Would a lower rated GPU such as a 760 utilize 50% of its VRAM where a 780 or 780ti would utilize less due to its higher processing power? If so would SLI then utilize an even smaller percent of its total VRAM? Yes I understand SLI does not double VRAM. Wanting to play Skyrim at 6000 x 1200 heavily modded and wondering which GPU to go with. Any input is appreciated
Wanting to understand the role of VRAM a bit better
- Thread starter Drommel
- Start date
OK, let me throw a crude analogy out there to see if it's helpful.
Imagine you are working from lots of pieces of paper, using information from the paper to produce a finished product. The finished product is also stored on a number of pieces of paper. Let's imagine too (for the sake of the analogy) that you can't stack paper on top of each other, so each piece of paper needs an amount of space accessible to it.
In that case, the size of your desk determines how many pieces of paper you can work from simultaneously, and how large your finished product can be.
Think of VRAM as the size of your desk. As long as everything you need fits on the desk, having extra space (or extra VRAM) won't make things any faster at all, it's just unused space. However, as soon as you run out of space, you run into big problems. If a graphics card runs out of VRAM it will start shifting things into system RAM, which (going back to the dodgy analogy) is little like using a filing cabinet. The information is still accessible if you need it, but it takes time (competitively a massive amount of time) to retrieve the information... and remember that before you can look at it, you have to file another piece of paper in system RAM to clear up desk space for the information you're trying to retrieve.
A faster GPU can process information more quickly (like having a more efficient worker), but it is VRAM that determines how much data a card can work with at any one time (how big the desk is).
In other words, I would expect VRAM usage to be more or less identical for different cards, provided you're using the same game at the same resolution and settings.
What needs more VRAM? Things like:
- High resolutions (like your 3x1200p setup), the "finished product" (the rendered frame) is much larger in this case, as well as the sheer amount of pixels that have to be processed and generated
- Post processing such as AA
- High resolution textures
- etc etc.
Hopefully that's helpful?
Or others may like to comment on this analogy?
Imagine you are working from lots of pieces of paper, using information from the paper to produce a finished product. The finished product is also stored on a number of pieces of paper. Let's imagine too (for the sake of the analogy) that you can't stack paper on top of each other, so each piece of paper needs an amount of space accessible to it.
In that case, the size of your desk determines how many pieces of paper you can work from simultaneously, and how large your finished product can be.
Think of VRAM as the size of your desk. As long as everything you need fits on the desk, having extra space (or extra VRAM) won't make things any faster at all, it's just unused space. However, as soon as you run out of space, you run into big problems. If a graphics card runs out of VRAM it will start shifting things into system RAM, which (going back to the dodgy analogy) is little like using a filing cabinet. The information is still accessible if you need it, but it takes time (competitively a massive amount of time) to retrieve the information... and remember that before you can look at it, you have to file another piece of paper in system RAM to clear up desk space for the information you're trying to retrieve.
A faster GPU can process information more quickly (like having a more efficient worker), but it is VRAM that determines how much data a card can work with at any one time (how big the desk is).
In other words, I would expect VRAM usage to be more or less identical for different cards, provided you're using the same game at the same resolution and settings.
What needs more VRAM? Things like:
- High resolutions (like your 3x1200p setup), the "finished product" (the rendered frame) is much larger in this case, as well as the sheer amount of pixels that have to be processed and generated
- Post processing such as AA
- High resolution textures
- etc etc.
Hopefully that's helpful?
Or others may like to comment on this analogy?
leeb2013
Honorable
- Aug 5, 2013
- 1,967
- 0
- 12,460
very good analogy. I have a few more questions that I'd like to tag on this thread too;
how come so much Vram is needed when a 1080 display is only 2MP (an uncompressed RAW 12MP image from my camera is only 4MBytes). What is being stored in the Vram other than frames?
Having recently bought a R9-290 with 4GB Vram, I've found that Titalfall is quite happy to use 3.7GB of it on one monitor, clearly it's making full use of it, but as it uses about the same for 3 monitors and will also run on a 1GB GPU (the graphics aren't exactly 'next gen') what is it using all the Vram for and what advantage is there in filling it as much as possible?
Thanks
how come so much Vram is needed when a 1080 display is only 2MP (an uncompressed RAW 12MP image from my camera is only 4MBytes). What is being stored in the Vram other than frames?
Having recently bought a R9-290 with 4GB Vram, I've found that Titalfall is quite happy to use 3.7GB of it on one monitor, clearly it's making full use of it, but as it uses about the same for 3 monitors and will also run on a 1GB GPU (the graphics aren't exactly 'next gen') what is it using all the Vram for and what advantage is there in filling it as much as possible?
Thanks
You'll need to wait for someone who knows more to answer your questions in detail. I know that all the textures and frame buffers are stored there, but I don't claim to be an expert. Also, I think you'll find a 12MB RAW image should be more like 15MB+, but still, nowhere near a big chunk of your VRAM I agree.
In terms of RAM usage with Titanfall I could have a guess... Given that unused RAM give you literally nothing, you may as well leave data in there if there's a chance it'll be needed again in future. Your Windows Task Manager for example tells you how much "Cached" data is stored in memory. This is data that isn't specifically needed for anything, but the OS has decided there's a chance it might be needed and may as well leave it memory in case. It's possible that Titanfall knows how much VRAM you have and is doing a similar thing. That's entirely speculation on my part however, again, I don't claim to be an expert.
Thanks for the reply rhysiam. That analogy basically fits my understanding of how it works although I was thinking that a faster processing GPU would be able to allocate VRAM storage faster than say a card with less power. Looking back perhaps my question was not worded as well I would like. I get that items like AA, resolution and HD textures would utilize more VRAM however, I was thinking that while a lower powered card might reach its max and begin to bog down and begin to use system RAM that a higher powered card would better manage its VRAM due to its faster processing. To use my own crude analogy say we had two GPUs, a low end card and a high end both with 2G of VRAM and a game were to require a definite 2G of VRAM, would both be hindered by the amount of VRAM and therefore begin to bog down equally?
leeb2013
Honorable
- Aug 5, 2013
- 1,967
- 0
- 12,460
I think with the low end card, you wouldn't have playable fps with everything set to max, hence lowering the settings would use less Vram. With the high end card, you could max out everything including AA and possibly not have enough Vram. I noticed GTA shows how much Vram will be used for each setting and as you say, AA needs a large portion.
I agree with Leeb. My understanding is (not claiming to speak definitively here) that a card is only ever working on one frame at a time (though they do store multiple completed frames in the buffer). That means the complexity of the frame (resolution, textures and post processing) is what determines the amount of memory required. Everything required for a frame is stored in VRAM while the frame is being processed. In that sense it doesn't really matter from a VRAM capacity point of view whether the GPU takes 20ms or 40ms to render a frame, it still requires sufficient VRAM to store the requisite data for the frame.
In that sense, while a 780ti may be able to process frames twice as quickly as a 760 at the same resolution and settings, each frame requires identical data to produce and thus would therefore (I assume) require the same amount of VRAM.
Obviously as you process frames faster, VRAM bandwidth (speed) becomes increasingly important as you're trying to load more data in and out of memory (you're producing twice the number of frames for example, so twice the data in and out), but I don't believe that capacity requirements change.
As Leeb points out, faster GPUs make it possible to achieve playable frame rates at higher detail settings, but it's the higher detail required from each frame which increases the demand for VRAM capacity, not the speed at which the frames are rendered.
In that sense, while a 780ti may be able to process frames twice as quickly as a 760 at the same resolution and settings, each frame requires identical data to produce and thus would therefore (I assume) require the same amount of VRAM.
Obviously as you process frames faster, VRAM bandwidth (speed) becomes increasingly important as you're trying to load more data in and out of memory (you're producing twice the number of frames for example, so twice the data in and out), but I don't believe that capacity requirements change.
As Leeb points out, faster GPUs make it possible to achieve playable frame rates at higher detail settings, but it's the higher detail required from each frame which increases the demand for VRAM capacity, not the speed at which the frames are rendered.
(I'm theorising here)... Both GPUs would spend the same amount of time waiting for the data to be copied to the VRAM from system RAM, so the bottleneck would be the same for both. Obviously once the data did arrive the faster card could process it more quickly, so no, they wouldn't be completely equally bogged down. However I think once you seriously run out of VRAM you're into the 5-15 frames per second range at best, irrespective of your GPU.
Going back to your original question RE modded Skyrim @ 3x1200p, I would strongly suggest that a single high end card would be better than two mid range cards in that scenario. Those texture packs in Skyrim are VRAM hogs as I understand, and I don't think the 2GB RAM on a 760 (if that's what you were considering) is going to cut it. The 780/780ti also have much higher memory bandwidth which becomes increasingly important at high resolutions. If you're set on SLI, that's one case where getting the 4GB 760s would be a better idea (I usually wouldn't recommend it, but 3x1200p for modded Skyrim is a different story!). But I still think a single 780 or 780ti would be the better choice personally.
TRENDING THREADS
-
News Firefox user loses 7,470 opened tabs saved over two years after they can’t restore browsing session- Started by Admin
- Replies: 19
-
-
-
-
Question HELP !!!!! My Desktop PC randomly sends monitors to standby???
- Started by EpicsGamer
- Replies: 7
-
T-Mobile Home Internet: Revisiting 5G connectivity for the home after two years
- Started by Admin
- Replies: 6
Facebook
Twitter
Instagram