I thought Nvidia OpenGL performance was supposed to be a tier above their AMD counterparts? What gives?
I thought Nvidia OpenGL performance was supposed to be a tier above their AMD counterparts? What gives?
Again, there's absolutely no good reason why that should happen, especially since both nVidia parts sound like jet engines and are absolute furnaces in comparison.
edit : He states , if you are a watching this and you a Nvidia fan , chances are you have a decent Nvidia card and will be fine playing the game.September 23, 2011
By Andrew Burnes
id Software have created some of the most loved and most played games of all time, such as Return To Castle Wolfenstein, Doom, and Quake. Early next month the Texan studio will launch Rage, a brand new intellectual property full of new ideas, varied gameplay, co-op, and multiplayer, and it is these points that we discussed with Studio Head Tim Willits at EuroGamer Expo 2011.
We also spoke about tech with Tim, taking a moment to work through a special Transcoding option for those with fast GPUs.
“When you have a high-end graphics card,” says Tim, “you can apply [it] to Transcoding, so when you whip around fast in Rage the world comes into focus extremely quickly. And of course, if you have a high-end video card with lots of video RAM, your texture size will be larger, which will give you much crisper textures much farther away than you would normally see on a console.”
We also talk about scalability and catering to those with mid and low-range systems, so if you want to learn more click on the video above.
http://www.geforce.com/News/article...how-rage-takes-advantage-of-your-high-end-gpu
A developer speaks about the game in this short video.
edit : He states , if you are a watching this and you a Nvidia fan , chances are you have a decent Nvidia card and will be fine playing the game.
What are you basing that statement on? In these tests the GTX470 has a higher Z fillrate than the 5870:Zixel fill rate is the issue.
What are you basing that statement on? In these tests the GTX470 has a higher Z fillrate than the 5870:
Doom3 engines + Nvidia always have the worst black levels.
It's why I said Zixel fill instead of stencil fill.
http://www.beyond3d.com/images/reviews/Slimer-arch/TexturedFillrateBilinearAbsolute-big.jpg
Combine the two and it paints a fairly clear picture of exactly where the limitations are. Moving from 4x to 8x AA on the 470 should roughly cut framerates in half under D3 20x16(shouldn't be quite linear, but close).
Back in the 5800 days nVidia and ATi were close on pixel fill, but nV killed them on stencil giving them a big overall edge in Zixel. Now they are both close on stencil, but ATi clobbers nV on pixel giving them a fairly large overall edge.
FYI- nVidia was refrast for DooM3, if it isn't as dark as nV it's wrong(Carmack is the one that said that nV was refrast, not nV).
The OpenGL performance problems section on wikipedias Fermi page has vanished. How interesting.
Rollo must of been doing some edit work.
Don't care. It looks wrong. Too dark.
Many PS3 games have this problem as well (very noticeable when playing Portal 2), and even when doing things like playing Netflix. I maintain that it's either a hardware fault or nVidia driver issue.
http://www.youtube.com/watch?v=kI5JVL7M1yI&feature=player_detailpage#t=27s
My graph isn’t of stencil fill, it’s depth (Z). The GTX470 is faster than the 5870.It's why I said Zixel fill instead of stencil fill.
That’s texel fillrate, absolutely nothing to do with zixel.
You seem to be mixing three different things here. Zixel is not the same as texel which is not the same as pixel. I’m not even sure what you’re trying to infer.Combine the two and it paints a fairly clear picture of exactly where the limitations are. Moving from 4x to 8x AA on the 470 should roughly cut framerates in half under D3 20x16(shouldn't be quite linear, but close).
Back in the 5800 days nVidia and ATi were close on pixel fill, but nV killed them on stencil giving them a big overall edge in Zixel. Now they are both close on stencil, but ATi clobbers nV on pixel giving them a fairly large overall edge.
Fermi does 33% more depth/stencil (the Depth/stencil hardware function is combined) samples per clock than AMD's GPUs do, which is why AMD partially makes up for the performance difference and they use more aggressive z clear and/or lossy depth buffer compression. Even though AMD says their Depth buffer compression is lossless, it's not unless they're just clearing their zbuffer faster than necessary.My graph isnt of stencil fill, its depth (Z). The GTX470 is faster than the 5870.
Thats texel fillrate, absolutely nothing to do with zixel.
You seem to be mixing three different things here. Zixel is not the same as texel which is not the same as pixel. Im not even sure what youre trying to infer.
The pixel and texel fillrate specs for a 6850 are well below that of a GTX580, yet the former still runs faster in Doom 3. Specs cant explain that, but poor OpenGL drivers can.
Not to mention that none of this explains the severe performance slowdowns in professional OpenGL applications or synthetic OpenGL tests (e.g. glReadPixels being up to 10 times slower on Fermi than previous hardware).
Interestingly programmers report some of the performance goes back to where it should be when a Cuda implementation is done instead. That wouldn't happen if it was a hardware limitation based on specs. Again, this is pointing to poor OpenGL drivers.
It's not like this is anything new - G80 drivers were absolutely woeful in OpenGL and it took them a good two years to get everything where it should be.
Official specs confirm that nvidia does 8 zixels/clock and that AMD does 6. As for my comments on AMD's optimizations, I've observed it myself--their depth range is definitely not long as nvidia's. Not many people notice it, so that's why they get away with it.^^You got a source for that?
Can you please show us some documented evidence of this? Thanks.Even though AMD says their Depth buffer compression is lossless, it's not unless they're just clearing their zbuffer faster than necessary.
That's not evidence. Can you please show us evidence to back your claims? Thanks.As for my comments on AMD's optimizations, I've observed it myself--their depth range is definitely not long as nvidia's. Not many people notice it, so that's why they get away with it.
Official specs confirm that nvidia does 8 zixels/clock and that AMD does 6. As for my comments on AMD's optimizations, I've observed it myself--their depth range is definitely not long as nvidia's. Not many people notice it, so that's why they get away with it.