Video memory - do I need more?

[BFG10K]

Originally posted by: Azn

I never denied Ultra wasn't better with AA. I said some games 9800gtx wins even with AA over the ultra. I guess ultra wins more with AA in average. 2560 is more of a vram abnormally than anything else.

Forget about 2560*1600; the point is the other figures show the Ultra being 5%-10% faster or thereabouts.

So therefore my tests of a 8800 Ultra against a GT260+ are basically identical to testing a 9800GTX+ vs GTX260, tests that showed a large difference.

 

[AzN]

Originally posted by: BFG10K

Originally posted by: Azn

I never denied Ultra wasn't better with AA. I said some games 9800gtx wins even with AA over the ultra. I guess ultra wins more with AA in average. 2560 is more of a vram abnormally than anything else.

Forget about 2560*1600; the point is the other figures show the Ultra being 5%-10% faster or thereabouts.

So therefore my tests of a 8800 Ultra against a GT260+ are basically identical to testing a 9800GTX+ vs GTX260, tests that showed a large difference.

For you maybe it's showing a large difference at the settings you desire but in most hardware sites it does not. Did you record those fps? You came up with a conclusion on 1 web based benchmarks over 9800gtx+?

Well here's a recent test that includes 9800gtx+ vs all kinds of 260...

http://www.hardwarecanucks.com...eo-card-review-10.html

When AA isn't applied it looks like GTX 260/192 isn't all that much faster over 9800gtx+. Maybe 10~15% average. With AA about 15~25% long as it's not running out of vram. So 30% faster in average is a little excessive.

 

[BFG10K]

Originally posted by: Azn

For you maybe it's showing a large difference at the settings you desire but in most hardware sites it does not. Did you record those fps? You came up with a conclusion on 1 web based benchmarks over 9800gtx+?

Heh. I write video card reviews for a website, so of course I wrote them down. 😉

When AA isn't applied it looks like GTX 260/192 isn't all that much faster over 9800gtx+.

If it?s not much faster then you?re testing CPU limited settings, plain and simple.

I mean why test CPU limited settings (i.e. without AA) when comparing video cards? Of course you?re going to flat-line but what does that prove, other than CPU limitations at play?

Again it?s really quite easy to see:

http://www.computerbase.de/art...rmancerating_qualitaet

1280x1024x4: 14% faster.
1280x1024x8: 36% faster.
1600x1200x4: 20% faster.
1600x1200x8: 52% faster.

Again, the GTX260 is substantially faster overall in a range of games with 4xAA or 8xAA over the 9800 GTX+. And again I?ve tested a ton of games (much more than a typical website does) and my findings back my claims.

 

[AzN]

Originally posted by: BFG10K

If it?s not much faster then you?re testing CPU limited settings, plain and simple.

I mean why test CPU limited settings (i.e. without AA) when comparing video cards? Of course you?re going to flat-line but what does that prove, other than CPU limitations at play?

Again it?s really quite easy to see:

http://www.computerbase.de/art...rmancerating_qualitaet

1280x1024x4: 14% faster.
1280x1024x8: 36% faster.
1600x1200x4: 20% faster.
1600x1200x8: 52% faster.

Again, the GTX260 is substantially faster overall in a range of games with 4xAA or 8xAA over the 9800 GTX+. And again I?ve tested a ton of games (much more than a typical website does) and my findings back my claims.

CPU limitations? How do you figure when different cards were tested in same setup all the way upto 2560x1600 resolution. It's not the cpu when testing across multiple resolutions. It's the strength of the GPU core and SP that is being pushed here. Of course bandwidth does play a role here but not as much when considering with no AA. I don't know why you even mention CPU limitations. It's not like G92 won't gain performance with faster CPU as it will with GTX series.

Just look at COD4 with no AA @ 2560x1600...

9800gtx+ = 41.61 fps

260-192 = 46.78 fps

That's something like 12% difference.

Now when you look at 1680x1050 4xAA

9800gtx+ = 53.83 fps

260-192 = 67.03 fps

Here you see 25% better performance with 4xAA. i'm sure it gets worse as you push 8xAA.

What does this mean? That in bandwidth limited situations like AA 9800gtx+ does really bad compared to 260gtx considering bandwidth is the biggest difference between these 2 cards. I don't consider a 260gtx a more powerful core because it has more bandwidth. On the contrary. If G92 had gddr5 and 1gig of vram it would easily match GTX260 level of performance even with AA.

You might have tested a lot of games but you seem to be confused what is pushing what. Biggest difference when considering AA and higher levels of AA which you seem to be testing your GTX260 compared to 9800gtx is bandwidth. This is the same reason why 9800gtx loses to 8800ultra with AA even though G92 is more powerful than G80.

 

[Paratus]

@Apoppin

I'd be careful. I dont't know if you have 1200 hours to blow on another game.

😉

 

[Dadofamunky]

Originally posted by: BFG10K

Originally posted by: Azn

I wouldn't go far as 9800gtx being slower than the ultra.

With AA it is:

http://www.computerbase.de/art...rmancerating_qualitaet

Overall in all tested games, with 4xAA, the 8800 Ultra is 5% faster @ 1280x1024, 7% faster @ 1600x1200 and 32% faster @ 2560x1600.

With 8xAA the 8800 Ultra is 9% faster @ 1280x1024, 48% faster @ 1600x1200 and 379% faster @ 2560x1600.

So yes, the 8800 Ultra is faster with 4xAA and massively faster with 8xAA.

:Q

+1

 

[BFG10K]

Originally posted by: Azn

CPU limitations? How do you figure

Because the scores were not changing relative to the GPUs? relative strength until AA was enabled. Ergo, neither card was being saturated enough to reach its full performance threshold, so you mistakenly claimed the cards were close to each other.

when different cards were tested in same setup all the way upto 2560x1600 resolution.

And? You seem to somehow have a mistaken idea that 2560x1600 automatically implies a GPU limitation. Anyway, I wasn?t even talking about 2560x1600 so stop bringing it up.

It's not the cpu when testing across multiple resolutions.

It is if the resolution is not placing a high enough load on the GPU. Again just because 2560x1600 is used it does not automatically imply a GPU limitation. It can do, but that?s not always the case.

Just look at COD4 with no AA @ 2560x1600...

9800gtx+ = 41.61 fps

260-192 = 46.78 fps

That's something like 12% difference.

Now when you look at 1680x1050 4xAA

9800gtx+ = 53.83 fps

260-192 = 67.03 fps

Here you see 25% better performance with 4xAA. i'm sure it gets worse as you push 8xAA.

Right, that?s my point exactly. When AA is enabled it pushes the cards closer to their performance threshold and starts painting a truer picture of their relative standing. Without AA you?re running CPU limited tests and thus the smaller differences aren?t an accurate gauge of the relative standing of the cards.

I don't consider a 260gtx a more powerful core because it has more bandwidth. On the contrary.

Who cares about the core? Do you use a core or a whole video card? My original claim was that the GTX260 is quite a bit faster than the 9800 GTX+ (when tested properly), and that?s quite true.

Biggest difference when considering AA and higher levels of AA which you seem to be testing your GTX260 compared to 9800gtx is bandwidth. This is the same reason why 9800gtx loses to 8800ultra with AA even though G92 is more powerful than G80.

Again I never argued what the reason was, I simply argued that GTX260 was much more powerful despite your claims to the contrary. I also pointed out that your scores were CPU limited by demonstrating benchmarks with AA showing performance more in line with the expected performance differences.

 

[AzN]

Originally posted by: BFG10K

Because the scores were not changing relative to the GPUs? relative strength until AA was enabled. Ergo, neither card was being saturated enough to reach its full performance threshold, so you mistakenly claimed the cards were close to each other.

What does that have to do with CPU limitations?

And? You seem to somehow have a mistaken idea that 2560x1600 automatically implies a GPU limitation. Anyway, I wasn?t even talking about 2560x1600 so stop bringing it up.

It is testing for GPU limitations when the benchmarks were done with exact same setup and when testing across all resolutions with different cards which you have mistaken., If you look at 4870x2 with the exact same setup it is pumping out nearly twice as much frame rates @ 2560x1600 with no AA.

260/192 = 46.78 fps
4870x2 = 77.78 fps

It is if the resolution is not placing a high enough load on the GPU. Again just because 2560x1600 is used it does not automatically imply a GPU limitation. It can do, but that?s not always the case.

Right.. :laugh: But they tested across all resolutions upto 2560x1600 with different GPU's to find out GPU limitations not CPU.

Right, that?s my point exactly. When AA is enabled it pushes the cards closer to their performance threshold and starts painting a truer picture of their relative standing.

When AA is involved it is mostly testing for bandwidth limitations as AA needs more bandwidth to run more efficiently.

Without AA you?re running CPU limited tests and thus the smaller differences aren?t an accurate gauge of the relative standing of the cards.

Prove it. 😛

2560x1600 no AA
260/192 = 46.78 fps
4870x2 = 77.78 fps

Who cares about the core? Do you use a core or a whole video card? My original claim was that the GTX260 is quite a bit faster than the 9800 GTX+ (when tested properly), and that?s quite true.

Who cares? Core tells how powerful a chip is and it's capability when bandwidth isn't a issue. Tis the same reason why 9800gtx+ does so relatively well again GTX260/192 without AA. With bigger bandwidth G92 can be good as GTX260 even with AA. That's easy as slapping GDDR5 on a g92 to make it competitive against GTX260. There's a point of no return when bandwidth has been fully saturated > 2900xt which GTX260 seems to fully saturated. GTX260 has nothing more to give back at this point when more bandwidth is added but a G92. 😀

Again I never argued what the reason was, I simply argued that GTX260 was much more powerful despite your claims to the contrary. I also pointed out that your scores were CPU limited by demonstrating benchmarks with AA showing performance more in line with the expected performance differences.

I never denied the facts that 260 wasn't faster but to say a card is 30% faster was the problem I had initially. That would depend entirely on settings of Card A and B strengths.

 

[BFG10K]

Originally posted by: Azn

What does that have to do with CPU limitations?

The GPUs aren?t saturated enough to show true differences. You said it yourself: AA hits the memory more.

When AA is involved it is mostly testing for bandwidth limitations as AA needs more bandwidth to run more efficiently.

Right, exactly, and more pixel fillrate from the ROPs.

Prove it.

Again this is very easy to prove:

2560x1600x0:
GTX260: 46.78
9800GTX+: 41.61.
% difference: 12%.

Now add 4xAA to the mix:

GTX260: 42.85.
9800GTX+: 32.01.
% difference: 38%.

Given the difference in cards goes from 12% to 38%, it?s obvious the 9800GTX+ isn?t being taxed enough when it?s benched without AA.

Who cares? Core tells how powerful a chip is and it's capability when bandwidth isn't a issue.

So? Again, do you drop a GPU core straight onto your motherboard or do you drop a video card that includes VRAM?

You can?t just dismiss differences by saying ?well this is happening but it wouldn?t if it had more bandwidth, so it doesn?t count?. That?s irrelevant. The point is that it?s happening because the video card is a total package.

I never denied the facts that 260 wasn't faster but to say a card is 30% faster was the problem I had initially.

I?ve demonstrated repeatedly that the performance hit can be 30% (or more) and your decision to pretend it?s not happening by making theoretical comments about the core doesn?t change anything.

 

[AzN]

Originally posted by: BFG10K
Again this is very easy to prove:

2560x1600x0:
GTX260: 46.78
9800GTX+: 41.61.
% difference: 12%.

Now add 4xAA to the mix:

GTX260: 42.85.
9800GTX+: 32.01.
% difference: 38%.

Given the difference in cards goes from 12% to 38%, it?s obvious the 9800GTX+ isn?t being taxed enough when it?s benched without AA.

Now you have changed your original statement that was questioned. You specifically said CPU limitations which I was asking about not GPU limitations. I quote your original statement.

Without AA you?re running CPU limited tests and thus the smaller differences aren?t an accurate gauge of the relative standing of the cards.

All cards perform better with faster CPU long as as it's not GPU bound. This should be a given yet the benchmarks have nothing to do with CPU limitations because performance was measured with the exact same setup testing GPU limitations. Now if you test with faster CPU, performance will go up for both 9800gtx and GTX260 long as it's not GPU bound. You get me?

You can just look at my prior post. Here I will quote it again.

Just look at COD4 with no AA @ 2560x1600...
9800gtx+ = 41.61 fps
260-192 = 46.78 fps
That's something like 12% difference.

Now when you look at 1680x1050 4xAA
9800gtx+ = 53.83 fps
260-192 = 67.03 fps

Here you see 25% better performance with 4xAA. i'm sure it gets worse as you push 8xAA.

If you look at those benchmarks 260 seems to be better with AA because of it's massive bandwidth even on a mainstream resolution like 1680x1050 but at 2560x1600 no AA the bandwidth limitations have been lifted for the most part and painting truth about GTX260 far as how strong the core really is. Without the bandwidth G92 as powerful as GTX260. This is what I was trying to get at from the get go.

So? Again, do you drop a GPU core straight onto your motherboard or do you drop a video card that includes VRAM? You can?t just dismiss differences by saying ?well this is happening but it wouldn?t if it had more bandwidth, so it doesn?t count?. That?s irrelevant. The point is that it?s happening because the video card is a total package.

Did I say I that? Anyways. my original point was that GTX260 isn't that much more powerful compared to G92 given a G92 can easily match a 260 if it supported GDDR5 circuitry and 1 gig of vram. That was my point. And not everyone cares for AA filtering. I'm sure lot of people do. AA isn't a standard yet. I know you care about AA. I use a low pixel pitch monitor I could care less about AA and I'd rather use higher resolutions. When games like Crysis hit everyone stopped using AA because the performance wasn't there with modern GPU not to mention there are games released this year with NO AA support.

I?ve demonstrated repeatedly that the performance hit can be 30% (or more) and your decision to pretend it?s not happening by making theoretical comments about the core doesn?t change anything.

Yes you have but depending on the settings that 30% shrinks to 12%. That's my point.