Current crop of graphics cards compared, ranked by price
Although NVIDIA's pledged to simplify its lineup for consumers and ATI's been getting better, the current state of the graphics card market is still a pretty wild alphabet soup of model numbers and specs lists, so the crew over at The Tech Report decided to break things down using the only stat that matters: price. While the results aren't exactly shocking (surprise: more dollars equals more FPS), what's interesting is that multi-GPU rigs are really quite cost-effective, delivering performance on par with higher-end cards at significantly lower prices. For example, two Radeon HD 3850s run nearly as fast as a single Radeon HD 3870 X2, even though they cost a fair bit less, and two GeForce 9600 GTs can potentially outgun a GeForce 8800 Ultra. That's always been the promise of SLI and CrossFire, and it looks like it's paying off -- any system-builders out there care to share their experiences?



















Reader Comments (Page 1 of 2)
Thi mam(kris120890) @ May 9th 2008 6:34AM
That graph is genius and simplifies it brilliantly. hats off to the creater
Microswirl @ May 9th 2008 6:51AM
René Descartes FTW!
Chris @ May 9th 2008 11:40AM
if you actually go read the article, they left out certain cards for certain tests, thus rendering the overall average chart useless. good idea, poor implementation.
Eric @ May 9th 2008 1:14PM
The one big problem I have with this article is that with the exception of Crysis, all games are run at 2560x1600. I realize that they had to run at that resolution to show off the multi-GPU solutions, but really, what percentage of gamers run on a 30" display? 0.2%?
rock99rock @ May 9th 2008 2:39PM
I dont think its a matter of how many gamers run at that res, but how many SHOULD run at that res.
Ghen @ May 9th 2008 5:34PM
A realistic resolution would be better for testing purposes because some cards perform remarkably different at lower res with the same options. Usually its the difference between older high end cards versus newer low end cards.
tristan @ May 14th 2008 3:09PM
Not sure if anyone has broken down how much it costs per FPS on each card, but I did and here are the results I came up with:
Top 3 (30+ FPS)
GeForce 9600 GT at $4.85 per frame
GeForce 8800 GT at $5.59 per frame
GeForce 8800 GTS 512 at $6.71 per frame
Top 3 (40+ FPS)
Radeon HD 3850 CF at $6.60 per frame
GeForce 9800 GTX at $7.56 per frame
Radeon HD 3870 x2 at $7.96 per frame
Top 3 (50+ FPS)
GeForce 9600 GT SLI at $5.93 per frame
Radeon HD 3870 CF at $6.60 per frame
GeForce 8800 GT SLI at $6.91 per frame
These are all calculated with the individual cards listed in the test. If you took the single cards and doubled them up your gonna see quite a bit of difference in the results.
Andrew @ May 9th 2008 6:36AM
Interesting graph. In an ideal world the 'perfect' graphics card would sit in the top left hand corner ie. cost $0, framerate 90+
Its interesting to note that there appears to be a 'sweet' spot on the graph with a number of graphic cards centred near to the Nvidia 8800GT, thats obviously where the market is I would suspect.
Nvidia really do need to look at what they are doing. Multiple graphic cards using thousands of watts of power, churning out all that heat and costing an arm and a leg doesn't seem economical. How about doing an 'Intel' and bringing out some multiple core chips with lower Mhz but able to multithread stuff.
michas_pi @ May 9th 2008 6:40AM
The "perfect" graphics card you speak of already exists.
It's our eyes and visual cortex.
K @ Jun 4th 2008 10:58AM
Most softwares still aren't capable of utilizing multithreaded hardware so there's no need for that... yet
Twitchy @ May 9th 2008 7:25AM
"The "perfect" graphics card you speak of already exists.
It's our eyes and visual cortex."
I bet your "...eyes and visual cortex..." can't play Crysis though, can they? If so, will you let me tear the necessary parts out of your head and mount them on my PC?
Brian @ May 9th 2008 7:48AM
Twitchy: I don't know, Crysis looks pretty freakin' sweet when I close my eyes and pretend I'm running around throwing barrels at trucks and tearing down huts with grenades. I think that's got even the eyeball beat in terms of entertainment.
Ivan @ May 9th 2008 7:59AM
zing!
Andir3.0 @ May 9th 2008 8:28AM
Using the logic of the OP (which makes sense... the best bang for the buck is the card closest to the top left corner) the 9600GX2 seems to be the average winner in all this.
Andir3.0 @ May 9th 2008 8:32AM
*9800GX2 (grr, morning fingers...)
OneLove @ May 9th 2008 10:12AM
Spending $600+ for a graphics card is not winning.
Ruben @ May 9th 2008 10:35AM
"Nvidia really do need to look at what they are doing. Multiple graphic cards using thousands of watts of power, churning out all that heat and costing an arm and a leg doesn't seem economical."
Their cards (ATI's included) are already efficient at what they do. Thats what it takes to generate the graphics that are output by these graphics cards. You underestimate the power required to generate the graphics created by games like Crysis, COD4, etc. One can make the argument that Consoles can already output such graphics, but that person would be completely and utterly wrong. No console can output 2560x1600 (hell, none can output at 1920x1080) at a reasonable frame rate. But PC's can, and with quality and AA settings set to the maximum allowed. In fact, i would make the claim that CPU's waste way too much power for what they are capable of processing. Modern processors can compute 15 - 20 GFlops at usage ratings of 65 - 100 watts, meanwhile, graphics processors compute 250GFlops or more at 200 to 250 watts.
michael @ May 9th 2008 4:43PM
"The "perfect" graphics card you speak of already exists.
It's our eyes and visual cortex."
Well, ... uh ... no.
Graphic cards render or 'draw' the picture they don't 'see' the picture. The human equivalent would be your hand and a set of colored pencils or paint.
Now, you may be an artist, but for me it would be like one frame per hour. And the speed wouldn't really matter because the quality would be so bad you wouldn't be able to recognize the picture anyway.
DaveyG @ May 9th 2008 12:42PM
"In an ideal world the 'perfect' graphics card would sit in the top left hand corner ie. cost $0, framerate 90+"
Wow, you must be a genius. Did you figure that out all by yourself?
Carl Vitullo @ May 9th 2008 6:48AM
link plz?
it seems to be down, goes to 404.
OneLove @ May 9th 2008 10:13AM
you need a better graphics card, please pick one from the graph above.
Bill @ May 9th 2008 7:28AM
Though it is obvious that two cards are better than one with two chips in them it is also a bit misleading. I am saying that because a good SLI or CF able motherboard costs twice as much as one that does not offer dual PCI-16x slots.
And depending on the cards I would assume that two cards consume more power than one with two chips in them - depending of course they use the same chips.
All things considered though that 9600GT in SLI seems one great deal at the moment....
Ryan Singer @ May 9th 2008 7:37AM
HIGH 5!
THANKS FOR BREAKING IT DOWN THIS WAY:
FRAMES PER SECOND VS CHA-CHING!
Ihar `Philips` Filipau @ May 9th 2008 7:41AM
They should have also included power consumption.
Even my good ol' 420W power supply isn't sufficient (if dedicated) to power any of the high-priced high-fps cards.
I was pretty appalled when I have seen that 3870 X2 power consumption is 400W at load. Though priced competitively at $400, one would need extra power supply just for the baby. It's bit better on nVidia's side - but not much better.
To be perfectly honest, the current run for the best performance leaves many customers as victims in its wake. They both - ATI, nVidia - should slow-down a bit and stabilize/improve what they have. And cooperate more with game producers so that games would run on current generation of GPUs - not on next next one. Considering that majority uses middle-range cards, that would make life a lot easier for many consumers.
But then, since most of buying decisions based solely on vocal minority of gamers who have a buck to drop on every possible upgrade, GPU manufacturers have no choice as to make the run for the obscure performance figures.
Seems like market is in unrepairable state.
iofthestorm @ May 9th 2008 11:55AM
A 3870 single GPU should run on that PSU though, and if not probably a 3850.
Ghen @ May 9th 2008 5:40PM
420W isn't a gaming power supply, its that simple.
My wife needed a new power supply for her computer recently, she plays WoW and yet I still bought her a 600W supply. Why? Efficiency. Its not that the game will use the extra wattage but that a higher wattage supply will provide power more efficiently than one close to the limit required.
Ghen @ May 9th 2008 5:41PM
and no, its not crap brand 600 either. I know the difference.
John @ May 9th 2008 7:51AM
I just rebuilt my system this week, going from 2 cheaper SLI cards to 1 more expensive one.
I went from an EVGA 680i board w/ 2x 8800GTS-640 to an Asus X48 board (the DDR2 one) with a single 9800GX2-SSC. In giving up the capability to do nvidia SLI I've gained a MASSIVE amount of stability, especially utilizing all 8 gigs of memory under Vista-64, siginificantly increased RAID-0 performance for my dual Raptor system volume, and 2000 points higher in 3dmark (17949 under XP, 17137 under Vista).
So, while yeah... you can get a good cost-performance ratio out of a couple of SLI'ed cheaper cards, the choices in motherboard in order to do that may have other downsides besides cost (because my new Rampage Formula board wasn't cheap either). After having 2 680's fail, continued stability problems, and reading issues with the new generation of nvidia boards on the forums, there was no way that I'd go down that road again. Intel chipset FTW!
Kizorblade @ May 9th 2008 9:03AM
Well, the 9800GX2 is as the name implies, two graphics cards, so you still are using SLI I'm afraid
Zachary Pruckowski @ May 10th 2008 2:56AM
Yeah. I think the other issue here is cost of mother board. Any motherboard will have a PCIe 16x slot, but you may have to pay more to get an SLI motherboard. That's got to be factored into the price of an SLI solution.
moe @ May 9th 2008 7:52AM
if you take a closer look at what is in the tests and and what they are saying you will see flaws.. first before one can say the GT single or sli is the best bang you have to consider for the price where the cards sit on graph. the gts 512 out shines the GT in every graph but at a price that is competitive to the gtx and gx2. Second, they did not include the gts 512 in sli which would show the same results as above and would in fact beat the gx2 sli in every test making the best priced sli option.
the only thing that impressed me with these tests was the use of higher resolutions as most reviews cheat and use lower resolutions especially in 3dmark6. The GT is a good deal but as you said when shaders become more used in games the ones with the extra shaders will win and as sli gets better dual cards will also win. the gts 512 is the clear winner in these tests.
p.s. the prices used in these graphs are outdated.
wittepens @ May 9th 2008 7:52AM
mmm theres one left out
the 8800 gtx is still available for 200$
its almost as good as a 8800 ultra.
so for 600$ you can get close to the results of a tri-sli 8800 ultra system.
Majortom1981 @ May 9th 2008 11:59AM
My next card will be the nvidia 9600 the one with the display port. Unless another cheap card that is faster that has a display port comes out
Blastar @ May 9th 2008 8:18AM
I Wonder how low is my HD 2900 XT...
Alex Terry @ May 9th 2008 8:27AM
Is this going to be an ongoing comparison with updates on costs?
Be interesting to add the specific cards into the equation. BFG vs Asus etc.
Love the web, power to the consumer!
yacoub @ May 9th 2008 9:01AM
What a terrible chart. Also, anything "SLI" is immediately written off, as I (and most people) are only interested in what the single-card implementations have to offer. SLI and Crossfire are just for people with more money than sense.
Jomo @ May 9th 2008 9:05AM
While I'm sure the techreport guys know what they are doing, I'm not sure these results are totally accurate...
There are a few glaring errors that strike me with it:
- the difference between an 8800gt and an 8800gts 512 looks disproportionately large, considering a gts is just a gt with slightly higher clocks.
- elsewhere I've seen very few benchmarks that put ATI cards (3850/3870) way below a 9600 in performance. At worst, the 3870 is usually in between the 9600 and 8800gt.
- there are several tests ive seen that show even an 8800gts 512 to beat an 8800 ultra, and a 9800gtx (an overclocked 8800 gts) is even better, often outright beating the ultra.
All the same, its a nice graph, but i'm inclined to take this with a BIG pinch of salt, as this goes against a number of other reviews/roundups
sockatume @ May 9th 2008 9:48AM
That graph is much closer to linear than I would've expected. I'm used to thinking of the price-performance curve levelling off as soon as you hit $500, but this stays straight to around the $1000 mark. Excellent job, Tech Report!
OneLove @ May 9th 2008 10:19AM
I get 95 fps with my 8500 GT in cod4.
SweetSauce @ May 10th 2008 1:50PM
At 2560x1600 and on a 30" dislay?
If not, try comparing apples to apples.
huh @ May 9th 2008 10:21AM
It's a nice start, but really these sites should have been providing these kinds of stats for a long time. I should be able to find a card that fits my form factor, noise and heat requirements, performance and price easily, for example. Storage Review has been the leader in this area for at least a decade, with consistent reviews using comparable headings and most importantly collecting discrete comparable data, and long term user's observations. This serves everyone, but most sites would rather just type what they feel like for a particular review, without any consistency or data collection. When other sites started going to Flash based graphs I knew it was going sideways, this is at least smarter rather than "flashier." There are very nice browser based graphing packages these days so there's no excuse not to have any kind of stat you'd want comparing any device that's been reviewed. Review sites have been around for over 15 years, so it's about time when you consider what's happening elsewhere on the Web. I hope to see sharing data across sites for the next phase. :)
FrankieJoe @ May 9th 2008 10:43AM
Did they completely forget about Crossfire X?
FrankieJoe @ May 9th 2008 10:46AM
ah, nevermind. It appears that this picture is rigged.
dave @ May 9th 2008 10:55AM
I think the best bargain is the 8800 GT SLI setup (what I have). It benchmarks at just under 15k on 3Dmark05. An even better SLI setup which is just as good as the 9800 GTX, is a 8800 GTS 512mb SLI setup...it performs exactly the same for around 100 bux less. Too bad it's not on that chart...good stuff though.
JakeB @ May 9th 2008 11:20AM
Since the card I got (8800 GTS 512) is on there, I was disappointed to see it not in SLI form. Anyone got a link to those numbers? I dont want to spend the money if its not worth it.
dave @ May 9th 2008 11:42AM
Its worth it especially now since the prices on the 8800 gts are really low...$229 at Tiger. SLI two of them and you'll get the same performance as the 9800 GX2 and save about $150 too :)
JakeB @ May 9th 2008 2:09PM
I dont buy from Tiger, only Newegg. Seeing that it is the same price and I have a $20 gift card to use I think I will get it next paycheck
michael @ May 9th 2008 11:38AM
I have found better prices than they used. I suspect that spot bargains are available throughout the graph, and it doesn't really change the shape of the graph so it may not matter. However, for example, I recently saw the 9600GT for $129 instead of the $159 that they use.
Ridgecity @ May 9th 2008 1:10PM
just remember to turn off your fridge before turning your pc on, since you will probably blow a fuse, and an eye vein once you see the extra juice your quad card pc add to the electricity bill. I rather have A/C than 20 more fps...
DssTrainer @ May 9th 2008 1:22PM
Fairly certain 8800GT out performs the 8800GTS at stock speeds.