Video Card Memory Analysis: 256MB vs. 512MB
How many forums have you visited in the last month with threads discussing whether 512MB video cards are really necessary? Well hopefully this article will shed some light on this topic.
A few weeks ago I reviewed the Gainward BLISS 7800 GT Goes like Hell, which is a supercharged 7800 GT with 512MB memory. It’s clocked at 450/650 (standard GT is 400/500 ) and that is undoubtedly the main reason for its superior performance, but an interesting question arose after the review went live – Did the 512MB of memory make any difference? I don’t believe anyone reading this would disagree that 512MB cards will become a firm requirement of high end gaming very soon, but the key question here is does 512MB of VRAM offer anything to us now, other than future proofing? To answer this we first need to understand exactly what happens to a 256MB card when you max out its memory (VRAM).
Simply put, the data that cannot be stored locally in VRAM is transferred to system memory where you will incur large latency penalties across the board. Technically the only penalty incurred should be fps losses, but a typical symptom you’ll probably encounter is stuttering caused by texture swapping from the system RAM to the VRAM, generally known as cache thrashing. This takes place because textures needing to be displayed on screen obtain absolute priority, so if a texture needing to go on screen is currently being held in system RAM it will need to be moved to the video card ram before it can be displayed. Like I said, technically this should not happen, but inefficient memory management in many game engines guarantees the stuttering issue a common one.
With AGP there is an option within your motherboard bios that allows you to select how much system memory can be used by the GPU. This is known as the AGP aperture size. You can choose various options (64, 128, 256 etc) but most people recommended you use one quarter the capacity of your system memory. With PCIe there is no such configuration left to the end user, it’s all handled automatically by the video card driver. From what I can see, its decision is sensitive to the amount of system memory you have available and will at maximum only ever double the amount of available memory. I’m not fully sure this is the case as I don’t have 2x1GB of memory here to see what happens. I’m confident that a 256MB card will still be doubled to 512MB regardless and a 512MB card will have 1GB available. I’ll be able to confirm this in the next few days and will update this article with my findings.