Memory bandwidth for video cards...

[imzjustplayin]

I'm trying to wrap my head around why the 6600GT can perform about the same or better than the 9800pro. The 6600GT has much less memory bandwidth than the 9800pro but appears to make up for this shortcoming by having a ridiculously high core speed. Is it that the games the 6600GT and 9800pro can play are more demanding on the pixel filling instead of the bandwidth?


Err, I'm not quite sure what exactly stresses the memory bandwidth aspect of a video card besides the loading textures within the game (depending on quality). They say that X800XL video card with 512MB ram isn't any faster than the 256MB model because 'it can't utilize the memory' due to its speed and this same arguement was used with the 9800pro 128 v. 256mb version but I believe this to be false.. I'm just confused I guess as I can't seem to grasp at the situation with the 9800pro and 6600GT. I do know that memory bandwidth plays a factor as you increase the screen resolution so I guess it should win at 1600X1200 but then I'm told that at that resolution the cards can't handle any game anyways... Are games just improperly tuned for these card(s)?

The reason I'm so baffled is because at 1280X1024 the 9800pro and 6600GT should be on par/6600gt faster and the 6600GT has much faster core speed but half the mem bandwidth due to the 128bit. With these facts, the games that these two cards are capable of playing seem to depend more on the core speed. So with this train of thought why when overclocking do I notice a small performance improvement when overclocking just the core speed or just the memory speed but a significant with both??? I tried this out in Futuremark programs and in counterstrike FPS test which BTW made me realize that the new Steam patch has killed my FPS from what it used to be.. (re ran some tests and got lower scores)

 

[MrFace]

Both chips are completely different cores. There is more to how a video card works than memory size and "LOLMHZ".

 

[mavalpha]

The shaders in the 6600 are much more efficient than the 9800's over all; they can do the same amount of work with less available resources.
Why was the Pentium-M as powerful as the Pentium4, even when the P-M was still stuck in a single-channel-memory chipset?

 

[imzjustplayin]

The shaders in the 6600 are much more efficient than the 9800's over all; they can do the same amount of work with less available resources.
Why was the Pentium-M as powerful as the Pentium4, even when the P-M was still stuck in a single-channel-memory chipset?

I don't believe thats true for ALL tasks as the P4 is still very good at multimedia and high bandwidth requirement applications. Its like its got really big hands and strong arms and can really move stuff but when it comes to things like soldering, has a very difficult time doing it.

Both chips are completely different cores. There is more to how a video card works than memory size and "LOLMHZ".

Yes but what I was trying to get at was not so much the core speed differences but that it had a fast core but limited memory bandwidth. It just seems kind of odd and that definately doesn't explain why when overclocking the 9800pro, I have to overclock the memory and core to get anything significant when on the 6600GT it is on par/out performs and it has half the memory bus speed...

 

[MrFace]

I don't believe thats true for ALL tasks as the P4 is still very good at multimedia and high bandwidth requirement applications. Its like its got really big hands and strong arms and can really move stuff but when it comes to things like soldering, has a very difficult time doing it.



Yes but what I was trying to get at was not so much the core speed differences but that it had a fast core but limited memory bandwidth. It just seems kind of odd and that definately doesn't explain why when overclocking the 9800pro, I have to overclock the memory and core to get anything significant when on the 6600GT it is on par/out performs and it has half the memory bus speed...

That's because the 6600gt core is so much better that the memory "limitation" doesn't affect it whatsoever.

 

[Marvelous]

6600gt has higher fillrate and slightly lower memory bandwidth...

It's also one generation ahead of 9800pro.

 

[Anarchist4000]

AA is what really sucks up memory bandwidth a lot of the time. That or using insane amounts of overlapping textures in a game, I mean using them not actually storing them on the video card.

The 6600 looks to have 1/3 more transistors than the 9800 so you're looking a faily significant increase in capability. The clockspeed also looks like it's significantly higher so the only way it'd be close is if you were doing some serious AA and texturing in a game.

 

[imzjustplayin]

AA is what really sucks up memory bandwidth a lot of the time. That or using insane amounts of overlapping textures in a game, I mean using them not actually storing them on the video card.

The 6600 looks to have 1/3 more transistors than the 9800 so you're looking a faily significant increase in capability. The clockspeed also looks like it's significantly higher so the only way it'd be close is if you were doing some serious AA and texturing in a game.

Forgot about AA, you're right about that. Turns out that (from what I saw) that the 6600GT doesn't support anything higher than 4X AA while the 9800pro has 6Xaa. (Noticed this when I was on my friends computer configuring BF2 video settings and AA only had 2 and 4X while mine has 6X.. Question is, is the 9800pro able to handle 6X aa in a game like BF2 over the 6600GT...

What I really wonder is, if the video card has the memorybandwidth available, shouldn't turning on AA not kill performance? So why is there a performance hit then when cards turn on AA in video games? Is there a program to measure how much bandwidth or % of the card's capacity is being used?