BF2S Forums Difference between DDR2 and DDR3

..teddy..jimmy wrote:

sorry i cant help..dnt understand anything about computers

So why answer then?

no no, the 7600GS is both GDDR2, and the 7600GT is GDDR3

basically its the speed at which things flow through the memory. GDDR2 is kinda old, and GDDR3 is faster. there are some cards that will feature GDDR4 very soon.

sooo...512 of GDDR2 is better than 256 of GDDR3?

There are GDDR4 cards out already.. ATI Radeon X1950

Just a little curious, but what exactly are you talking about?  Reason i ask is cause the only thing i know that has the abbreviations DDR is Dance Dance Revolution...LOL, but i surely know thats not what you are all talking about.  So, would someone enlighten me?

DazBurt wrote:

..teddy..jimmy wrote:

sorry i cant help..dnt understand anything about computers

So why POST then?

..teddy..jimmy wrote:

sorry i cant help..dnt understand anything about computers

Or spelling.

Twist wrote:

.|microphage wrote:

sooo...512 of GDDR2 is better than 256 of GDDR3?

Nope. The GDDR3 is faster, and the extra memory is even MORE of a disadvantege, because you will need to address more memory, thus slowing down the memory operations. In general, ALWAYS choose the faster memory, and ALWAYS buy the lowest possible amount of memory that you can make due with.
For instance, if you have a monitor that will go 1280x1024 max, and you want to run it at true color (32 or 36 bits depending on your preference), you multiply the factors, and you will have your GPU memory consumption (number of horizontal pixels, times number of vertical pixels times number of bits for color in each pixel plus hamming codes for each block, blocksize may vary depending on manufacturer) when running in a "normal" windows environment. So you will notice that for windows usage you will "only" need around 40 MB of GPU memory (64MB for 1600x1280). Generally speaking a good GPU will also be able to handle artifacts, vectors, surfaces, etc. etc. Shading, AA and other "effects" are calculated in realtime, so you normally dont need a lot of memory for these, though it IS calculated in BLOCKS, so you will need SOME memory. Therefor you also need to "make room" for these things in your memory. So double up on your memory for games, and you will find yourself needing a 24'+ monitor to utilize more than 128 MB of GPU memory !

If you doubt my calculation, try running a program to determine the GPU memory usage while running an app. You will notice that you rarely use all of it. So if the extra memory is adressed, you waste memory bandwidth by having more memory than you need, and thus you drag down your performance.

Mostly true. I disagree totally about needing a 24" monitor to utilize more than 128MB of GPU memory. More memory is good. It helps a lot. The more stuff you can have stored on your video memory means less transfering stuff onto it, the less you have to move stuff about the faster it all goes.

Your calculations are flawed. You have been working stuff out for a single frame. At 1024x768 with 32-bit colour running at 50 fps you are looking at about 15GB/sec of memory bandwidth. Lots of video memory getting used there.

Cards with more video memory run faster. The 7800GTX 512 was faster than the 7800GTX 256. It may be that the addressing takes longer, but that rarely leads to slowdowns in my experience - 64-bit processors run faster not slower due to extended memory addressing techniques.

512MB of DDR2 could be faster than 256 of DDR3, it depends on the App. In your case you have 2 different GPUs as well, so no it wouldn't be faster.

Last edited by Bertster7 (2006-09-24 06:53:36)

Bertster7 wrote:

Twist wrote:

If you doubt my calculation, try running a program to determine the GPU memory usage while running an app. You will notice that you rarely use all of it. So if the extra memory is adressed, you waste memory bandwidth by having more memory than you need, and thus you drag down your performance.

Mostly true. I disagree totally about needing a 24" monitor to utilize more than 128MB of GPU memory. More memory is good. It helps a lot. The more stuff you can have stored on your video memory means less transfering stuff onto it, the less you have to move stuff about the faster it all goes.

Your calculations are flawed. You have been working stuff out for a single frame. At 1024x768 with 32-bit colour running at 50 fps you are looking at about 15GB/sec of memory bandwidth. Lots of video memory getting used there.

Cards with more video memory run faster. The 7800GTX 512 was faster than the 7800GTX 256. It may be that the addressing takes longer, but that rarely leads to slowdowns in my experience - 64-bit processors run faster not slower due to extended memory addressing techniques.

512MB of DDR2 could be faster than 256 of DDR3, it depends on the App. In your case you have 2 different GPUs as well, so no it wouldn't be faster.

Again... I refer you to my willingness to have you test this theorem. Simply download a program that will tell you how much memory you use...

And you are correct in assuming that the BANDWIDTH will be very high. This is why you need a FAST card on a FAST bus. but you don't need MORE memory to support a higher bandwidth.
As for the specific cards I have never EVER seen a higher mem card (assuming all else is equal) perform better than a lower mem card, EXCEPT when the memory is exhausted. And it CAN get exhausted because the frames (of which there's usually only ONE in memory) may take up a lot of memory, and leave no room for "other stuff". You will NEVER EVER see a GPU that holds two frames in memory (unless the second frame is for a second monitor), because the GPU isn't pshycic, it cant tell what's gonna come next. It will only post the changes to the frame as instructed to. And yes, this requires HUGE amounts of bandwidth, I'll never argue with that. But you can run BF2 just fine on a 128MB card, if you can find one. The "other stuff" doesn't take all that much memory.
This is also the reason why in the "old" days you'd be working with a "stationary" background, and then have "sprites", that moved ontop of the background. Because the less you needed the background image to change, the less bandwidth you needed to show the graphics. Ofcourse, this was only good if the sprites would not "change" anything except their postition on the screen. This would allow you to "package" your frame somewhat... A sort of early form of compression. A predecessor to the early movie compression algorithms.
GPU construction hasn't changed THAT much over the last 20 years. Nowadays the engineers just waste bandwidth like mad instead of trying to focus on the stuff that matters, like speed, color recreation, functionality, price etc. etc.