll be all to happy to try it out!
Does the SLI rig gotta go back Monday or do they have someone there hovering over ya still hand-cuffed to it? :|DaveBaumann said:Provide me with some test material very fast and I'll be all to happy to try it out!
thegrommit
Newcomer
digitalwanderer said:
Why limit oneself to that resolution when HL2 supports widescreen gaming at 1980x1020?
Ooooh! How about the classic "benchtest"?
Y'know, where you hold the whole rig bench-high off the floor (36" or so, that's a little under a meter for you metric folks) and drop it to see if it survives ok? :|
Y'know, where you hold the whole rig bench-high off the floor (36" or so, that's a little under a meter for you metric folks) and drop it to see if it survives ok? :|
Kombatant said:
Wouldn't it be approximately twice the power consumption for a single card? Noise is another thing I don't see as particulary pertinent since it will just be 'louder'. It becomes even less relevant if you can slide two silencers in there. Let's see some performance numbers!!!
Well, noise won't necessarily scale according to the noise of a single board - you now have more components generating more heat so the fans are likely to need to spin faster.
I'm beginning to get the impression the entire thing is a bit of a farce right now...
That would be another interesting little thing to report about it.DaveBaumann said:I'm beginning to get the impression the entire thing is a bit of a farce right now...
You mean a farce as in "it ain't twice the power of one" or as in "this bloody thing is bug-ridden and not ready for primetime!"? Or something else entirely? :|
EDITED BITS: Oh, and could you run 3dm2k1se just for us old skool crowd please?
More along the lines of "you won't see any performance increase unless its in our list of precoded app detections, and if it is we won't give you any control over the rendering mode it uses".
