CrossFire/SLI Memory

[jamie0226]

I heard that in Crossfire or SLI if there were 2GB cards and there are two of them overall memory will still remaining 2GB cause all other cards memory will mirror the main card. If this is true how come in multi-screen its better off with 2 or more cards, will there be any benefits from memory perspective of 2 or more card? Exclude GPU perspective.

Thanks

 

[MadMarc]

No you won't get any benefits in terms of memory, since each card has to load the same things into its memory and so the usable space remains 2 GB. The extra performance comes just from the added gpu power.

 

[KING_OF_SAND]

Yes. When you are using 2 2gb cards you have 4GB of memory but only 2GB of usable memory. You can use as many screens as you want, as until you hit your max vRAM limit then you just get some big performance hits.

 

[Rastalovich]

What big performance hits are these ?

 

[Master&Puppet]

What big performance hits are these ?

Well there are some good examples where I believe a larger vRAM has given a card an advantage over it's competitors.

If you compare something like a 7950 to a 580, which are pretty evenly matched gpu wise (but the 7950 has twice the vRAM), you can see the 7950 pull away in certain situations:

http://www.techpower...HD_7950/21.html

If you take a look at that review you can see that in the most modern games with (which are the most likely to use more textures) the 7950 chases down/pulls away from the 580 at higher resolutions (my educated guess is that the vRAM is allowing this).

But that only applies to games which are able to fill up the 580 - metro, bf3, shogun 2,Skyrim, dirt 3 etc... in less demanding/older games the relationship between the cards stays the same i.e. in the original crysis the 7950 is always 2-4 fps ahead regardless of the resolution.

 

[Rastalovich]

If you take a look at that review you can see that in the most modern games with (which are the most likely to use more textures) the 7950 chases down/pulls away from the 580 at higher resolutions (my educated guess is that the vRAM is allowing this).

But that only applies to games which are able to fill up the 580 - metro, bf3, shogun 2,Skyrim, dirt 3 etc... in less demanding/older games the relationship between the cards stays the same i.e. in the original crysis the 7950 is always 2-4 fps ahead regardless of the resolution.

We can't accept that without the same reviews showing the vram being filled up. We can assume that the memory is filled up, but if we look at other examples within the same review, the likes of metro trade off 5-6 frames on a 2560 display against a 480 with half the memory. Half the memory, plus noticing the card is generically slower, for 5-6 frames on a 2560 display doesn't read big performance hits or on the other hand that more than 1.5g of memory is possibly being used - which we can't establish as it's not recorded.

EDIT: only 1 test will put this thing to bed once and for all. A 1.5g (or 2) version of a card, versus a 3g (or 4) version. Loads of titles, same settings, 3 or so resolutions. One of which is a multi-screen. Same clocks, same windows install, same drivers, usual recordings plus vram/cpu/mem usage. I would personally put cash on there not being a 'big performance hit' without many prior facts.

 

[Master&Puppet]

Well I've also found this - it's a little bit dated but I'll try and dig up some more. I never said huge performance hit by the way - just under increased textural loads it can help! I would like to test it myself but I don't have the cards available...

http://www.tomshardw...870,2428-5.html

I'd assume that a multiple monitor setup (i.e. a higher resolution) would increase the advantage. That's doesn't mean that you need a lot of vRAM for 1080p though. 1.5-2gb looks plenty to me at the moment even though all that testing you've got in your thread suggests that some games are able to fill up more it's not necessarily an advantage to fps.

 

[Rastalovich]

Mad isn't it. There's one result on the 1680x1050 where the 512 card goes faster on average, all the way up to their highest resolution where, although there is minimal average drop (saying minimal, 2-3 frames and we're talking about the crysis-hog here) from as small as a 512m card to a 2G card. I would have expect a bigger swing on that myself.

 

[Master&Puppet]

Yea - especially at the lower resolutions the scores are so close (sometimes within 0.5 fps) that even tiny uncontrollable little changes, like a single background thread deciding to do a tiny bit more work, changes the place of the cards. I consider them all equal unless there's more than a 5% difference in fps over several tests. But that review does suggest there was an advantage - but that's the only review I've found on it so far...

 

[006409]

remember windows is clever so system memory can be alicated to the graphics card.

To view how much memory has been alicated to your graphics card in windows vista, 7 & 8;

start >

right click on 'computer' >

properties >

select 'performance information and tools' in the bottom left hand corner >

then select ' view and print detailed performance and system information' on the right hand side >

Then under graphics it will say 'total available graphics memory' of which under it splits between whats shared system memory and graphics memory.

hope that helps.

 

[Rastalovich]

I think, with todays hardware, the way Windows allows all modern setups to use 4g standard as vRam, in whatever ratio - obviously it'd be preferable to have it mostly on the gfxcard for that much of an edge, a lot of games make use of this.

I also think to a lesser extent, a 775/ddr2/ddr3 higher end system can cope just about as well. Perhaps as the spec goes down it impacts more and more. I'm sure as an excessive example, DDR1 @266 would be a nightmare. DDR2 @ 533+.. ok not bad. DDR3.. possibly 667+ is hardly noticeable.

I'm not totally convinced that all game engines use this tho. Maybe Dx9 doesn't allow it, maybe it does, I don't know. Maybe if you code the game correctly, and not make assumptions as is a common bad practice, it'll never work with the engine.

Massive study would sort of all this out.

 

[jamie0226]

my problem is solved thanks to all you guys, thanks

 

[KING_OF_SAND]

What big performance hits are these ?

Have you played BF3? When you run out of vRAM you know it fast. You go from a steady 60FPS to a jumpy 40 - 60 FPS. Crysis 2 you go from 60FPS to 30FPS and occasionally 40FPS.

 

[Rastalovich]

I play everything.

Whether I continue to play everything depends on the game(s).

 

[Master&Puppet]

Yea you still don't need more than 2gb vRAM for any game at 1080p and I think that gives you medium term future proofing too.

 

[Rastalovich]

1080 seems to be the most popular resolution these days with the mass production of these monitors.

Whilst I prime a sole computer for games playing and nothing else, I do keep in mind that not everyone can do this and at the same time, if someone is getting performance hits, like KOS does, you can't dismiss the fact that they get them. What the factors are you can't comment on without getting your grubby mits on their computer.

It is alarming for a minority selection of games that vram usage does seem to spiral out of control on some setups - whilst on others you see different results.

I looked at this guy earlier http://www.youtube.com/watch?v=NTlgOs2EI-M hitting some massive fps with a "new" 680, and the 670 4g were coming out and people asked him if it were better to get that than his 680 - so you'd deduce he's on a 2G 680. Sure he's on 1920x1080 and cuts out blurring etc, but even so, he goes around 1450 vram and much large fps than I get, somewhere from 50 to 100.

And yet KOS goes from 60 to 30 - which is half the performance. To me there must be a reason for this. Mods ? i dunno, I play the game but I'm not into any 'scene' with it.

If people get results - they get them.