Published on 16th December 2013 by
Originally Posted by SchizoFrogTell me something though, at 1080 res are we starting to see 2GB of memory being maxed out by newer games?
Originally Posted by SchizoFrogShould gamers be looking for 3GB (am I right in thinking you need a memory bandwidth of 384 to use 3GB?) and 4GB cards instead if we want our cards to live longer than the next 12 months? Also, can someone just briefly explain again about the limits factored in by the memory bandwidth as I seem to remember previous comments about a memory bandwidth of 256 not being able to use all of 4GB of memory.
Originally Posted by Corky42
To access all 3GB in one cycle you would need 384bit bus, as each RAM module uses a 32bit bus AFAIK. You can get cards with more ram like you say with a 256bit bus and 4GB ram, but the ram modules will split the bus between them.
Originally Posted by Dogbert666
...at 1080p I would say that 2GB of VRAM is currently not a performance limit by itself. As others have pointed out, the various other specs of GPUs like the 270X are going to limit it before the 2GB does, especially at this resolution.
Whether gamers should be looking at 3GB/4GB cards is honestly hard to say. Such cards usually come with more powerful GPUs anyway, so if you want it to last longer then the standard advice of investing more holds true. But whether you should get a 4GB version of the GTX 770 or R9 270X is where it gets a little trickier - I haven't tested either but right now I reckon you'd see zero difference at 1080p. That said, since both consoles now have 8GB of memory theoretically available to the APUs (less once other system resources are accounted for, but still plenty more than 2GB), we could see games starting to be developed that benefit from more than 2GB, which would carry over into the PC space. That said, the actual console APUs are relatively underwhelming compared to most modern GPUs, so it all depends how well developers are able to optimise games for the hardware available to them. It's an annoying case of having to wait and see.
Originally Posted by Dogbert666I'll continue with these cards as examples. On their 384-bit bus (6 x 64-bit memory controllers), whether you have GTX 780's 3GB (12 x 256MB memory modules, 2 per controller) or Titan's 6GB (24 modules, 4 per controller), the entire address space (3GB or 6GB) is available to the controllers in every clock cycle, as the ability to double up GDDR5 SDRAM modules (like Titan does) without performance loss is built into the specification. Therefore, for GTX Titan, the GDDR5 will be set to operate in x16 mode (or 'Clamshell Mode') rather than x32, so the 32 bits of data can be split between the modules in each cycle.
Originally Posted by SchizoFrogI think this would be a great article to put together to compare not only AMD against nVidia but also say, 2 different versions (2GB/4GB) of 2 different cards GTX760/GTX770 (on nVidia's side and likewise a similar match up for AMD) to see what difference the memory makes between the mid level GPUs and the high end GPUs when used at 1080 (a higher resolution would start to effect the GPU and then wouldn't give true memory differences, at least I think that would be the issue).
Originally Posted by Corky42Sorry if i come across as being a bit dense, but if i have understood correctly in x16 mode (or 'Clamshell Mode') would that mean the ram module only contains half the data sent from the controller? (why do i feel like i have just asked the dumbest question in the world) :o
You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.
17th January 2017
13th January 2017
© Copyright bit-tech