![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
Graphics memory, a primer?
- Thread starter Ravien
- Start date
Oops. My fault for thinking an answer is simple. 
The entire question is specifically for gaming. My budget is undetermined, and I'm not exactly sure what card I'm looking for, just mainly looking for a card on a good deal (I'm nearly an internet styled version of an extreme couponer, lurking until good deals, then snatching it up)
It's mostly for gaming, trying to push the farthest with the least for the longest amount of time.
I mean, at what point is the difference so miniscule it doesn't matter?
Like, 128->256, I can imagine could be a huge difference. I could be wrong. I would assume not a big difference between 3XXbit or whatever is is and 448 though, because at that point you're so wide open it doesn't matter. I could be wrong though. Which is why I ask this question.
I assume memory size plays alot into the bit rate as well, as 1gb at 128 is probably fine, but 2gb at 256 is much better than 2 at 128, but 1 at 256 is overkill. I don't know. But I want to know. And I'm not an engineer, so my brain can't handle detailed explanations of how GPUs work that every place puts out about new cards. God I hate those things.
The entire question is specifically for gaming. My budget is undetermined, and I'm not exactly sure what card I'm looking for, just mainly looking for a card on a good deal (I'm nearly an internet styled version of an extreme couponer, lurking until good deals, then snatching it up)
It's mostly for gaming, trying to push the farthest with the least for the longest amount of time.
I mean, at what point is the difference so miniscule it doesn't matter?
Like, 128->256, I can imagine could be a huge difference. I could be wrong. I would assume not a big difference between 3XXbit or whatever is is and 448 though, because at that point you're so wide open it doesn't matter. I could be wrong though. Which is why I ask this question.
I assume memory size plays alot into the bit rate as well, as 1gb at 128 is probably fine, but 2gb at 256 is much better than 2 at 128, but 1 at 256 is overkill. I don't know. But I want to know. And I'm not an engineer, so my brain can't handle detailed explanations of how GPUs work that every place puts out about new cards. God I hate those things.
If you're just looking to learn for the purpose of buying a gaming card, not worth worrying about VRAM specs. We could lecture all day on it and it wouldn't help your buying decisions. As pointed out:
Yep, there are too many factors at play to compare cards on VRAM specs alone. You're better off buying based on reviews, where real-world performance is tested.
You guys I think the OP is just asking for some rules of thumb.
Memory clock speed and bus width are the two variables that determine memory bandwidth. Memory bandwidth is a better overall indicator of performance than just clock speed or bus width alone. Let me give you a simple example.
Now that I told you about memory bandwidth, let me throw a wrench into the gears and say that you can NOT compare memory bandwidth across radically different architectures. I'll give you an example of this too. Please look at this performance summary at TechPowerUp. The Radeon HD 7970 has 264 GB of memory bandwidth, while the GTX 670 "only" has 192 GB of memory bandwidth. But in real life the two cards are tied in performance until you get up to very high resolutions, then the 7970 gets ahead by about 10%.
But the most important thing you should know has already been posted by someone else: real world performance in reviews is more important than theoretical performance.
Memory clock speed and bus width are the two variables that determine memory bandwidth. Memory bandwidth is a better overall indicator of performance than just clock speed or bus width alone. Let me give you a simple example.
- An older video card, the nVidia 9600GT, has a 256 bit bus. A newer graphic card, the GTX 460 768MB, has a 192 bit bus. At first, one would assume that the older 9600GT has better performance.
- The older 9600GT has its memory clocked at 1800 MHz. The newer GTX 460 768MB has its memory clocked at 3.6 GHz. At first, one would assume that the GTX 460 768MB has better performance.
- The reality is that the 9600GT has 57 GB per second of memory bandwidth, and the GTX 460 768MB has 86 GB per second of memory bandwidth. The GTX 460 is much faster in terms of memory.
Now that I told you about memory bandwidth, let me throw a wrench into the gears and say that you can NOT compare memory bandwidth across radically different architectures. I'll give you an example of this too. Please look at this performance summary at TechPowerUp. The Radeon HD 7970 has 264 GB of memory bandwidth, while the GTX 670 "only" has 192 GB of memory bandwidth. But in real life the two cards are tied in performance until you get up to very high resolutions, then the 7970 gets ahead by about 10%.
But the most important thing you should know has already been posted by someone else: real world performance in reviews is more important than theoretical performance.
The rare times the difference in memory is relevant are when you're comparing OEM versions of cards (usually the mobile versions), which occasionally have different memory speeds/types/bus widths than the retail models.
The other (rare) time it was relevant was the GTX460 which came in versions with different memory sizes, and since the memory controller was linked to the GPU's shaders, turning off a small faction of memory also turned off those shader units (making the card slower). But that's one single instance.
Other than those, as stated already, it's pointless to compare memory bus widths, types, and speeds because you won't find the same card configured two different ways. You can only compare the whole package, unless you have a purely academic interest in it.
The other (rare) time it was relevant was the GTX460 which came in versions with different memory sizes, and since the memory controller was linked to the GPU's shaders, turning off a small faction of memory also turned off those shader units (making the card slower). But that's one single instance.
Other than those, as stated already, it's pointless to compare memory bus widths, types, and speeds because you won't find the same card configured two different ways. You can only compare the whole package, unless you have a purely academic interest in it.
FTFY
Legitimately the only difference between the 660Ti and the 670 is the memory bus. And the 670 is 20% faster so... 192bit ->256bit = 20% performance.
However, as others have said different architectures handle memory differently.
Well, for the most part, I'm looking to compare inside of the same architecture. I'm pretty sure which card I want, but I'm wondering if juicing out for that extra gig of memory, or 64 more bits of bandwidth will really matter that much, from one 6870 to another (is just an example, dont get clinincal on me mentioning that specific card - I dont have my wishlist handy at work, so I don't remember what card it is I'm bearing down on)
Like for example, if I had a GTX 670 w/:
1gb @ 192bit,
And two more w/:
2gb @ 192bit,
1gb @ 256bit,
which would be better, and why? In addition, if I could find the best one,
2gb @ 256bit, how much better than either of them would it be, and if it is any better, would it be worth paying the premium?
Like for example, if I had a GTX 670 w/:
1gb @ 192bit,
And two more w/:
2gb @ 192bit,
1gb @ 256bit,
which would be better, and why? In addition, if I could find the best one,
2gb @ 256bit, how much better than either of them would it be, and if it is any better, would it be worth paying the premium?
grandfatman
Limp Gawd
- Joined
- Sep 6, 2012
- Messages
- 188
what "difference" are you looking for?
memory speed, FPS,games, design?
5 is always better than 3, right?
If you read the excellent review on the new GTX 660Ti, the memory bus size is not all that significant.
Don't they go up to 11?
(Wait, that Spinal Tap reference MIGHT not work here! ) lolThe 2gb @ 256 bit is best because, as we know, that particular card improves when you give it more memory bandwidth. But the comparison is purely academic because that is the only configuration of that card that exists or will ever exist (excluding the 3rd party "double VRAM" models that are for running multi monitor setups and has no benefit for normal use).
Trepidati0n
[H]F Junkie
- Joined
- Oct 26, 2004
- Messages
- 9,269
Your quesiton does not compute. The 670 doesn't give you choices of bus widths and neither does the 660Ti. You get one or the other.
Your quesiton boils down to two parts.
1. How much performace do you want? (660 Ti vs 670)
2. What level of textures/AA are you running (2 vs 3GB or 2 vs 4GB)
For the most part the questions are mutually exclusive. If you run out of memory, your performance will drop to shit regardless of how much raw GPU power you have. Unless you are focusing on a specific game with specific AA/texture features (e.g. Skyrim) the industry standards are sufficient (there are a ton of [H] articles on this already for both single and SLI). Therefore it boils down to #1...how much performance do you want. Memory size, for the most part, is a red-herring, under the nominal case.
Alright then, lets peg this at a specific model then.
http://www.newegg.com/Product/Produ...0006662&isNodeId=1&Description=560+ti&x=0&y=0
560 ti. I can get it in 1gb 256bit, 2gb 256bit, or 1gb 320bit. Which is better? Is 2gb much better than 1gb for gaming at 1080p?
http://www.newegg.com/Product/Produ...0006662&isNodeId=1&Description=560+ti&x=0&y=0
560 ti. I can get it in 1gb 256bit, 2gb 256bit, or 1gb 320bit. Which is better? Is 2gb much better than 1gb for gaming at 1080p?
The cards you are quoting are not the same. Nvidia's atrocious naming scheme is to blame for that. The full name of that last 560 Ti is the "560 Ti with 448 Cores", a nonsensical title which indicates that it is in fact a cut-down GTX 570, a card that uses the GF110 GPU rather than the GF114 that the 560 uses, and as a GF110 card it has a 320-bit memory bus.
So the 560-Ti-448 is the best card, but because it's an intentionally mislabeled GTX 570 (you can tell by the number of shaders it has), not necessarily because of its memory bus width.
edit: And for the record, I have heard that the 560-Ti-448 is a really good deal for its price.
only this matters
http://lmgtfy.com/?q=gpu+benchmarks
btw instead of 560ti 448 you can get 7870 or 7850, same or cheaper depending where you buy them from and they beat that card at stock, consume less and can overclock for way more
7870 is only about 8% better than the 7850 at the same clocks, and I calculated this in 3dmark (compared performance clock to clock to the benchmarks in a review of 7870 XFX i think), real world performance would most likely be lower, almost equal, and they both overclock the same. 8% means something like 3fps at 50-60fps
http://www.guru3d.com/article/geforce-gtx-560-ti-448-core-review/22
http://www.3dmark.com/3dm11/4354358
stock clocks of 900mhz gives me around p5500-p5700 so they're basically equal, in some games that favour nvidia the 560 will have slightly more fps. I've noticed this in Portal 2 for example.
So what you're saying is that performance-wise, these three cards are pretty much identical, but the raedons are current gen and the 580 is last gen, and they all cost about the same, right?