Not a single game use 2gb on 1080p, nowadays you can perfectly play with 1gb in this resolution (a 1gb gtx460 well oc can run almost all games near maximum setting), but 512mo is over. The best choice is a 3gb card only because they have a better bandwith, thats why 7970 are better than a 680.
With Dolphin I think the only things that will limit you willl be the software and the processor, I think 1gb card should enough to run 1080p.
There are a few pc games that use well over 1GB at max settings at 1920 x 1080.
Cards with more video ram don't necessarily have more video ram bandwidth. In your particular comparison the 7970 does indeed have more video memory bandwidth primarily due to its 384 bit buswidth (as compared to 256 bit).
Yeah games overpass 1gb but they are still perfectly playable on the most powerfull 1gb cards, I dont think you are so limited with 1gb, you will be limited by the power of the card. My gtx460 was running crysis 3 1080p almost max settings, max textures at 42fps, the 1gb are fast to saturate but you can still up your settings without losing fps, curious but true.
RAGE is a game which will fill any quantity of VRAM whatever settings it uses. It has almost 20GB of textures on your HDD, and loads the highest resolution versions which will fit into VRAM. If you don't have much, regardless of your settings, it will use small low res textures. If you keep the same settings, and switch from a 1GB card to a 2GB card, even if they're identical in any other way, and settings are left the same, you'll get shown higher resolution textures, and the game will look better.
Just one example of how quality of a game at what would appear to be max settings can be dependant on VRAM quantity.
Actually a lot of modern AAA games employ similar techniques. Executing different branches of code and loading different assets depending on the amount of vram detected to prevent framerate drops, stutters, and crashing on cards with less vram. Though I can't think of any that do it in a way that is as obvious as RAGE.
Quote:Yeah games overpass 1gb but they are still perfectly playable on the most powerfull 1gb cards, I dont think you are so limited with 1gb, you will be limited by the power of the card. My gtx460 was running crysis 3 1080p almost max settings, max textures at 42fps, the 1gb are fast to saturate but you can still up your settings without losing fps, curious but true.
You will get stuttering when you get above your gfx cards memory max. In some games this is very noticeable, in others not so much.
The interesting part is how many cards now have the upper limit of vram on lower bandwidth, like my GTX 660 and GTX660ti. The firs 1536MB is connected thru 192bit, where as the last 512MB is only at 64 bit. Not that it has much of an real world application, as most games that have settings that utilize more than 1.5GB will not be playable at 60 FPS on those cards. The last 512MB is still much faster than the DDR3 you have for RAM which would be the next step for texture storage.
cluthz Wrote:The firs 1536MB is connected thru 192bit, where as the last 512MB is only at 64 bit.
cluthz Wrote:The last 512MB is still much faster than the DDR3 you have for RAM which would be the next step for texture storage.
These two sentences don't make any sense. What do you mean by "the last 512MB"?
If you're saying what I think you're saying then you're wrong.
On a GTX 660 and 660 TI 2GB edition 1GB of memory is wired through a 64 bit bus. Then you have two other 512MB memory sets that are also wired through two separate 64 bit memory buses. You end up with 192 bit total buswidth (3 x 64 bit) and 2GB of total vram.
Source:
http://www.tomshardware.com/reviews/geforce-gtx-660-geforce-gtx-650-benchmark,3297-2.html
Like the GeForce GTX 660 Ti, the reference 660 includes 2 GB of memory. The only way Nvidia is able to achieve this using three 64-bit memory controllers is by mixing capacities. Also like the 660 Ti, the only way this amount of memory can be handled over a 192-bit bus is with mixed-density ICs. It works like this: the three 64-bit controllers divide the total memory into 512 MB chunks, which are accessed at the full 192 bits. The remaining 512 MB is addressed by just one 64-bit controller in a completely separate transaction.
Edit: Ah. I see your statement is copy/pasta from tomshardware. In that case I can blame them for wording it in a confusing manner and spouting things which are clearly wrong.
tomshardware Wrote:the three 64-bit controllers divide the total memory into 512 MB chunks, which are accessed at the full 192 bits
This makes it seem like the ICs are all wired to all three controllers. Which of course they are not. Each IC is only wired to one controller. So they are not "accessed at the full 192 bits". Each IC can transfer at most 64 bits at a time over the memory bus.
tomshardware Wrote:The only way Nvidia is able to achieve this using three 64-bit memory controllers is by mixing capacities.
This makes it sound like they are using mixed IC densities, which they aren't.
tomshardware Wrote:Also like the 660 Ti, the only way this amount of memory can be handled over a 192-bit bus is with mixed-density ICs.
Bullshit. Explain to me tomshardware how 8 doesn't go into 2048? Because I'm pretty sure the answer is 256. That's right. If you pop off the HSF and count there are 8 ram ICs total, 4 on each side. Each one is 256MB. You can confirm this by reading the info. on the chip. No mixed density ICs are used. And this is definitely not the only way this can be handled over a 192 bit memory bus. You don't have to wire the same number of ICs to each controller you know. Which nvidia didn't.
This is such an obvious mistake and so easy to verify that I'm amazed that not only did tomshardware make it but they never went back and fixed it. Which means that nobody called them out on it. They seem to be operating under the assumption that there are 6 ICs, which would indeed require mixed IC densities. But if they just opened the card and counted they would see that this isn't the case.
tomshardware Wrote:The remaining 512 MB is addressed by just one 64-bit controller in a completely separate transaction.
This makes it sound like it has its own memory controller. In reality it just shares one with another pair of 256MB ICs.
The wiring is really simple. There are 8 x 256MB ICs on the board, 2GB total. They are divided into 4 pairs each containing 2 x 256MB ICs for 512MB of total memory. The first two pairs are wired to their own 64 bit memory controller. The last two pairs are both wired to the last 64 bit memory controller. In other words the first two memory controller each have 2 ICs wired to them and the last memory controller has 4 ICs wired to it. No mixed density ICs are needed for this to work. Which is why nvidia didn't bother with it. There is also a 4 IC version with higher denstity ICs (512MB per IC) where only one chip is wired to each of the first two memory controllers and two chips are wired to the last memory controller. And there are likely 6 and 12 chip 3GB versions as well according to the docs.
I have now spent the last 15 minutes on google trying see if there was indeed a 6 chips version out there that was wired as tomshardware describes but I simply cannot find one. Every single PCB shot I could find of a GTX 660 or GTX 660 TI has had either 4 or 8 ICs. And I've looked at over a dozen designs now including the reference design which has 8! The 4 chis versions usually have all 4 chips on the top side (under the HSF) but some have 2 chips on each side. The 8 chip versions usually have 4 chips on each side but some have 6 chips on one side and 2 chips on the other side. I also checked my own GTX 660 and it has 8. Since tomshardware didn't provide any PCB shots from underneath the HSF I'm going to assume that they're wrong since everything I've been able to find from every other source including my own GTX 660 contradicts their claim.
This of course makes your last two posts completely incorrect. I certainly don't blame you though. Tomshardware is usually a reputable source and you were just following them. Which is why I'm so surprised that they would get something so obvious so wrong.
I'm not sure what to believe. You sound like you know well what you are talking about, but both tomshardware and anandtech claims otherwise.
Anandtech:
http://www.anandtech.com/show/6159/the-geforce-gtx-660-ti-review/2
The best case scenario is always going to be that the entire 192bit bus is in use by interleaving a memory operation across all 3 controllers, giving the card 144GB/sec of memory bandwidth (192bit * 6GHz / 8). But that can only be done at up to 1.5GB of memory; the final 512MB of memory is attached to a single memory controller. This invokes the worst case scenario, where only 1 64-bit memory controller is in use and thereby reducing memory bandwidth to a much more modest 48GB/sec.