I'm not talking about driver profiles at all. I'm talking about the way AMD want to hack up tessellation so that in processing it always hits the sweet spot on their hardware rather than them implement the hardware properly to handle the input data.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Amd Project Mjölnir
- Thread starter Gerard
- Start date
Soldato
And as i said it can be turned off.
And i don't agree with implementing things in hardware to handle input data at a level that 99% of people would never notice.
Soldato
Do you not think Nvidia isn't already doing the same? They implement some kind of optimization in all there profiles to get maximum performance.
Soldato
hhmm, that does not sound right, why would they not try to optimize it by as you say ''hack up tessellation'' as I know no one from AMD is going to come round my house and start soldering new chips to my graphics cards
Soldato
- Joined
- 11 May 2006
- Posts
- 5,786
Is this only HD6000 series?
Hope the 5000 series is forgotten so early.
Hope the 5000 series is forgotten so early.
Soldato
You're speculating on stuff you clearly have no idea about, never mind if IT IS actually happening or not, which again, you don't know.
Whats the point?
Is this only HD6000 series?
Hope the 5000 series is forgotten so early.
Why would it be forgotten? You'll get bug fixes and minor perf increases going forward but the drivers are mature now, so forget magic drivers that give to 10% increase across the board as they likely would have happened a long time ago...
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Yes theres somewhere inbetween, you both don't want AMD to cheap out on it, which personally I don't think they are doing. If Nvidia GENUINELY improves IQ a worthwhile amount, then its fine, if they introduce purposefully overly complex code just because their gpu can handle it and AMD can't, but it offers no worthwhile advantage, its just Nvidia being a-holes because they can be and that helps no one at all.
I mean if you add tesselation to a degree that say IQ improves dramatically for a 30% performance drop, it depends, if IQ is that much better and you're going from 100 to 70fps, thats great, IQ where we need it.
If you're going from 100fps to 50fps, and the IQ doesn't even visibly change on Nvidia hardware, but it goes from 90fps to 15fps on AMD hardware.......... well some might call that, for want of a better word, sabotage.
Unfortunately we've seen Nvidia do this before with Physx, with AA< with specific titles where they have a feature "added" to a game by donating engineers, and sackloads of cash to a dev, and you end up with a crap feature that runs like crap, offers little, but it runs slightly less like crap on Nvidia hardware.
Then you've got the simple situation of pushing software so it does make hardware slow, so future generations deal with it well, but thats more like what Crytek(used to) do, rather than adding a fairly poor feature killing performance.
I mean physx in Mafia 2 kills performance, and its woeful, shockingly unrealistic and stupid, a lot of it is faked effects for one thing, glass breaking, with two incredibly obvious "we've put these in so we can showcase physx" glass walls to be broken, that take up 20 seconds of a 5 hour game and the rest of that time a glass wall is no where to be seen.
I do find AMD's take on tesselation is, take AA, 2xaa offers the biggest benefit, 4xaa kills 80-90% of jaggies in the vast majority of games(some just have so many lines because of level design or style of game and some just don't work well with AA), 8xaa is for all intents and purposes perfect but takes a lot of performance, 16xaa offers almost no IQ improvement yet kills performance again, 372xaa is just worthless but officially it would actually improve IQ< but for the performance and the fact you can't notice those improvements, its stupid.
Tesselation IS the same, theres only so many points you can tesselate in a tiny area before you just can't see the difference and as with AA, performance is exponential.
Frankly both companies, AMD and Nvidia could do a lot more to PROVE which way is right, have dev's show off tesselation, have them visually show you the difference, make a demo that can go from 1 to 16 point detail in tesselation and see where you hit massively diminishing gains, both companies are pretty rubbish when it comes to actually proving their points.
Please point out where he has no idea what he is talking about, people are very quick to attack but never beck it up with facts. I am very interested in learning from people who are in the know.
nVidia does not at this point do any selective opptimisation for tessellation performance.
I'm all for opptimising for better performance where visual quality is unaffected but due to the way tessellation works you have to be really careful how you go about it and its best just left alone - you can't easily predict the outcome of a particular level of tessellation in an application i.e. what might be a totally overkill level in one might be required to get proper coverage in another. If you start selectively opptomising tessellation you end up with a complicated situation where developers can't predict what their original raw input data will look like once its gone through the drivers meaning you either potentially have issues like holes, seams, z-fighting, etc. in geometry or have to put a massive amount of extra effort into QA and can't account for future driver revisions where the system might change.
I'm all for opptimising for better performance where visual quality is unaffected but due to the way tessellation works you have to be really careful how you go about it and its best just left alone - you can't easily predict the outcome of a particular level of tessellation in an application i.e. what might be a totally overkill level in one might be required to get proper coverage in another. If you start selectively opptomising tessellation you end up with a complicated situation where developers can't predict what their original raw input data will look like once its gone through the drivers meaning you either potentially have issues like holes, seams, z-fighting, etc. in geometry or have to put a massive amount of extra effort into QA and can't account for future driver revisions where the system might change.
Theres atleast 3 missions with heavy use of PhysX tho thats still only about 30 minutes out of 5 hours of gameplay, the brewery, penthouse/hotel and the chinatown/triad restraunt/HQ which all have extensive breakable geometry and other physics effects... I'm not a fan tho of how its used situationally i.e. things that can be broken in those missions are static unmoveable, unbreakable objects in other missions like the prison and the WW2 section.
Soldato
And that already happens with drivers anyway on other areas with drivers optimization settings giving unintended results so what's new.
We will just have to wait & see & as i had said it can be turned off so is no big deal.
Last edited:
Having a few slightly more blurry textures than normal is one thing, having say objects appearing to be floating unsupported or the player able to see through parts of a wall they aren't supposed to is another matter again... not as simple as the user turning it off - not even sure at this point if there is an option anyway but even if there was some people are going to abuse it to take advantage i.e. being able to see things other players can't or don't expect - ok this is somewhat more towards the extreme end of the scale but a perfectly possible outcome.
Soldato
Already happens in games for years, its not exclusive to tessellation.
You can take anything to the extreme you will find possible negatives.
Radeon Pro & such programs for NV cards can already make complete mess of games if you set them so.
I would rather have the possibility of user controlled Driver tweaks that give no noticeable or minimum visual impact to gain acceptable performance than do it through the game that normally will gave a much bigger negative visual impact.
No Pain No Risk No Gain.
Last edited:
Soldato
