Eyefinity: Two mid-range cards in x-fire, or single high-end card?

eipwithnumbers

Honorable
Feb 4, 2013
2
0
10,510
Hello everybody,

I will soon set up Eyefinity with three screens, so I'll have to be pumping out 5760x1080. My predicament is which graphics option I should go with:

I already have a 7850 2GB OC edition (this one, which is discontinued: http://www.newegg.com/Product/Product.aspx?Item=N82E16814102984)

I could get another one that seems to be identical (this one, only difference (I think) is cooler/port placement: http://www.newegg.com/Product/Product.aspx?Item=N82E16814102999)

The problem is, my motherboard has two pci-express slots, but only one of them is 3.0, and the other 2.0 one only runs at x4. And it might even cripple the speeds more in crossfire. (http://www.biostar-usa.com/app/en-us/mb/introduction.php?S_ID=579#spec)

So, should I try to sell the 7850 I have now and get a single 7970? Or go ahead with the second 7850 and try to sell and replace the motherboard? Or would the bandwidth issue with the pci-e slots not even be noticeable? I've heard that two cards are better than one when dealing with eyefinity, but I don't know if that's true.

Thanks!
 


Yes, always go with one gpu vs 2 when feaseable. CF, not all games
benefit from CF, more heat generated, more PSU power needed,
more microstuttering issues, more complicated upgrade path,etc.

Tom
 
Actually, there would not really be more heat generated and power consumed by 7850 CF compared to a similarly performing single GPU card, almost all modern games work excellently in CF with GCN GPUs, and stutter is a mostly rare issue nowadays, especially since there are now software tools to fix it in CF.

The PCIe issue is not a cause for alarm. PCIe 2.0 x4 is still generally not much of a bottle-neck for modern graphics cards in gaming.

The two card option here would mean more memory bandwidth and ROP throughput across both cards than any current single GPU card, so in theory, it could be a little better for Eyefinity, but a single GPU solution such as a Radeon 7970 or an overclocked Radeon 7950 would probably be fine and more easily upgraded as well. It'd probably be easier overall to just go for a single GPU solution.
 



Nitpicking a little - 2 7850's(req 575w) do use more power than a 7950(500w req)
or 7970(550w req), and if they are using more power than they will generate more heat.
It's not a huge diff, but there is a diff.
 


As far as i know you only get the memory bandwidth of one of the cards - and that being the lowest of the two. The vram is used to store the same exact data on both cards in sli/crossfire so adding more cards will not give you more memory bandwidth... From what i understand.

As for going crossfire or not. I would recommend always going with a single solution - unless a single card cannot meet the requirements you need, and in that case you should use 2 higher end mid cards at the least(for most cases).

In your case you already have a 7850 - buying a second one will most likely be the most beneficial price/performance wise. Unless of course you can sell your 7850 for $150+ if so i would sell it and get a 7970.
 


You have confused memory bandwidth for memory capacity. The capacity is equal to that of the lowest capacity graphics card in the array, but memory bandwidth works a little differently. Performance-wise, memory bandwidth almsot doubles with two cards- almost. That would be more effective bandwidth (technically, bandwidth isn't doubling, it's just that you have two GPUs each with a 256 bit memory interface instead of a single GPU with a 256 or 384 bit memory interface of the higher end single GPU cards), which is good for high resolutions, high quality settings, and high levels of MSAA.
 


That is wrong. AMD's recommended power supply wattage has nothing to do with real-world power consumption and is a guideline for people who don't understand how the power consumption and power supplies actually related to one another. Two Radeon 7850s will use only a little more power on average than a Radeon 7970, approximately comparable power consumption to a Radeon 7970 GHz Edition. That's only about 20-30 watts more load power consumption than the 7970 for a proper power supply for the situation.
 


Yes i understand that the capacity is equal to that of the lowest... But you have double bandwidth that is used to fill double the vram (2gb from each card even if they hold the exact same data the bandwidth to fill each is still used)... so that extra bandwidth is being used to fill BOTH cards vram. Effectively giving you the bandwidth equal to a single card.
 


Perhaps you misunderstood, so I'll clarify. The Radeon 7850 has a 256 bit wide memory interface at around 1.2GHz to 1.4GHz, depending on the model. Let's assume 1.2GHz for the math. That's about 154GB/s. The Radeon 7970 and Radeon 7950 both have a 384 bit bus which is 50% wider than a 256 bit bus and they have a fairly similar, if somewhat higher, memory frequency giving them at best around 60-70% more more memory bandwidth than a Radeon 7850. Two 7850s have two GPUs and each GPU has its own memory interface. Although each GPU's memory must store almost exactly identical memory, each GPU does not necessarily use the same data from that memory pool- they only use at a given time what data is necessary for their frame. Each GPU renders a frame on its own with current Crossfire and SLI technologies, so from a raw performance perspective, the memory accesses of each GPU are separate too.

Basically, from a raw performance perspective, you can treat the GPUs as a single GPU because how many GPUs you have doesn't change what FPS is, it only changes how you get it. In doing so, their aggregate memory bandwidth is effectively double, just like GPU count, compared to a single Radeon 7850. Technically, bandwidth is not doubled the same way technically, GDDR5's quad data rate doesn't quadruple the frequency of the memory, but performance treats it as quadruple for GDDR5 just as it treats memory bandwidth as double for Crossfire and SLI.

Do you understand what I'm saying? This is also proven by doing memory bandwidth tests on Crossfire and SLI systems: the benchmarks effectively have double the memory bandwidth to work with just like they have effectively double the GPU performance to work with. Games work the same way, granted they don't scale quite perfectly.

EDIT: I'm still not saying that I don't recommend a single GPU solution; I was simply trying to be informative of the options. A single GPU solution such as an overclocked Radeon 7950 or a Radeon 7970 would probably be easier to work with in the short term and easier to upgrade/sell in the long term.
 


I didn't use AMD's website,I do know how psu's and real world use are related.
I used realhardtechX recommended requirement, I then
did the math(watts & amps from the sites that list tdp for those 3 cards,
and 2 7850's use more power than either one 7950 or one 7970, plain and simple.
As I said above it's not a huge diff, but a diff nonetheless.
 

eipwithnumbers

Honorable
Feb 4, 2013
2
0
10,510
Thank you all for your replies, they are very informative (kind of), but I still would like to know:

Which configuration would be better for high resolutions like eyefinity: two 7850's, or a single 7970?

I know that a single card would be 'less hassle' and 'more compatible', but I'm willing to work through the problems of CF and find solutions to them when they come up. The only thing I'm concerned about, really, is which setup can handle eyefinity better.
 

ikaz

Distinguished
Well this probably won't help too much but you have to remember a 7970 also comes with 1 gig of memory at 1080p that really doesn't mean much in the real world but @5760x1080 you could be getting close to the 2gig video card (some test show skyrim can use 1.7 gigs with mods). I would go with a single 7970 if money not an issue if so then the 7850. Also if you are Oc'ing your current card you could have issues if you try to do that in xfire (with two different models doesn't help).
 


TDP is useless in determining power consumption and a power supply's total wattage doesn't even necessarily relate to how much power it can deliver to graphics and CPUs. No offense intended, but any site using such information is mistaken. Even if you don't use AMD's info and go off another, the answer will generally be wrong as anything more than an extremely general guideline for someone who doesn't know how to do the math properly. I'm not trying to mock your intelligence and knowledge by saying that; I'm just stating facts.

I know from experience how much power the cards consume. I'll start off by repeating that TDP and power consumption are not necessarily related. TDP has more to do with maximum heat generation IIRC, but I know that it is not power consumption because that is easily proven. For example, the GTX 580 consumes significantly more power than the Radeon 6970, yet the Radeon 6970 has a higher TDP (compare 244W TDP to 250W TDP and the GTX 580, despite having a lower TDP, still consumes dozens of watts more power at load). To make matters worse, how TDP is determined varies between companies and how it related to power consumption can vary between different products even from the same product line of a given company. So, TDP is pretty much useless for us consumers.

The only way to truly get accurate power consumption numbers is to measure the power consumption of the cards in different workloads. Two Radeon 7850s, in real-world tests, perform around a Radeon 7970 GHz Edition (a little better than a 7970) while consuming similar power to the Radeon 7970 GHz Edition (a little more than the 7970, but only slightly and very far from that 75W number your numbers describe). In fact, it's often more power efficient to CF two lower end or mid-ranged cards of a given generation than using a single higher end model of that generation.

For example, two GTX 660 cards are more power efficient than a single GTX 680 while also outperforming it. Radeon 7850s are only comparably power efficient to a Radeon 7970 or 7970 GHz Edition, but they're still not generating more heat and due to the superior cooling (two coolers instead of one since the heat is spread across two cards), unless the cards are right on top of each other, temps are oftentimes better.

Still, I don't change my recommendation from a single GPU solution. They're simpler, take less space, and easier to upgrade and often easier to sell when replaced.
 


You won't be anywhere near a memory capacity bottle-neck with 2GB for two 7850s in 5760x1080. Games often use more memory than is necessary when they have spare memory, but that doesn't mean that they're anywhere near a bottle-neck. There is no situation except when you have two or more GTX 670 or 680 cards in SLI where 5760x1080 starts to run into memory capacity bottle-necks even in the most memory-intensive games and even then, it's rare.
 

It is a little confusing what you are saying... But i think what you mean is the 2gb from the 670 and 680 can be a bottleneck because those cards can function at playable frames with the settings that would use over 2 gb of ram (making the ram a bottleneck) where as the 7850 won't, making the 2 gb from the 7850 a non issue. Is that correct? just trying to clear it up for myself and anyone else reading it. If so I agree for the most part, there will always be situations that this won't apply though.
 


Correct. It takes about two or more GTX 670/680 cards in SLI for 2GB to occasionally have problems. Yes, 2GB on two Radeon 7850 cards will not be a problem because multiple 7850s, unlike multiple higher end cards such as the 670 and 680, are not fast enough to play games in resolutions and settings that are bottle-necked by 2GB.

So, for two Radeon 7850s, 2GB is plenty of memory. 1GB would be a bottle-neck very often, but 2GB is enough to never be a bottle-neck in gaming for two 7850s. The 3GB on the 79xx cards is only important in extremely high end setups that cost several thousand dollars.