Sign in with
Sign up | Sign in
Your question
Solved

Can someone explain the math?

Last response: in Graphics & Displays
Share
February 6, 2013 7:19:53 PM

Greetings!

So instead of getting into a heated debate that eventually leads to fanboy fights and FPS charts and whatnot, I'm doing the research myself but not finding one kinda critical explanation - the direct correlation between the core clock and the memory interface. It seems a simple enough question, but every site I go to goes south or doesn't fully explain it.

I suppose the easiest way to phrase my question in a manner relevant to my study is does the faster core clock on a GPU bottleneck because of the memory interface and, if so, is there a reference guide out there that explains it? I ask because I'm having a hard time deciding if, for what I do, having a higher core clock with a smaller interface can equivocate (relatively) to having a higher interface with a lower core clock.

i.e. Is having 1GHz + 2GB 256-bit going to have a similar output to a 800MHz +3GB 384-bit. I understand having more memory on the card has an effect as well, but I'm not playing high-end games on ultra settings and watching the FPS counter. I'm playing stuff like Civ 5, SCII, CoH and I'm not trying to push the card to the limits. However, I do want to a) understand the math and b) want to get the best bang for the buck.

Thanks ahead of time!

S

More about : explain math

a b U Graphics card
February 6, 2013 7:27:32 PM

Apples and oranges.
Core clock comparison is meaningless between different chips.
7970 at 925MHz (stock) will blow 1,1GHz 650's head off and *** down its neck. (Yeah, that was Duke Nukem reference)

If you're comparing same GPUs, higher core clock will always result in better overall performance than higher VRAM clock.
m
0
l
February 6, 2013 7:35:28 PM

You need more VRAM for higher resolutions It doesn't increase performance directly. More than 2GB of VRAM is useless unless you are on multiple monitors. The bandwidth (e.t. 256/384-bit) Is important. Powerful graphic cards need more bandwidth otherwise they will get a bottle necked (wider pipes can push more water through :D  ). Clock speed is what really matters. But still you need to show us both cards in order to decide witch one is better. New generation cards have more cores and shaders and may use a little lower clock speeds.
m
0
l
Related resources
a b U Graphics card
February 6, 2013 8:15:56 PM

lyrius said:
You need more VRAM for higher resolutions It doesn't increase performance directly. More than 2GB of VRAM is useless unless you are on multiple monitors. The bandwidth (e.t. 256/384-bit) Is important. Powerful graphic cards need more bandwidth otherwise they will get a bottle necked (wider pipes can push more water through :D  ). Clock speed is what really matters. But still you need to show us both cards in order to decide witch one is better. New generation cards have more cores and shaders and may use a little lower clock speeds.

That is completely false, I have a 7950 and my 3gb of vram is being maxed by my modded skyrim. 110+ mods. One monitior that is 1080p.
m
0
l
February 6, 2013 11:02:24 PM

lyrius said:
You need more VRAM for higher resolutions It doesn't increase performance directly. More than 2GB of VRAM is useless unless you are on multiple monitors. The bandwidth (e.t. 256/384-bit) Is important. Powerful graphic cards need more bandwidth otherwise they will get a bottle necked (wider pipes can push more water through :D  ). Clock speed is what really matters. But still you need to show us both cards in order to decide witch one is better. New generation cards have more cores and shaders and may use a little lower clock speeds.



I kinda wanted to avoid the fanboy fight, but since that's impossible apparently, it's the difference between going for the 7870 Ghz or the 7950. I have no intention of running Skyrim, let alone with millions of mods. I'm not a "gamer", I just play a couple of less intense games and am wondering if jumping up to the 7950 is necessary if I'm not power gaming or if there is a drastic difference between the two.
m
0
l
February 6, 2013 11:47:29 PM

Quote:
More than 2GB of VRAM is useless unless you are on multiple monitors.


Nothing fanboy about it. This is just incorrect.

But the 7950 is a fantastic card for the games you play. you won't be disappointed.
m
0
l
February 6, 2013 11:59:19 PM

A base game that is unmodded usually doesn't need more than 2GB of VRAM. There is an exception for a few games but there aren't that many games that need it.

This excludes skyrim with a lot of mods.
m
0
l
a b U Graphics card
February 7, 2013 12:20:23 AM

seafire01 said:
A base game that is unmodded usually doesn't need more than 2GB of VRAM. There is an exception for a few games but there aren't that many games that need it.

This excludes skyrim with a lot of mods.

A notable game i have encountered un-modded is Farcry 3 hits the 2.1gb mark every once in awhile. But Seinnhai in all honesty if you are not going to play intense games i recommend something of the 2 gb 7770 nature.
m
0
l
a b U Graphics card
February 7, 2013 12:29:30 AM

To answer your original question the amount transistors a GPU has combined with the core clock speed determines the actual performance or speed of processing of the GPU. The memory bit interface is something entirely different and has to do with the maximum amount of information being able to travel from the CPU to the VRAM to the GPU. Think of a bit interface being the cap on a bottle of water, the larger the cap or opening the more water can flow in or out respectively.
m
0
l
a b U Graphics card
a b } Memory
February 7, 2013 1:02:05 AM

im not familiar with how the core clock and ram relates, but for memory bandwidth, it goes something like this (please correct me if im wrong)

busWidth x memFreq (x factor depending whether it is ddr3 or ddr5, etc)

but let's say they are both ddr3 then we can remove that factor. if we have a 128bit vs a 256.
to be equal (theoretical), the 128bit should have double the memory speed of the 256bit to be able to match it. like so: (assuming 1000mhz speed)

1000 x 256 =
1000 x 128 =

This is theoretical and performance will vary from architecture to architecture. you also have Rops, etc. but i don't know how those work, esp mathematically



m
0
l

Best solution

a b U Graphics card
February 7, 2013 1:09:15 AM
Share
a c 176 U Graphics card
a b } Memory
February 7, 2013 1:17:06 AM

Soda's answer is closest. You seem to be looking at only one part of video cards. The clock speed. I thought the "mega hertz myth" was busted a long time. Why only for CPUs? You can't look at just one (or two) specs on two very different parts and know which is faster. Or how it will compare to other cards. You need to consider the card as a whole. There is no formula you can use.
m
0
l
February 13, 2013 10:56:57 PM

Best answer selected by Seinnhai.
m
0
l
!