2 gfx cards performance in charts?

[Fourstaff]

This question is specifically targeted at W1zzard, since that he is the one who does all of the graphics card benching, but others are welcome to voice their displeasure/support.

I was wondering, since that sometimes 2 mid end graphics card offers a lot more bang for buck compared to an expensive single card solution, can you include them in the comparison chart too? I understand that there is more a lot more work to do but I believe its useful to have them included. There are special SLi and Crossfire reviews going around, but those are (relatively) few and far between, and the chart is not always up to date (for example, comparing GTX580 and 6870CF). If its more work for not enough benefits, I completely understand as its not too difficult or time consuming for us to open 2 browser tabs and compare the two graphs. Thanks for reading.

Edit: if this matter have been discussed before, then this thread serves no purpose. I did a quick sweep around the comments and feedback section but came out nothing.

 

[El_Mayo]

I agree

 

[Inioch]

I have also lately been wondering how especially the new cards would fare against say 460 SLI or 6850/6870 CF. So +1 to this.

 

[qubit]

2 GPUs hardly ever make sense

I've noticed that when reviewers recommend two cheaper cards that together cost less than one expensive high end card and outperform it, that they forget about the cost of an SLI or Crossfire motherboard - these things are expensive and actually make the dual GPU more expensive overall, so this should be factored into the recommendation.

Of course, all the usual issues of dual GPU running, such as scaling, noise, heat and power usage should be factored in too.

When you put all these variables into the melting pot and give it a really good stir, you find that the single GPU solution is usually best. About the only time where a dual GPU solution seems to make sense to me, is where some uber enthusiast with very deep pockets runs two top of the range cards together, with maxed out components for the rest of the system to let them achieve their full potential.

This system will obviously give you the best frame rates, regardless of all the other factors, which makes sense in order to achieve the best possible performance to show off to your not so rich friends.

 

[Fourstaff]

Take for example: 890G-UD3 and 880G-UD3. Less than £20 difference. If you look at the markets right now, a GTX 480 costs about £330, while a 6850 costs about £150. They both perform about the same. If you don't have the money to afford a GTX480 right now (or similar), then you can get a 6850 and then another 6850 later and that will give you as much performance. I know there is a lot of other factors like power input, scaling, noise etc, but its still something which people should take a good look at.

 

[Inioch]

These days CF or SLI (8x+8x) doesn't really cost all that much more. It's built in many of todays middle range boards and chipsets. And power requirements are also not that much bigger, often equal or sometimes even smaller.
Power supplies used to run top cards often have the necessary connections to run a multi-card system too.

 

[qubit]

That is a small difference between the mobo prices, I'll give ya that - and that does help the dual GPU argument.

Thing is, your spending basically the same money, but ending up with two GPUs with all the downsides. Even if you wait to get the second GPU a while later when it's cheaper, this still doesn't work: you'll have spent the extra money on the better mobo up front and the high end card you're comparing to will have dropped in price as well, making the price difference very small and potentially a net negative. Not to mention that product ranges move on, your card may be discontinued or they may be a really great performing card came out since at a good price, which ruins the value if your upgrade completely.

Therefore, I maintain that a dual GPU card makes the most sense when you max out your system like in my example. Or perhaps getting a dual GPU card such as a 5970 or GTX 295 can make sense. These cost more, but not usually double more, you don't have to get a more pricey mobo and you get the performance boost now.

I reckon if scaling could have been 100% in all applications, then going the dual card route would have made much more sense. Unfortunately, the technology never even gets close to this.

 

[n-ster]

If you have an X58 platform, the mobo cost point is VERY moot

I to would JUMP IN JOY to see CF and SLI cards in the charts

 

[Fourstaff]

Well, if you cannot afford a high end graphics card right now, what are you going to use between building your new shiny system and getting a high end graphics card, onboard perhaps? High end goods may be cheaper in the future, but there is always the second hand market. We have just transitioned into DX11, and that might be a bit of a damper to the argument, but (for the sake of example) if you currently have a 5850 That's not going to push enough pixels for, say, Metro 2033 (don't bash the game please), you have 2 options: get another 5850 or sell it and get a GTX 580 or GTX480 etc. If you cannot sell the 5850, it makes perfect sense to Crossfire your 5850 with another 5850 if your mobo can support it, rather than leave the 5850 to dust and buy a spanking new top-of-the-range graphics card. Back when the 5850 was released, its probably considered a bit high end(ish), so you most likely would have gotten a nice crossfire-capable board too (again, just for the sake of example).

 

[qubit]

The affordability argument over time has some legs. But then, just to be awkward, I could suggest that one gets a small loan to buy the high end stuff first anyway?! lol But yeah, that would have it's own pitfalls and makes a grey area even greyer, so I'm pullin' ya leg.

So as a way to boost performance without splashing out in one go, I guess it works, but still with caveats in my mind, the scaling issue being the biggest one and with both AMD & nvidia equally.

You make an honest point about the features argument such as DX11. However, those basic feature enhancements don't come that often and then take ages to be implemented fully, so don't really have to be worried about so much, but not disregarded either.

Just look at the HD2900 from May 2007. It was very advanced in that it had a tesselator and we now know just how awesome that feature is! But it was never used and now that implementation is out of date anyway and won't work with DX11, so I doubt a single HD2900 owner ever used it. :shadedshu

Despite my negative view of dual card setups as a practical purchase, they are still interesting from a technical point of view, so I have no objection to your idea of including them in reviews in the way you suggest.

Oh and on the subject of Metro 2033...

 

[EarthDog]

The difference between a SLI/Crossfire board vs non is almost negligable in the overall picture. The point to me is this....

If I buy 2 cards I want perfect, or as close to it scaling ALL the time. SLI/Crossfire can scale from nothing to 100% with an average it seems of around 70%. So explain to me why would I pay 2x if If not getting 2x performance 90%+ of time? Not to mention the cost difference in power.

Take a typical res of 1920x1080. A 470 or 5850 can charge through that res with every game any most any AA you want on most any game. Those are $250. Why would I get 2 460's to run it faster with more AA that I cant notice after 8x really anyway?

Just my .02.

EDIT: That said, it should still be included in the tests.

 

[Completely Bonkers]

I think ONE or TWO benchmark SLI/CF cards would be a useful addition. But I dont want the tables riddled with 2x cards.

What I find more important in the tables are the following:

1./ A "legacy" card or two, that would be indicative of what people are upgrading FROM. I care less about the difference between a 5750 and a 5770 as I want to know about how much better these things are over a 3870 (e.g.) A single card from a previous generation (ie cards 18-36 months old) would make a good BENCHMARK. I think most people here are UPGRADING cards, so it is directly relevant to us
2./ Often there is missing data in the stats. Case in point: look at the comparative tables at the end of the review, you will see GTX460 in most but not all of the tables. Q. How noisy is the GTX460 compared to others... data not there.

 

[n-ster]

the popular SLI/CFX setups only ofc

 

[theubersmurf]

Providing it's not inordinately much work (and clearly doing all the benching for each review is substantial...unless he's using older benches at times) I'd like to set this a bit myself.

 

[Oxford]

Performance per watt and noise with SLI/CrossFire are the issues I'm mainly interested in. Pricing can fluctuate quite a bit, depending upon slickdeals-type specials.

Also, increased input lag -- some say -- can be an issue with dual GPU setups. Is that the case?

Just look at the HD2900 from May 2007. It was very advanced in that it had a tesselator and we now know just how awesome that feature is! But it was never used and now that implementation is out of date anyway and won't work with DX11, so I doubt a single HD2900 owner ever used it.

Microsoft may be largely to blame for that. Couldn't DirectX 10 have been patched to support the tessellator?

 

[qubit]

Microsoft may be largely to blame for that. Couldn't DirectX 10 have been patched to support the tessellator?

I'd read that it was actually nvidia holding back killer features from the DX10 spec, because they weren't ready to implement it at the time. Apparently, today's DX11 is what the original DX10 should have been.

However, I think it was Charlie Demrjian of SemiAccurate who said this, so I'll take it with a pinch of salt.

 

[FreedomEclipse]

FIRST OFF --

you could probably go to TOMSHARDWARE for CF/SLI/Single card comparisons - they have pretty decent charts - but of course TOMSHARDWARE has fallen from grace with the tech community since the site was bought by someone and to simply put it - their the clown of the tech community and whatever you hear from their site should be taken with a pinch of salt.

I personally find their graphics charts to be pretty decent.

Providing it's not inordinately much work (and clearly doing all the benching for each review is substantial...unless he's using older benches at times) I'd like to set this a bit myself.

Its a stupid amount of work testing graphics cards especially when you have to move from the ATi 4xxx series through to the 6xxx (4850, 4870, 4980, 4770, 5770, 5830, 5850,5870, 5970, 6850, 6870, 6950, 6970) - Just ATi cards ALONE would probably take you upto 1day to test them all if you average 45mins-1hr to test each setup. even if you exclude the 4xxx series - theres still a lot of work and thats NOT including Nvidia cards.

I agree that it would be a total kick ass addition/feature to the site and no doubt it would bring it a lot of traffic - but its very time consuming unless W1zzard can send the cards to someone to bench them for him, otherwise I dont think he has much time to do it all himself