Does Memory Bus Width Double in SLI mode?

[thisisrmm]

I'm doing a system build for a friend and he is going to get 2x GTX 760s, but he's very unsure about whether to get the 2GB version or the 4GB version. He's asked me for some assistance on this decision.

I know that 4GB can be somewhat overkill, since most games don't go over 2GB usage. I also know that the Memory Bus of the GTX 760 (256-bit) is too weak to actually even use the 4GB.

But after days of research on the differences, I came across a comment that said having the 2x 4GB versions actually doubles the Memory Bus (256 to 512-bit); this makes it so that the 4GB can actually be utilized.

Is this true?

 

[sam_p_lay]

Nope - each GPU has access only to its own memory architecture. It's honestly not that important though - all that counts is the end result (frames per second). There's been this stupid obsession with memory bus width ever since Igor's stupid comments in his GTX660 Ti review. Fact is, bus width is means to an ends (the ends being bandwidth - divide bus width by eight (bits to bytes) and multiply by effective memory clock frequency to get bandwidth in MB/s). Bandwidth is in turn another means to an ends (framerates). Since nobody is capable of replacing/upgrading the memory architecture on a card they buy, it makes exactly zero sense to single it out as a performance factor any more important than cores, ROPs, fill rate or anything else.

Think about it - how does the 192-bit GTX660 beat the 256-bit Radeon 7850? Hell even the GTX660 Ti is 192-bit and that beats the hell out of the 7850 despite a narrower bus width. My old 8800GTX from 2006 had a 384-bit bus width (look it up if you don't believe me) and that would get absolutely trampled today even by a basic GTX650. Since it used DDR3 and not GDDR5, data transfers per clock cycle were halved, so half effective bandwidth. Still just a means to ends.

 

[thisisrmm]

Nope - each GPU has access only to its own memory architecture. It's honestly not that important though - all that counts is the end result (frames per second). There's been this stupid obsession with memory bus width ever since Igor's stupid comments in his GTX660 Ti review. Fact is, bus width is means to an ends (the ends being bandwidth - divide bus width by eight (bits to bytes) and multiply by effective memory clock frequency to get bandwidth in MB/s). Bandwidth is in turn another means to an ends (framerates). Since nobody is capable of replacing/upgrading the memory architecture on a card they buy, it makes exactly zero sense to single it out as a performance factor any more important than cores, ROPs, fill rate or anything else.

Think about it - how does the 192-bit GTX660 beat the 256-bit Radeon 7850? Hell even the GTX660 Ti is 192-bit and that beats the hell out of the 7850 despite a narrower bus width. My old 8800GTX from 2006 had a 384-bit bus width (look it up if you don't believe me) and that would get absolutely trampled today even by a basic GTX650. Since it used DDR3 and not GDDR5, data transfers per clock cycle were halved, so half effective bandwidth. Still just a means to ends.

So you're saying that over Bus Width and what not, Memory Clock Speed reigns supreme? And which should my friend get? The 2GB or the 4GB?

 

[sam_p_lay]

No mate - they're two parts of a whole. The formula:

Clock freq (MHz) * (bus width (bits) / 8) * transfers per clock cycle (4 for GDDR5) = bandwidth (MB/s)

But the more important thing I said is that it doesn't even matter 🙂

The theoretical side of technology is interesting but utterly useless to the consumer. Real-world performance testing is both interesting AND actually useful! After all, it's not like you can buy a GTX660 and replace its memory interface with a 384-bit one! Take the component as a whole, since that's the only way it can be used.

And I'd definitely take 4GB models provided they're not too much more expensive. If you hit a RAM wall you'll simply have to drop settings (texture resolution and anti-aliasing will be most helpful). But obviously it's preferable to not have to drop settings.

 

[thisisrmm]

No mate - they're two parts of a whole. The formula:

Clock freq (MHz) * (bus width (bits) / 8) * transfers per clock cycle (4 for GDDR5) = bandwidth (MB/s)

But the more important thing I said is that it doesn't even matter 🙂

The theoretical side of technology is interesting but utterly useless to the consumer. Real-world performance testing is both interesting AND actually useful! After all, it's not like you can buy a GTX660 and replace its memory interface with a 384-bit one! Take the component as a whole, since that's the only way it can be used.

And I'd definitely take 4GB models provided they're not too much more expensive. If you hit a RAM wall you'll simply have to drop settings (texture resolution and anti-aliasing will be most helpful). But obviously it's preferable to not have to drop settings.

Thanks for the help sam_p_lay; I understand it much better now. I'm definitely going to let my friend know to pick up the 4GB versions then.

 

[sam_p_lay]

Glad I could help 🙂 Good luck with the build.

 

[thisisrmm]

Glad I could help 🙂 Good luck with the build.

sam_p_lay,

Do you think that 2GB would suffice for gaming in the next few years? Especially with new consoles coming out, my friend says he's worried that 2GB won't be "enough."

 

[sam_p_lay]

Honestly I'd feel happier with more than 2GB, though I'd only actually get a 4GB model is price isn't too much higher. If it's costing well over 50 more for the extra gig then I wouldn't bother. I think it's highly likely that within a couple of years (and almost certainly within three) that having more than 2GB will be beneficial at 1080p. Not currently the case though.

Like I say though, when that much memory does become necessary for max settings, I'd just drop texture res a little. Not sure about the memory consumption of SMAA (anybody?) but I'd imagine that's a memory-efficient technique for effective anti-aliasing too.

 

[thisisrmm]

Honestly I'd feel happier with more than 2GB, though I'd only actually get a 4GB model is price isn't too much higher. If it's costing well over 50 more for the extra gig then I wouldn't bother. I think it's highly likely that within a couple of years (and almost certainly within three) that having more than 2GB will be beneficial at 1080p. Not currently the case though.

Like I say though, when that much memory does become necessary for max settings, I'd just drop texture res a little. Not sure about the memory consumption of SMAA (anybody?) but I'd imagine that's a memory-efficient technique for effective anti-aliasing too.

I agree; with the coming of new consoles there will most definitely be the coming of more graphic-intensive games, so it's almost certain more VRAM will be needed.

He's really struggling in which GTX 700 series card to pick:

- 2x GTX 760 SLI (beats 770 and 780 in FPS, but more power, heat, noise, etc.)
- 1x GTX 770 (middle of the line)
- 1x GTX 780 (horrible price to performance: $650? really?)

Suggestions?

 

[sam_p_lay]

Absolutely agree about the GTX780. Prices on them will drop when AMD's new lineup launches and early adopters will regret not being patient! I'd honestly take a single GTX760 unless he absolutely has to max Crysis 3 and Metro. The GTX770 is far better price/performance than GTX780, but not great compared to the GTX760. You're looking at only a little over 20% framerate gain for the price difference. I'm really not sure that's worth it. Anyway, definitely not GTX780 🙂

 

[Williamcg1979]

I know this is an old topic, but I'm trying to decide between a gtx 780 and a 960 or 970 for fallout 4. Prior to reading the posts about the memory bus size, I had been concerned with exactly that. I guess my question is, would a 960 or 970 be sufficient if it calls for a 780 under "recommended"?

 

[sam_p_lay]

Impossible to say until the game gets benchmarked but I'd be very surprised if a GTX970 won't max it at 1080p.