9700 Pro Fillrate / Memory bandwidth balance

[Joe DeFuria]

There was some discussion here fairly recently about the suppossed "balancedness" of the new high-end architectures. It looks like X-Bit labs has gathered some good data in this respect with lots of overclocking benchmarks. (Including overclocking core and memory separately from one another.)

http://www.xbitlabs.com/video/radeon9700pro-overclocking/

Haven't read it in detail yet, but in addition to getting the core / memory up to 450/800 (!!) the results show the 9700 Pro looks to be pretty "balanced"...at least in those benchmarks tested.

 

[Doomtrooper]

I think this gives us a idea what a R350 on .15 might be able to achieve.

 

[Joe DeFuria]

Agreed.

Also interestingly, Hynix (an ATI memory partner) has announed production of 500 Mhz DDR SDRAM. (NOT DDR-II). One has to wonder if that might make its way into R350...

http://www.digit-life.com/news.html#943

 

[Bambers]

450/800

Even then was that chip even hot with the water cooling?

Interesting to note the aniso gains with memory increase and with gf fx 'only' having 16gb/s would it take more of a hit?

EDIT:
< can't read graphs, ignore this post

 

[Dave Baumann]

Given that its texture sample limited with Aniso I wonder if thats what they are going to do with R350 - increase the texture sampling per cycle. Probably not where your want to be ultimately as shaders are more important, but might see you good for current titles.

 

[arjan de lumens]

Hmmm - I seem to remember from when geforcefx was announced that some ATI engineers said that the R300 was more limited by memory latency than bandwidth. If that is the case, you could get good scaling of performance with memory clock speed even if the memory bandwidth is totally overkill in the first place ...

 

[antlers]

450/800

Even then was that chip even hot with the water cooling?

Interesting to note the aniso gains with memory increase and with gf fx 'only' having 16gb/s would it take more of a hit?

With aniso they showed performance is more dependent on GPU overclocking and somewhat less on memory overclocking--therefore, if the GeForce FX has a similar aniso implementation, it will be less constrained by memory bus (relatively) under aniso than without aniso.

It will certainly be a change if an nVidia chip benchmarks most favorably against the 9700 with anisotropic filtering applied!

 

[Hellbinder]

It took me a while.. But I finally figured out that my post got shang-hi'd by a mod and moved to a different forum...

Yes the post talked about the released CAT 2.5's.. However it was NOT intended as a discussion about drivers. It was is direct relation to the Performance increase seen in X-Bits test, And The relationship to the Drivers used.

In other words.. X-bit saw an approx 30% increase based on OCing. The Cat 2.5's have a verifiable 7-20 FPS (as illustrated by my examples) depending on the game. Thus combined with the info that the next driver set is supposed to increase performance even more...

It simply makes one wonder what kind of overall improvement an Extreme OCing test like this would result.. Perhaps a 40-50% increase in some cases??? Aslo, It would be nice to see whether they Addressed the Texture sampling/AF issue with Drivers to some degree.

 

[Bambers]

With aniso they showed performance is more dependent on GPU overclocking and somewhat less on memory overclocking--therefore, if the GeForce FX has a similar aniso implementation, it will be less constrained by memory bus (relatively) under aniso than without aniso.

It will certainly be a change if an nVidia chip benchmarks most favorably against the 9700 with anisotropic filtering applied!

lol, I was reading the graphs for that section the wrong way around
I did think it was a bit odd to have aniso depending as much on memory as I read it.