another SLI question

[SuperDuper]

I have a 1 GB GTS250 and was looking to go SLI. Would SLI work if the 2nd GTS250 was only 512MB? What if I used a 9800GT?

 

[vengence]

It would "work" with the GTS250 512, but it would be a bad idea if you are running 1080p or better.

 

[SuperDuper]

Thanks

 

[WildHawg]

Can I mix and match graphics cards that have different GPUs?
No. For example, an XXXGT cannot be paired with a XXXGTX in an SLI configuration.

Can I mix and match graphics cards with different sizes of memory?
When purchasing a second graphics card, you should try to match the memory size so that you are ensured full value and performance from your purchase. However, while it is not recommended, NVIDIA does offer the flexibility to run graphics cards with different sized memory by using CoolBits. Using CoolBits (value set to 18), you can force both of the cards to use the lower of the two memory sizes and operate them together in SLI mode. When dissimilar memory sizes are enabled to work together using CoolBits, the effective memory size for each card becomes the smaller of the memory sizes.

 

[SuperDuper]

Thanks

 

[obobski]

It would "work" with the GTS250 512, but it would be a bad idea if you are running 1080p or better.

I'm not so sure on this, I'm not saying you're wrong, I'm just not sure if it'd really kill performance, got a benchmark or something to read? (I'm saying this because I don't know if having 1GB actually helps the GTS 250 that much at high res, or if the GPU itself becomes the issue, if the GPU is more at fault, SLI will help, if its the memory, yeah you're right)

Can I mix and match graphics cards that have different GPUs?
No. For example, an XXXGT cannot be paired with a XXXGTX in an SLI configuration.

Can I mix and match graphics cards with different sizes of memory?
When purchasing a second graphics card, you should try to match the memory size so that you are ensured full value and performance from your purchase. However, while it is not recommended, NVIDIA does offer the flexibility to run graphics cards with different sized memory by using CoolBits. Using CoolBits (value set to 18), you can force both of the cards to use the lower of the two memory sizes and operate them together in SLI mode. When dissimilar memory sizes are enabled to work together using CoolBits, the effective memory size for each card becomes the smaller of the memory sizes.

exactly what I was going to post, for reference, this is from nVidia's website

 

[vengence]

I'm not so sure on this, I'm not saying you're wrong, I'm just not sure if it'd really kill performance, got a benchmark or something to read? (I'm saying this because I don't know if having 1GB actually helps the GTS 250 that much at high res, or if the GPU itself becomes the issue, if the GPU is more at fault, SLI will help, if its the memory, yeah you're right)

A single GTS 250 perhaps not. I'm pretty sure a pair of GTS 250s in SLI is as powerful as a single 4870 (or very near it) which has been shown in plenty of benchmarks to need the 1GB at higher resolutions.

 

[obobski]

A single GTS 250 perhaps not. I'm pretty sure a pair of GTS 250s in SLI is as powerful as a single 4870 (or very near it) which has been shown in plenty of benchmarks to need the 1GB at higher resolutions.

but you're comparing apples to oranges here, if the GTS 250 doesn't benefit from 1GB, a pair of them won't, because they don't just "combine" into a monolith, they're an SMP system, and since about 2001, no mainstream GPU vendor has been crazy enough to try and push cache coherent memory archs on their SMP solutions, because it eats up too much memory

basically if one card can't do it, two very likely can't as well

furthermore, where are you getting that "1GB is required" for the 4870?
http://www.tomshardware.com/reviews/graphics-ram-4870,2428-5.html

512MB vs 1GB vs 2GB, makes a very minimal difference in most benchmarks (basically no improvement for the money, especially the 2GB, excepting GTA4, which...c'mon)

I'm really doubting the need for a pair of 1GB boards, unless CoolBits/ForceWare has an issue with blocking off a segment of memory (as in, if the software doesn't work as cleanly as nVidia promises)

either way, comparing the 4870's to GTS 250 is really unrealistic, because you're talking about a single monolith vs an SMP setup, and a few other factors (like nVidia vs ATi) that make the comparison fairly illogical

if we compare 9800GTX+ 512MB to GTS 250 1GB (close enough for me):
http://www.legitreviews.com/article/913/5/

we see VERY little difference, and this small difference will be exaggerated when you're talking about two of these cards, again, I'm doubting the need for a pair of 1GB boards, at least from a performance perspective

also, a pair of GTS 250's in SLI should be easily faster than a 4870

 

[kumquatsrus]

in 95% of games 1gb won't make a difference over 512mb with a 250, but a few will. far cry 2 is one of them depending on the settings:

http://www.anandtech.com/video/showdoc.aspx?i=3523&p=6

@ 1680x1050 dx10 4xAA, the 1gb card was about 50% faster than the 512mb card. with no AA they would probably perform the same though.

so basically i would go 512mb unless the price differential was small enough to justify going 1gb (like $10 extra bucks).

and yes 2x gts250 > 1x 4870 most of the time.

 

[vengence]

furthermore, where are you getting that "1GB is required" for the 4870?
http://www.tomshardware.com/reviews/graphics-ram-4870,2428-5.html

I dunno, maybe you should try reading [H]ardocp's reviews if you are going to post on ther forums.

 

[obobski]

first: we aren't even talking about the 4870, so its performance is irrelevant

second: [H]'s data supports my assertion that the difference is minimal at best, the only thing that "improved" was the minimum measured FPS, although average and max are more or less on top of each other (and their graphs are more or less the same line)

third: kumquatsrus' post supports the [H] data (meaning that [H] agrees with 3rd party sources), in other words, we're talking about one of the rare instances that the extra memory makes a slight difference, and it isn't "plenty of benchmarks proving it needs 1GB" (as most of the benchmarks provided in this thread show quite the opposite)

so I said before, the difference is negligble, but even if it were to matter, you're still talking about an AMD card (and said card's SINGLE processor performance), and not the nVidia board (actually PAIR) in question, so the point is moot

 

[SuperDuper]

Thanks for all of the replies
I've yet another question. Should I notice any input lag with a 2nd card?

 

[vengence]

wow, what a warm and polite attitude you've taken with me

I don't bother with pleasentries when people are spreading fud.

first: we aren't even talking about the 4870, so its performance is irrelevant

Yeah memory is magically different between a 250 and a 4870.

second: [H]'s data supports my assertion that the difference is minimal at best, the only thing that "improved" was the minimum measured FPS, although average and max are more or less on top of each other (and their graphs are more or less the same line)

ummm try reading, 27.6 vs 21.6. That's an average of 28% more fps. That's minimal? I want some of what ever it is your smoking.

third: kumquatsrus' post supports the [H] data (meaning that [H] agrees with 3rd party sources), in other words, we're talking about one of the rare instances that the extra memory makes a slight difference, and it isn't "plenty of benchmarks proving it needs 1GB" (as most of the benchmarks provided in this thread show quite the opposite)

Slight? he was talking about 50% increase in fps? What the hell does it have to be to be a big difference to you?

so I said before, the difference is negligble, but even if it were to matter, you're still talking about an AMD card (and said card's SINGLE processor performance), and not the nVidia board (actually PAIR) in question, so the point is moot

The point isn't moot because the pair of nvidia cards is near the shadder power of a single 4870. Your complete failure to understand any of this is why I didn't bother to be polite. You would just assume cripple his preformance in a couple of games to save 5 or 10$. I will not do that to him.

 

[obobski]

Thanks for all of the replies
I've yet another question. Should I notice any input lag with a 2nd card?

honestly not sure on input lag, I'm doubting it very much though, theres a few threads regarding microstutter/input lag with SMP setups, both sides are pretty die-hard to their cause

what I can tell you is, most people I know with SLI and CF aren't complaining, so I wouldn't be too worried there

snip

reasons you can't be making this 4870 vs GTS 250 comparison:
- different drivers, GPU, memory controller, memory type, memory bus, hardware, etc
- you're trying to compare 2 GTS 250's across the board to a single 4870, this isn't generally possible (you're comparing a monolith to an SMP and arguing architectural/implementation specifics, you can compare the end result performance, but not the "path taken" to get there)


its like saying because apples have a lot of sugar, if you eat a ton of them, you'll know what oranges taste like
or because a Vortec is an engine, you can use its performance data to know exactly how a VTEC is going to perform

theres too many variables being changed to make the comparison

as far as the GTS 250 benchmarks, the performance gains may very well be 25-50%, however if this %delta isn't the difference between playable and unplayable, so if you start at 15 FPS and go to 20 FPS, so what if its gained performance?

furthermore, the gains from SLI are going to be fairly big either way, as I said some posts back, unless the CoolBits trick really cripples things (i.e: the drivers hate it), it shouldn't be an issue either way, although (again as I said earlier), if price isn't a huge difference, it doesn't matter either way what you do (and from your post, it would seem the difference is like what? $5? why even argue the technology at that point, just provide that info up front, if its only $5, who cares...)