Getting to the bottom of VRAM sizes, bus speeds, and performance. Novel questions, I swear.

cyjackx

Honorable
Dec 23, 2013
10
0
10,520
In my researching between the 2gb and 4gb models of various cards, I've encountered numerous articles and benchmarks mentioning the bus speed.

People say several cards don't have the ability to use more than 2gbs of VRAM.

In theory re: the analogies of bus speed as highway lanes, I understand. If you don't have enough bandwidth to access all 4gbs of VRAM in one "trip", then naturally that bandwidth bottlenecks you.

But do you NEED all 4gbs every trip? Perhaps this is an innate technical function of VRAM that I don't understand, but I don't see why a 2gb card and a 4gb card sometimes seem to have the same performance. Even if the bandwidth is the same, there really isn't anything that can be kept on the VRAM for quicker access than from the hard drive? Why must all data come thru the tunnel simultaneously? You can't make two trips?

Is this for gaming specifically? Can nothing be kept in the VRAM to be saved for another trip? Can't pack 4gbs of data to park inside the VRAM and only use the 2gb you need right now and save the other 2 for later? Or does the VRAM have to be emptied/reset every trip? Perhaps that's what I'm missing.

Is this software dependant? My intended purchase is for video editing and rendering, does this change how vram is used? What if 4k editing is a possibility?

I'm pretty cynical sometimes, but even I find it hard to believe that the extra VRAM on certain models is nothing more than a marketing gimmick and has no practical value.
 
Solution
The problem with "waiting for another trip" is that, in many situations, the processor is doing exactly that: waiting on the memory. This is especially prevalent at high-resolutions, where lots and lots of data needs to be processed.

High resolutions is also where having more memory comes in, though many times, it's a bit of a gimmick; for many mid-to-low range cards, anything that would require that much memory is going to be unplayable anyway.

Perhaps as a better illustration may be as such: You have a factory (the GPU), a warehouse separated by some distance (the VRAM), and a warehouse further away (the RAM). The factory is almost always running under-capacity: it needs more stuff from the warehouse. The highway between the factory...
When your using SLI setups to power multi monitors for gaming the VRAM on the second card is un-usuable
as a result, if your on 75hz or 144hz refresh rate monitors with only 1gb VRAM on the main, the GPU will tap into system resources which is considerably slower than the onboard graphics ram which plays a major role in high levels of AA , HBAO/HDAO and other such visual enhancing effects. Im not too knowledgeable on all applications of the card though.
 
The problem with "waiting for another trip" is that, in many situations, the processor is doing exactly that: waiting on the memory. This is especially prevalent at high-resolutions, where lots and lots of data needs to be processed.

High resolutions is also where having more memory comes in, though many times, it's a bit of a gimmick; for many mid-to-low range cards, anything that would require that much memory is going to be unplayable anyway.

Perhaps as a better illustration may be as such: You have a factory (the GPU), a warehouse separated by some distance (the VRAM), and a warehouse further away (the RAM). The factory is almost always running under-capacity: it needs more stuff from the warehouse. The highway between the factory and the warehouse is almost always completely full of traffic; you can't fit more on it, but if you could, the factory could operate closer to capacity. If the warehouse is not anywhere near full, even when the highway is, making the warehouse bigger won't really help anything, but if the warehouse does, indeed, get full, any extra storage has to be done in the warehouse that is further away (RAM) with smaller roads to get to it, and still has to go through the main warehouse and its highways. In the vast majority of cases, the warehouse isn't full, and the factory wouldn't be able to process all that material in a timely manner, anyway, even with a bigger highway.

There are, of course, some exceptions. The 780Ti, for instance, runs out of room in the warehouse without necessarily flooding the highway when running at 4k on certain games. With that specific chip, running at 4k doubling the VRAM is probably worth it. There's also SLi/Crossfire configurations, where there are more "factories" but the warehouses must be identical, and might as well be the same one (though it's a little more complicated than that), but the amount of "highway" is increased, as well. At high resolutions, you may again run into the warehouse being too small.

Anyway, that's a gross oversimplification, but for the most part, no, more RAM doesn't do anything for a GPU. You may run into situations in non-realtime rendering (like video rendering), however, where more RAM is helpful because of the high resolution / textures / etc. taking up more memory than the GPU has, since you're not worried about "playability" in that case.

TL;DR: Most games never use that much memory, unless played at too high res for the GPU to handle. Can be helpful with multi-gpu or where a minimum framerate doesn't need to be maintained (at high resolutions).
 
Solution

TRENDING THREADS