-
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.
Possible huge ATI tessellation performance boost with 10.5 drivers
- Thread starter RavenXXX2
- Start date
Permabanned
- Joined
- 22 Feb 2010
- Posts
- 1,629
Sounds very promising
if they do infact use the shaders i will have Great performance in ungen
if they do infact use the shaders i will have Great performance in ungen
Well it's true about Nvidia tessellation performance dropping in games compared to synthetic benchmarks they have optimized the cores for, check Dirt2 out when in the menu area where the tessellated crowd is, the 480 is pushing the same frames as the 5870.
Permabanned
- Joined
- 22 Feb 2010
- Posts
- 1,629
looks nice in metro2033
To be honest all ATI is going to acheive is better Heaven benchmark results. As they have stated themselves, once the gtx 480 is using its shaders for the graphics in the game heavily, it's tessellation performance drops. The same thing will happen with the 5 series.
Still nice to have and perhaps will still make a difference to some games and their performance with tessellation.
Still nice to have and perhaps will still make a difference to some games and their performance with tessellation.
You mad? Tessellation in metro2033 is even less noticable than in dirt 2! You need to take screenshots and look at them to see the difference and hence not worth the hit to framerates IMO.
Image an HT like situation for the 5xx0 series of cards.
Most games wont fill all 5 units of each shader, imagine if you could use the remaining ones for tessalation.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Taping out a core, even if it doesn't need a respin(its INCREDIBLY unlikely for ANY core to tape out and go into full scale production on one spin, its also unlikely and rare for 3 full respins to be needed). Basically tape out first date means 2-3 months minimum till retail availability, another respin add's 4-6 weeks.
As for benchmark performance, rather pointless waste of time, if they can use excess shader performance to do tesselation that would otherwise not be used, then fine, but the way the architecture works thats very unlikely, possible, but unlikely.
They really need to be pushing VERY HARD to get game developers to code better for the architecture. Its somewhat PS3/360 problems at the moment in that programming to get the full 1600shaders working most of the time is incredibly difficult and doesn't happen often. Getting 3 of the cluster of 5 shaders working is fairly easy, but 4, or 5 is very rare. At any one time its rarely more than 60-70% loaded which leaves them huge headroom. Though really thats the side effect of making a core with a huge number of shaders without the massive size.
Its very likely that when Nvidia make a similar core(going as small as possible and thereby having to go for efficient and complex shaders rather than simple and large ones) that all of a sudden all game dev's will be breaking their necks to optimise code for the AMD style architecture.
Lets get two thing straight... nVidia's tessellation processing is NOT aided by the stream processors (as in traditional compute - but is implemented as dedicated inline geometry unit)... and...
Using the stream processors to increase tessellation performance is going to be VERY hit and miss, your doing additional work that the nVidia polymorph engine doesn't do, deeper into the pipeline, as the ATI cards have masses of shaders its likely that in specific instances it can provide a decent speed up, but in others its going to have overhead penalties that may even cancel out the gains, so its going to make it more complex to program for. Outside of predictable targets like static benchmarks I think the writebacks/local memory issues will likely murder performance.
Using the stream processors to increase tessellation performance is going to be VERY hit and miss, your doing additional work that the nVidia polymorph engine doesn't do, deeper into the pipeline, as the ATI cards have masses of shaders its likely that in specific instances it can provide a decent speed up, but in others its going to have overhead penalties that may even cancel out the gains, so its going to make it more complex to program for. Outside of predictable targets like static benchmarks I think the writebacks/local memory issues will likely murder performance.
Last edited: