128 bit vs 256 bit

[fedex]

is there much difference with the 128 bit cards and the 256 bit cards?
say x700xt vs x800xt?

just one last thing what is the x700 like does anyone have one?
and got opinions on the card

[titan87]

a 256bit card will be faster than a 128 bit card as the 256bit card can move ~twice the ammount of data as the 128. but i may be wrong

[AMD64]

there is alot of differnce between 128 and 256 bit cards i dont know any of the technical stuff. but ppl will take 128mb 256 bit card over a 256mb 128 bit card

[1234]

go google x700xt and read a few reviews. x700xt sits with 6600gt in terms of performance. 6600gt will outperform or match an x800pro in doom3. in simple terms an x800xt is far superior to an x700xt however, both would be good for current games. just depends how much you are willing to spend.

1234

[ssX]

from wat a gather x700xt is canceled. x800 will replace it with x800xl the next card up from it.
(have a look on the news page for reviews.)
(tho i think that x700pro may still be around?)

as for the 128 vs 256 bit.
in modern games where there are thousands of textures, lighting etc to be processed a second, bandwidth is the key.
especially in high filtering, IQ and res environments.
wen the gpu is processing mass amounts of data, you want the data moving through ram as fast as possible. and as much data crammed down the 'pipe' as possible.


its not just that the x800xt has 256bit memory its that is also has 16pipes.

anyway someone may have a better explanation than that.

[phextwin]

256 bit eqauls twice the ram bandwidth for a given speed than 128bit ram.

More bandwith is good, as the gpu can acces the data needed for processing faster. Thus more fps.

[Fodder]

The bandwidth is only useful however when the GPU is fast enough to make use of it. Example, X700 is ~500MHz 8-pipe core/~500MHz 128-bit memory, but is solidly faster at times than the ~400MHz 8-pipe core/~400MHz 256-bit memory X800SE. In GPU dependant stuff like pixel shading (which is getting more and more common these days) memory bandwidth really doesn't matter much at all, which is how the dirt-cheap low-end 64-bit cards (ie. 9600SE) still manage to turn in acceptable framerates. We're currently reaching a point where we have ample fillrate and memory bandwidth (example, the 6800/X800/6600/X700 have significantly less bandwidth available per-pipe than the 9800/5900, yet still perform admirably), and right now the biggest (and easiest) gains to be had will come from boosting shading grunt even more, which helps to explain the push towards unified shader architectures.

[Rzeractor]

wouldnt you need an extremely high fill rate for a high shading power?

[Fodder]

Not necessarily, you can always include multiple shader units per pipeline. Not as effective as adding full pixel pipelines, but alot easier on the transistor budget.

[phathag]

generally speaking a 256 bit card with 128MB mem should outperform a 128bit of the same same card with 256MB of memory; problem is most people choose the other way around...

thus some manufacturers out there make cheap cards, stick 256 MB of memory in and droves of people end up misinformed and buy the entry level 256MB card than the better performing 128MB mainstream/high end cards.

[philip_pc]

if what is beign said here is true about 256bits being better than 128bits.....
then shoudlnt the old 9800XT/9800pro cards with 256bit beat the current 6600GT cards with 128bit............
or is it because the 6600GT's have been better refined and therfore perform better.......
i'm just not sure on this thing.....

[fedex]

woah old tread alert close peeps

[XST]

Originally Posted by philip_pc

if what is beign said here is true about 256bits being better than 128bits.....
then shoudlnt the old 9800XT/9800pro cards with 256bit beat the current 6600GT cards with 128bit............
or is it because the 6600GT's have been better refined and therfore perform better.......
i'm just not sure on this thing.....

woah, big bump.

its because theyre more refined. and bigger clocks.

similar to the geforce fx case, between 5800 and 5900. 5900 was more or less equal to 5800, in performance. better architecture and 256bit mem compensated for speeds.
well, in a nutcase that is

[Circastes]

I thought there was a large difference between 128bit and 256bit cards, until I read Fodder's post. I looked into some benchmarks, and it seems the bus width difference becomes more obvious when you enable IQ settings. I'm seeing a (128bit) 6600GT losing 17.2 FPS by enabling 4XAA (which, by the process that AA is, uses alot of memory bandwidth) in Doom3, while a (256bit) 9800XT only loses 5.4 FPS.

It cannot be put down to ATI's superior handling of AA modes, as when 4XAA is enabled, the two 256bit flagships (at the time of this review) the 6800U and X800XT PE trade places, with the 6800U taking the lead. The 6600GT is on the same core architecture as the 6800U.

[itsmydamnation]

well on my x800 overclocking my mem to 550ddr form 500ddr had very little difference. overclocking my GPU from 392 to 440 had about 1000points 3dmark 03 and very noticeable in games.

So this would suggest that bandwidth doesn't seems as important as raw clock speed.