• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

128-bit or 256-bit cards?

S

shipinabottle

Member
My old nVvidia card was 256-bit memory interface. Most of the ones nowadays are 128-bit. So, is there a huge difference between the two?

Researching online, it seems that there's no major difference.
 
I am not all that sure as I havnt kept up to date with all the new stuff out. But the ones that are 128bit if you look, should be DDR5 memory I believe, compared to 256bit which is DDR3. I believe DDR5 is still faster(would make sense right? Or else why use it heh). All the new cards these days are DDR5. All the older cards were DDR3 and the cards before that were DDR2.
 
256-bit memory interfaces are faster, and modern high-end cards still use 256 bit or greater interfaces. Mid to low-end cards use 128 bit interfaces. They're slower than the modern ones with 256-bit, but as fast as the older 256 bit cards because all new cards now use faster GDDR5 memory while older cards used slower GDDR3 memory.

As Schadenfroh said, this all factors into the card's total memory bandwidth, and that should be the main number to consider for a graphics card's memory speed.
 
Schadenfroh said:
Look at memory bandwidth instead of just the bus width, there are other factors to consider.
Click to expand...

bandwidth is indeed bits * clock... and indeed is more indicative of performance than looking at just how many bits yours video card has. But it is still a flawed way to look at it. It is still looking at an individual engineering decision and saying "AHA! This card must be faster because engineering decision number 263 on it is higher then on competing card".

The correct way is to look at benchmarks, which take everything into account concurrently by looking at actual performance rather then individual engineering decisions.

http://www.anandtech.com/bench/GPU11/188
 
That is why I never really pay much attention to or invest much effort into specifics on technology items. Just look at benchmarks, that is all that matters. Performance in real world scenarios is all I care about. From a strict logical point of view, the fact all the newer vid cards have (at least?) 128bit DDR5 memory instead of 256 DDR3 would seem to dictate that DDR5 128bit is "better". 😛
 
It's more cost and power efficient to use 128-bit wide buses in combination with GDDR5 memory than 256-bit with GDDR3, so that's why manufacturers made the shift. It gives comparable memory bandwidth stock and higher bandwidth if the memory is OCed, so it's more flexible as well.

The two factors that you take into account when measuring memory bandwidth are the effective memory frequency (eg. 4000MHz) and the bus width (eg. 128-bit).

To answer your question directly: your old card probably used slow GDDR3 memory with a 256-bit bus width, so memory bandwidth should actually be higher on something like a GTS 450.
 
Fallengod said:
From a strict logical point of view, the fact all the newer vid cards have (at least?) 128bit DDR5 memory instead of 256 DDR3 would seem to dictate that DDR5 128bit is "better". 😛
Click to expand...

actually no it doesn't, they are about equal in performance. It is a matter of costs.

Remember that bandwidth (total memory speed) is bits * mhz. GDDR5 has higher mhz, getting more "bits" on the memory controller means making the memory controller larger, which means making your GPU die larger; a 128bit controller transfers 128bit per clock cycle (measured in mhz), and a 256bit controller transfers 256bit each clock cycle. Nvidia for a time went with with a larger GPU die for the 256bit memory controller and coupled it with a GDDR3 (slower) ram and got the same overall bandwidth as AMD's GDDR5 + 128bit controller (Smaller GPU die). nVidia paid more on the GPU die fab, but could use GDDR3 which was cheap and widely available while AMD was stuck with the rare and expensive GDDR5 which limited their supplies.
Since then prices of GDDR5 went down and now its cheaper to use a smaller controller (smaller GPU die) with GDDR5.

Now, if you were to use 256bit GDDR5, that would be faster. (twice as fast as 128bit GDDR5)
 
Last edited:
LOL_Wut_Axel said:
It's more cost and power efficient to use 128-bit wide buses in combination with GDDR5 memory than 256-bit with GDDR3, so that's why manufacturers made the shift. It gives comparable memory bandwidth stock and higher bandwidth if the memory is OCed, so it's more flexible as well.

The two factors that you take into account when measuring memory bandwidth are the effective memory frequency (eg. 4000MHz) and the bus width (eg. 128-bit).

To answer your question directly: your old card probably used slow GDDR3 memory with a 256-bit bus width, so memory bandwidth should actually be higher on something like a GTS 450.
Click to expand...

Not exactly. Like taltamir said, a 256 bit memory bus width has the same advantage over 128 bit as GDDR5 memory has over GDDR3: both effectively double memory bandwidth. A graphics chip with 256 bit GDDR3 memory will have the exact same bandwidth as one with 128 bit GDDR5 memory clocked at the same memory frequency. Any advantages of 128 bit GDDR5 memory comes from the ability to push the clock frequency higher. If it's the same frequency it won't have more bandwidth.
 
Red Hawk said:
Not exactly. Like taltamir said, a 256 bit memory bus width has the same advantage over 128 bit as GDDR5 memory has over GDDR3: both effectively double memory bandwidth. A graphics chip with 256 bit GDDR3 memory will have the exact same bandwidth as one with 128 bit GDDR5 memory clocked at the same memory frequency. Any advantages of 128 bit GDDR5 memory comes from the ability to push the clock frequency higher. If it's the same frequency it won't have more bandwidth.
Click to expand...

Um, that's what I said. I don't know what you're arguing about.
 
nVidia also has 192-bit memory interfaces so they can cripple certain cards so they perform in certain price brackets.
 
Red Hawk said:
You implied that newer graphics card with 128 bit GDDR5 memory should have more bandwidth than one with 256 bit GDDR3 memory. That isn't necessarily the case.
Click to expand...

No, I didn't. I said that if it's an old card it'll have slower GDDR3, so yes, a newer card like a GTS 450 should have more bandwidth. Higher-specced GDDR3 runs at around 2000MHz effective, and then you could make that argument. I'm talking of GDDR3 around 1600MHz, which is what older chips ran at.
 
I'm not sure if this completely counts as a car analogy, but I know how we like them in VC&G, so here goes. 🙂

Think of the memory speed as a speed limit on a road and 64, 128, 192, 256, etc. bit, as how many lanes wide the road is. You can move the same amount of traffic on a two lane road with a 60mph speed limit as you you can on a four lane road with a 30mph speed limit.

Likewise 128 bit 4Gbps memory will have the same bandwidth as 256 bit 2Gbps memory.

The only reason you see one vs. the other is cost.
 
Memory buss by itself is meaningless. Same with bandwidth (which is buss memory x ram speed / 4, I think).

Just look at benchmarks, you'll find the best answer there.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top