Well, I forced 'low precision' right now, and the graphics are all messed up, shaders missing and messed sort of. Same way it was in Ruby 1 if forced in 'low precision'. As for 'partial precision' isn't that part DX8 part DX9? I don't think that would be possible here, anyways it's not an option in 3dAnalize.
New Ruby and Crytek Demo for Download
- Thread starter tEd
- Start date
At this point in time, I'd say not.
Read back over some of the philosophical debates about multi-precision architectures here (and related comments on the increased transistor cost of FP32 ALU's and why they wanted 90nm for SM3.0) and you'll see that there is quite an opposition to them. Fundamentally, if you build the architecture with FP32 in mind all the time (just as ATI did with FP24 all the time) there is not necessarily any reason to have multi-precisions.
And on that note, VS's are currently single precision - they are always FP32. Supporting multiple precision can actually require more transitors, so given that ATI has publically stated they are going unifed shading hardware its another question whether it would be a cost or a benefit to support multi-precisions on a unifed platform.
Read back over some of the philosophical debates about multi-precision architectures here (and related comments on the increased transistor cost of FP32 ALU's and why they wanted 90nm for SM3.0) and you'll see that there is quite an opposition to them. Fundamentally, if you build the architecture with FP32 in mind all the time (just as ATI did with FP24 all the time) there is not necessarily any reason to have multi-precisions.
And on that note, VS's are currently single precision - they are always FP32. Supporting multiple precision can actually require more transitors, so given that ATI has publically stated they are going unifed shading hardware its another question whether it would be a cost or a benefit to support multi-precisions on a unifed platform.
DaveBaumann said:
I think you might be overestimating the ability to do that though. Even assuming the architecture is built for FP32, ( I mean even the Geforce 6 series has register issues with FP32, Paticularly on long shaders. I dont see how ATI can completely alleviate them. If they are so against FP32. That is of course their perrogative. But I am willing to bet they will go for at least some sort of partial precision next year. Whether they will use FP24 or Fp16 for their PP is entirely speculative though.
Even with an amazing architecture. FP32 requires alot more work than FP24. I believe ATI would be leaving themselves open if they dont support a partial precision of some time.
You can never completely alleviate these things, and even with FP24 and R300/R420 they can never cater for maximum usage all the time; what you can do is build in predictible performance degredations, as they have with R300/R420 and there is no reason why that can't translate to FP32 in a similar manner.
Sireric has said that purely at a silicon level support for FP32 over FP24 would be an increase of 25% transitors (for that areas that deal with it); whilst that 25% pertains to the ALU widths, its a similar thing just to apply the to the registers. There no reasons why FP32 can't behave as FP24 does now, it just costs transitors and I suspect that this is wholly one of the reasons they said "No, 130nm / 110nm won't cut it for our shader 3.0, we'll wait for 90nm".
(On another note, if they were going down this route then they would have known it two years ago and they would have been advocating it to developers as a route to take for "future hardware"; not once have I seen it in any developer documentation nor has a developer said to me that they have been advocating it)
Sireric has said that purely at a silicon level support for FP32 over FP24 would be an increase of 25% transitors (for that areas that deal with it); whilst that 25% pertains to the ALU widths, its a similar thing just to apply the to the registers. There no reasons why FP32 can't behave as FP24 does now, it just costs transitors and I suspect that this is wholly one of the reasons they said "No, 130nm / 110nm won't cut it for our shader 3.0, we'll wait for 90nm".
(On another note, if they were going down this route then they would have known it two years ago and they would have been advocating it to developers as a route to take for "future hardware"; not once have I seen it in any developer documentation nor has a developer said to me that they have been advocating it)
Here are some screenshots of Crytek demo running at 1920x1200 widescreen with 2xAA + 3Dc enabled. (56kers, don't bother).
beginner16
Newcomer
What was the point of crytek demo?To tell us how powerful their new game engine is,or to tell us that their next game will have same graphic as this demo?
beginner16 said:
showing off the engine what it is capable of and get money from it
They doing alot of such things. Sometimes they team up with ati doing this and then team up with nvidia and doing that. It's a win win situation. They get good PR for the game(farcry) and engine (which they wanna license) and on top of it they can please both major IHV and fanboys of theirs
Bouncing Zabaglione Bros.
Legend
kyleb said:
I got the same thing on the 4.11 drivers, but I wasn't prepared to grumble about it as I had a load of other memory hungry apps running in the background which might have caused the problem. If it is something that is fixed in the 4.12 drivers, those should be out in the next week or so.
Unknown Soldier
Veteran
Or just d/l the Beta 4.12's
US
US
Similar threads
