3GHz 384bit vs 6GHz 192bit GDDR5

marshal11

Distinguished
Feb 13, 2012
641
0
19,160
i'm a little confused about memory speeds and bus width. would 6GHz GDDR5 192bit memory interface width be the same speed as a 3GHz 384bit memory interface width?

the reason i'm asking is because i'm debating between getting a 660ti or a 670. i'm worried about the 192bit interface keeping the 660ti from releasing it's full power at 1080P. should i fork out the extra 100$?

(just so there is no confusion, i was just using random examples for my first question, not real specs of GPUs.)
 
Solution


That's correct, they're just mentioning different metrics for the same measurement. The command clock, data clock, and transfer rate are all...
6Ghz GDDR5 is just a marketing term. The transfer rate for any DDR device is twice that of the reference clock. In the case of "6Ghz GDDR 5" the transfer rate is 6 giga transfers per second, and the data reference clock is 3Ghz. In some cases this can be very difficult to distinguish.

To make matters worse, the data clock on GDDR5 is twice that of the command clock which means that there are 4 IO transfers per physical command clock. This makes it QDR (Quad Data Rate) with respect to the command clock. This second clock is what separates GDD5 from GDDR4, as GDDR4 is otherwise very similar to DDR3.

Some programs such as MSI Afterburner report the command clock rate and allow you to modify it, while others report the data clock rate and allow you to modify that, still others allow you to modify the data rate itself and divide backwards to alter the command clock which is what actually gets changed.

DDR3 has standard specifications for DDR3-800 which has a transfer rate of 800 megatransfers per second (abbreviated as MT/s) and an IO bus clock of 400Mhz. Similarly, DDR3-1600 has a transfer rate of 1600 megatransfers per second, and an IO bus clock of 800Mhz. Without knowing whether the device in question is marketing its transfer rate or IO bus clock you can't really know for sure.

The 660Ti and 670 both use the same 6 gigatransfer SGRAM. However, some manufacturers may label it as 6Ghz for marketing purposes while others may label it as 3Ghz (data clock), or even 1.5Ghz (command clock) for technical accuracy. They may also increase or decrease the frequency for added performance. It is the same either way.
 

marshal11

Distinguished
Feb 13, 2012
641
0
19,160
ahh i see. i didn't even think about that because nVidia didn't double the transfer speeds with the fermi architecture, but with kepler they did, making it confusing to compare the two. (fermi: http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-560/specifications kepler: http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-660ti/specifications )

okay. so the 660 ti has 3GHz GDRR5 192bit. say another card has 1.5GHz GDDR5 384 bit. would this mean they technically have the same transfer speed?

which leads to my other question. is it a good idea to get the 670 over the 660 ti?
 

sherlockwing

Honorable
Aug 7, 2012
412
0
10,810


Depend on what game you want to play and on what resolution/FPS:

price/performance 670 isn't going to beat the 660 just like 680 could never beat 670, you could Overclock to get close the above card's stock performance but you are still behind when they OC:

Judging by this (stock performance)
BF3%20%201920.png


You will do just fine with a good Factory OCed(MSI/EVGA etc) 660Ti playing on a 60 FPS TN/IPS monitor as it should get fairly close to 60 FPS when OCed.

670's #1 selling point right now is that a pair of 670 in SLI can saturate a 120hz monitor with 120 FPS game play where as 660 SLIs falls short and 680 SLI/single 690 are +$200 more expensive while and the extra performance over 120 FPS isn't that noticeable.


Personally I use a Gigabyte N-670OC/GD2(Factory OC version of GTx 670) because I intend to set up a SLI next month to maximize my 120 Hz monitor.
 


That's correct, they're just mentioning different metrics for the same measurement. The command clock, data clock, and transfer rate are all multiples of each other so it doesn't matter which value is mentioned so long as it is clear what that value represents. 6GT/s (sometimes written as Ghz, but this is a bit of a misnomer), 3Ghz, and 1.5Ghz are all different metrics for the exact same device. Similarly, 5GT/s, 2.5Ghz, and 1.25Ghz would all be the same. It's possible to have an overlap like the DDR3 specification that I mentioned, but I don't know of any devices which would have GDDR5 memory that slow.

To answer your other question, the 660 Ti has been panned as having lackluster memory bandwidth and this can have an impact on anti-aliasing, high resolutions and other memory intensive processing effects.

The 670 is usually the king of the hill in terms of bang for buck at 1080p and would get my recommendation. There are more powerful cards out there but these are only necessary at higher resolutions such as 2560x1440 and 2560x1600. Since the 670 can play all but the most demanding games at maximum settings at 1080p anything more is just wasted power at that resolution.
 
Solution