Graphics Chips Compared Plus Test Configurations
We ran the following graphics cards on a PC with an Intel Core i7-920 CPU overclocked to 3.8 GHz (its default clock rate is 2.66 GHz), minimizing the possibility of a processor-based bottleneck.
| Nvidia Graphics Cards | ||||||
|---|---|---|---|---|---|---|
Vendor and GPU | Internal Reference | RAM | GPU Clock | Shader | Memory Clock | SPs |
GeForce GTX 295 | 2 x GT200b | 2 x 896 MB GDDR3 | 576 MHz | 4.0, 1,242 MHz | 2 x 999 MHz | 2 x 240 |
GeForce GTX 285 | GT200b | 1,024 MB GDDR3 | 648 MHz | 4.0, 1,476 MHz | 2 x 1,242 MHz | 240 |
GeForce GTX 280 | GT200 | 1,024 MB GDDR3 | 602 MHz | 4.0, 1,296 MHz | 2 x 1,107 MHz | 240 |
GeForce GTX 275 | GT200b | 896 GDDR3 | 633 MHz | 4.0, 1,404 MHz | 2 x 1,134 MHz | 240 |
Zotac GeForce GTX 260² (GTX 260 216 SPs) | GT200b | 896 GDDR3 | 576 MHz | 4.0, 1,242 MHz | 2 x 999 MHz | 216 |
GeForce GTX 260 216SPs | GT200b | 896 GDDR3 | 576 MHz | 4.0, 1,242 MHz | 2 x 999 MHz | 216 |
GeForce GTX 260 | GT200 | 896 GDDR3 | 576 MHz | 4.0, 1,242 MHz | 2 x 999 MHz | 192 |
Zotac GTS 250 1 GB (GTS 250) | G92b | 1,024 MB GDDR3 | 740 MHz | 4.0, 1,836 MHz | 2 x 1,100 MHz | 128 |
GeForce GTS 250 | G92b | 1,024 MB GDDR3 | 740 MHz | 4.0, 1,836 MHz | 2 x 1,100 MHz | 128 |
GeForce 9800 GX2 | 2 x G92 | 2 x 512 MB GDDR3 | 600 MHz | 4.0, 1,500 MHz | 2 x 1,000 MHz | 2 x 128 |
GeForce 9800 GTX+ | G92b | 512 MB GDDR3 | 738 MHz | 4.0, 1,836 MHz | 2 x 1100 MHz | 128 |
GeForce 9800 GTX | G92 | 512 MB GDDR3 | 675 MHz | 4.0, 1,688 MHz | 2 x 1,100 MHz | 128 |
GeForce 9600 GT | G94 | 1,024 MB GDDR3 | 650 MHz | 4.0, 1,625 MHz | 2 x 900 MHz | 64 |
GeForce 9600 GT | G94 | 512 MB GDDR3 | 650 MHz | 4.0, 1,625 MHz | 2 x 900 MHz | 64 |
GeForce 8800 GTS 512 | G92 | 512 MB GDDR3 | 650 MHz | 4.0, 1,625 MHz | 2 x 972 MHz | 128 |
GeForce 8800 GT | G92 | 1,024 MB GDDR3 | 600 MHz | 4.0, 1,500 MHz | 2 x 900 MHz | 112 |
GeForce 8800 GT | G92 | 512 MB GDDR3 | 600 MHz | 4.0, 1,500 MHz | 2 x 900 MHz | 112 |
GeForce 8800 Ultra | G80 | 768 MB GDDR3 | 612 MHz | 4.0, 1,512 MHz | 2 x 1,080 MHz | 128 |
GeForce 8800 GTX | G80 | 768 MB GDDR3 | 576 MHz | 4.0, 1,350 MHz | 2 x 900 MHz | 128 |
GeForce 8800 GTS | G80 | 640 MB GDDR3 | 513 MHz | 4.0, 1,188 MHz | 2 x 792 MHz | 96 |
GeForce 8800 GTS | G80 | 320 MB GDDR3 | 513 MHz | 4.0, 1,188 MHz | 2 x 792 MHz | 96 |
| ATI Graphics Cards | ||||||
|---|---|---|---|---|---|---|
Vendor and GPU | Codename | RAM | GPU Clock | Shader | Memory Clock | SPs |
Radeon HD 4890 | R790 | 1,024 MB GDDR5 | 850 MHz | 4.1 | 4 x 975 MHz | 800 |
Radeon HD 4870 X2 | R700 (2 x RV770) | 2 x 1,024 MB GDDR5 | 750 MHz | 4.1 | 4 x 900 MHz | 2 x 800 |
HIS H487QT1GP ICEQ4+ (HD 4870) | RV770 | 1,024 MB GDDR5 | 770 MHz | 4.1 | 4 x 1000 MHz | 800 |
Sapphire Vapor-X HD4870 2G (HD 4870) | RV770 | 2,048 MB GDDR5 | 750 MHz | 4.1 | 4 x 900 MHz | 800 |
Radeon HD 4870 | RV770 | 512 MB GDDR5 | 750 MHz | 4.1 | 4 x 900 MHz | 800 |
Sapphire HD4850 1G (HD 4850) | RV770 | 1,024 MB GDDR3 | 625 MHz | 4.1 | 2 x 993 MHz | 800 |
Radeon HD 4850 | RV770 | 512 MB GDDR3 | 625 MHz | 4.1 | 2 x 993 MHz | 800 |
Radeon HD 4770 | RV740 | 512 MB GDDR5 | 750 MHz | 4.1 | 4 x 800 MHz | 640 |
Radeon HD 4670 | RV730 | 512 MB GDDR3 | 750 MHz | 4.1 | 2 x 1,000 MHz | 320 |
Radeon HD 3870 X2 | R680 | 2 x 512 MB GDDR3 | 823 MHz | 4.1 | 2 x 900 MHz | 2 x 320 |
Radeon HD 3870 | RV670 | 512 MB GDDR4 | 776 MHz | 4.1 | 2 x 1,125 MHz | 320 |
Radeon HD 3850 | RV670 | 256 MB GDDR3 | 668 MHz | 4.1 | 2 x 829 MHz | 320 |
SPs=Stream Processors, R680=2xRV670, R700=2xRV770, Shader 2.0=DirectX 9.0, 3.0=DirectX 9.0c, 4.0=DirectX 10, Shader 4.1=DirectX 10.1
| Test System For Nvidia and ATI Graphics Cards | |
|---|---|
Motherboard | Asus P6T, PCIe 2.0, ICH10R, 3-Way SLI |
Chipset | Intel X58 |
Memory | Corsair, 3 x 2 GB DDR3, TR3X6G1600C8D, 2 x 570 MHz 8-8-8-20 |
Audio | Realtek ALC1200 |
LAN | Realtek RTL8111C |
HDDs | SATA, Western Digital, Raptor WD300HLFS, WD5000AAKS |
DVD | Gigabyte GO-D1600C |
Power Supply | CoolerMaster RS-850-EMBA 850 W |
Drivers and Configuration | |
Graphics | ATI Catalyst 9.5, Nvidia GeForce 185.85 WHQL |
OS | Windows Vista Ultimate 32-Bit, SP1 |
DirectX | 9, 10, and 10.1 |
Chipset Driver | Intel 9.1.0.1007 |
wtf?!
no 4850X2, are you people MAD?!
article after recent article tom's has recommended the 4850X2..so why isn't it in this chart???
cheers,
bill
p.s. stuff and nonsense: http://www.eupeople.net/forum
Good article. Can you do the same for SLI and Crossfire including the older cards? Then in the summary show the single and dual card tables side by side?
you guys should go see the comments on the US page...
Surprised that Sapphire submitted the old HD4850 model for testing, the newer one has a much better cooler and with the altered p-states (lowest is 160/250MHz!) its passive at idle.
LOL! Just had a gander at the US page as xizel suggested...
Wondered how Toms had suddenly reversed the set-in-stone law of HD4850>GTS250 and HD4870>(old)GTX260... Well, now we know. They cheated and used dodgy benchmarks (yet again, and I've been forced to say that line A LOT lately!) and then did their usual plug routine.
Looks like Toms is part of nVidia's TWIWMTBP program. Y'know, The Way It Was Meant To Be Paid. Mmm, payola time!
No idea why they still employ Kriess or Roos, no-one in their right mind can believe anything Toms says until those two are locked up on corruption charges in nVidia's inevitable Intel-style anti-competitive suit... XD
Tom's just wrote an article on the 4850X2, yet they fail to include it in the test.......smoke another one guys.
Yeah I agree Solitaire, they have always been Intel/Nvidia biased.
Seems The Last Remnant isn't Radeon optimised...wonder how much nVidia paid them
Upgrading soon, still debating whether to go Phenom II or Core i7 as the cost difference isn't as much as it used to be, but is still around 150+ quid more mainly because the motherboards are more expensive for Intel, plus the need for at least 3 sticks of DDR3...and it seems to me that using SLi/Crossfire on Intel chipsets appears to affect their performance compared to a dedicated chipset for SLi or Crossfire, as i get surprisingly similar results with a 4890 and an OLD Phenom 9850 @ just 2.8GHz...weird, isn't it? Pretty sure some driver trickery is going on here, too.
God, how big must their stockpile be? Or it's just a cheap easy way to make more money to bribe people. Lol.
God, TH has gone downhill hasn't it? I miss the old site layout too!
Also interesting that the Phenom/II based rigs still do really well in games...of course if i were encoding lots of porn to DVD i'd probably get a Core i7, like battle times of old Intel seems slightly faster at encoding.
Also, LOL @ nVidia for getting rid of more G92 based GPUs marked as a GTX250
Hmm.. down here, you can get 2x 4870 1Gb's for 150€/each, totalling the best thing you can get for under 300€. How's that for you?
well i like getting a good deal and i bought a radeon 4830 running it with a fx62 @2.8Ghz and i can play all games at HD with very playable framerates and i reckon if i crossfire it then i'll get about the same as a gtx280 for under £200 so why isnt the best budget card in the bench mark ?
ps toms hardware been using your site for years and years and its a great site well worth reading, sometimes does seem a bit biased one way or another which is fine cos if i was doing the job these big companys could buy my vote for a month or two as well
6gb ram on 32bit windows?
I am a big ATI Fan but upgraded my vga card about 7 months ago... went for the Asus 4870 1gb also tried the Shapire version but could not get it going on my dual screen vista 64 pc so had to swap it at the supplier for a Asus gtx 260 896mb .. must say great card and great software ... pitty I could not get the same support from ATi
I think the normalization is lacking a bit.
What I suggest is a method of weighing the average fps rates based on relevance: supposing the refresh rate of most monitors around these days is 60Hz, i would give 100 points to a card that achieves that value (60fps). A card that achieves less than that would get a percentage of those points (30fps->50 points), but for higher values the 'bonus' should not be as high (ie: for 120fps the card should NOT get 200 points because those extra 60fps are mostly irrelevant; maybe one point/fps over 60 resulting in 160points for 120fps or even half a point resulting in 130).
This would make extra 'power' more relevant in titles that actually NEED it.