Best Of The Best: High-End Graphics Card Roundup

Graphics Chips And Test Configuration

We used an Intel Core i7-920 CPU for our testing purposes, boosting its clock speed to 3.8 GHz in order to circumvent any potential CPU bottlenecks.

This time, our technical tables are more comprehensive to help us detail how our retail test cards differ from the clock rates for standard reference cards. You'll find three different generations of manufacturing technology represented here as well: 65, 55, and 40 nm all exert influence on maximum power consumption and operating temperatures for graphics chips. These tables also illustrate how overclocking on specialty models boosts fill rates and raw computing power for their graphics chips.

Nvidia Graphics Cards
Manufacturer and Chipset
Code NameGraphics RAM
GPU clock
ShaderRAM clock
SPs
EVGA GTX 295 Hydro Copper (GeForce GTX 295)2 x GT200b2 x 896 GDDR3720 MHz4.0, 1,548 MHz2 x 1,080 MHz2 x 240
GeForce GTX 2952 x GT200b2 x 896 GDDR3576 MHz4.0, 1,242 MHz2 x 999 MHz2 x 240
Zotac GTX285 AMP Edition (GeForce GTX 285)GT200b1,024 GDDR3702 MHz4.0, 1,512 MHz2 x 1,296 MHz240
MSI N285GTX SuperPipe OC (GeForce GTX 285)GT200b1,024 GDDR3680 MHz4.0, 1,476 MHz2 x 1,250 MHz240
GeForce GTX 285GT200b1,024 GDDR3648 MHz4.0, 1,476 MHz2 x 1,242 MHz240
MSI N280GTX OC HydroGen (GeForce GTX 280)GT2001,024 GDDR3700 MHz4.0, 1,400 MHz2 x 1,150 MHz240
GeForce GTX 280GT2001,024 GDDR3602 MHz4.0, 1,296 MHz2 x 1,107 MHz240
BFG GTX 275 (GeForce GTX 275)GT200b896 GDDR3648 MHz4.0, 1,440 MHz2 x 1,152 MHz240
GeForce GTX 275GT200b896 GDDR3633 MHz4.0, 1,404 MHz2 x 1,134 MHz240
GeForce GTX 260 216SPsGT200b896 GDDR3576 MHz4.0, 1,242 MHz2 x 999 MHz216
GeForce GTX 260GT200896 GDDR3576 MHz4.0, 1,242 MHz2 x 999 MHz192
GeForce 9800 GTX+G92b512 MB GDDR3738 MHz4.0, 1,836 MHz2 x 1,100 MHz128
GeForce 9800 GTXG92512 MB GDDR3675 MHz4.0, 1,688 MHz2 x 1,100 MHz128
Manufacturer and Chipset
Memory bus
Fab technology
TransistorsInterface
EVGA GTX 295 Hydro Copper (GeForce GTX 295)2 x 448-bit55 nm2 x 1,400 millionPCIe 2.0
GeForce GTX 2952 x 448-bit55 nm2 x 1,400 millionPCIe 2.0
Zotac GTX285 AMP Edition (GeForce GTX 285)512-bit55 nm1,400 millionPCIe 2.0
MSI N285GTX SuperPipe OC (GeForce GTX 285)512-bit55 nm1,400 millionPCIe 2.0
GeForce GTX 285512-bit55 nm1,400 millionPCIe 2.0
MSI N280GTX OC HydroGen (GeForce GTX 280)512-bit65 nm1,400 millionPCIe 2.0
GeForce GTX 280512-bit65 nm1,400 millionPCIe 2.0
BFG GTX 275 (GeForce GTX 275)448-bit55 nm1,400 millionPCIe 2.0
GeForce GTX 275448-bit55 nm1,400 millionPCIe 2.0
GeForce GTX 260 216SPs448-bit55 nm1,400 millionPCIe 2.0
GeForce GTX 260448-bit65 nm1,400 millionPCIe 2.0
GeForce 9800 GTX+256-bit55 nm754 millionPCIe 2.0
GeForce 9800 GTX256-bit65 nm754 millionPCIe 2.0
Manufacturer and Chipset
Pixel Fill Rate Gpixel/sTexture Fill Rate Gtexel/sMemory Bandwidth GB/sDie Size mm²
EVGA GTX 295 Hydro Copper (GeForce GTX 295)2 x 20.22 x 46.12 x 121.02 x 470
GeForce GTX 2952 x 16.12 x 36.92 x 111.92 x 470
Zotac GTX285 AMP Edition (GeForce GTX 285)22.551.4165.9470
MSI N285GTX SuperPipe OC (GeForce GTX 285)21.849.8160.0470
GeForce GTX 28520.747.4159.0470
MSI N280GTX OC HydroGen (GeForce GTX 280)22.4
51.2147.2576
GeForce GTX 28019.3
44.1141.7576
BFG GTX 275 (GeForce GTX 275)18.1
41.5129.0470
GeForce GTX 27517.7
40.5127.0470
GeForce GTX 260 216SPs16.136.9111.9470
GeForce GTX 26016.1
36.9111.9576
GeForce 9800 GTX+11.8
47.270.4276
GeForce 9800 GTX10.8
43.270.4330
ATI Graphics Cards
Manufacturer and Chipset
CodenameGraphics RAM
GPU Clock
ShaderMemory clock
SPs
Radeon HD 4890R7901,024 MB GDDR5850 MHz4.14 x 975 MHz800
Radeon HD 4870 X2R700 (2 x RV770)2 x 1,024 MB GDDR5750 MHz4.14 x 900 MHz2 x 800
Radeon HD 4870RV770512 MB GDDR5750 MHz4.14 x 900 MHz800
Radeon HD 4850RV770512 MB GDDR3625 MHz4.12 x 993 MHz800
Radeon HD 4770RV740512 MB GDDR5750 MHz4.14 x 800 MHz640
Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
111 comments
    Your comment
    Top Comments
  • Anonymous
    Only one ATi card? What happened to all those OC'd 4890s?
    23
  • Anonymous
    Weird test:
    1) Where are the overclocking results?
    2) Bad choice for benchmarks: Too many old DX9 based graphic engines (FEAR 2, Fallout 3, Left4Dead with >100FPS) or Endwar which is limited to 30FPS. Where is Crysis?
    3) 1900x1200 as highest resolution for high-end cards?
    16
  • curnel_D
    Ugh, please tell me that The Last Remnant hasnt been added to the benchmark suite.

    And I'm not exactly sure why the writer decided to bench on Endwar instead of World In Conflict. Why is that exactly?

    And despite Quarz2's apparent fanboism, I think HAWX would have been better benched under 10.1 for the ATI cards, and used the highest stable settings instead of dropping off to DX9.
    10
  • Other Comments
  • Anonymous
    Only one ATi card? What happened to all those OC'd 4890s?
    23
  • Anonymous
    And those HAWX benchmarks look ridiculous. ATi should wipe floor with nvidia with that. Of course you didn't put dx10.1 support on. Bastard...
    8
  • cangelini
    quarzOnly one ATi card? What happened to all those OC'd 4890s?


    These are the same boards that were included in the recent charts update, and are largely contingent on what vendors submit for evaluation. We have a review upcoming comparing Sapphire's new 1 GHz Radeon HD 4890 versus the stock 4890. It'll be up in the next couple of weeks, though.
    1
  • ohim
    Am i the only one that find this article akward since looking at the tests done on Ati cards on The Last Remnant game makes me wonder what went wrong ... i mean it`s UT3 engine ... why so low performance ?
    4
  • curnel_D
    Ugh, please tell me that The Last Remnant hasnt been added to the benchmark suite.

    And I'm not exactly sure why the writer decided to bench on Endwar instead of World In Conflict. Why is that exactly?

    And despite Quarz2's apparent fanboism, I think HAWX would have been better benched under 10.1 for the ATI cards, and used the highest stable settings instead of dropping off to DX9.
    10
  • anamaniac
    The EVGA 295 is the stuff gods game with.

    I would love that card. I would have to replace my whole system to work it properly however.
    I want $1500 now... i7 920 (why get better? They all seem to be godly overclockers) and EVGA 295.

    How about a test suit of the EVGA GTX 295 in crossfire for a quad-gpu configuration? I know there's driver issues, but it would be fun to see what it could do regardless. Along with seeing how far Toms can OC the EVGA GTX 295.
    Actually... Toms just needs to do a new system building recommendation roundup. I find them useful personally, and would have used it myself had my cash source had not lost his job...
    -12
  • Anonymous
    Weird test:
    1) Where are the overclocking results?
    2) Bad choice for benchmarks: Too many old DX9 based graphic engines (FEAR 2, Fallout 3, Left4Dead with >100FPS) or Endwar which is limited to 30FPS. Where is Crysis?
    3) 1900x1200 as highest resolution for high-end cards?
    16
  • EQPlayer
    Seems that the cumulative benchmark graphs are going to be a bit skewed if The Last Remnant results are included in there... it's fairly obvious something odd is going on looking at the numbers for that game.
    4
  • armistitiu
    Worst article in a long time. Why compare how old games perform on NVIDIA's high end graphic cards? Don't get me wrong i like them but where's all the Atomic stuff from Saphire, Asus and XFX had some good stuff from ATI too. So what.. you just took the reference cards from ATI and tested them? :| That is just wrong.
    9
  • pulasky
    WOW what a piece of s********** is this """"""review"""""" Noobidia pay good in this days.
    -7
  • darkpower45
    ok i tried playing The Last Remnant on my comp with my 4870x2 and it failed hardcore >.< the game itself is ridiculously boring too. sooo why is it added to the benching list?? *shakes head* makes me sad...
    -1
  • guusdekler
    I find it a lack this tests do not include the 3DMark Vantage suite.
    Ok, there aren't many games using DX10, but some very good ones do !.

    Thats the reason i've switched to vista.

    And with me enough people to justify a proper DX10 benchmark.
    1
  • Luscious
    No mention of the GTX 285 2GB version? I'm planning on picking up three of these for a tri-SLI Core i7 build, all water-cooled and overclocked.
    -6
  • Ellimist
    well i'm running an factory overclocked gtx285. only because i like solid drivers and DAAMIT doesn't seem to be able to provide these consistently. thats been my biggest problem in picking up an ATI card.

    This review however is terrible. the benchmark selection is dated if nothing else. even toms other reviews of recent have used better benchmarks than this.
    -9
  • sosofm
    This benchmark is not fair for Ati !!!
    10
  • IronRyan21
    Lets see some 3dmarkVantage pls
    3
  • drealar
    Like car review magazine (like the one my friend is working for), I THINK they only have cards that were submitted to them and (not sure if this is the case with Tom's) they're only lended for a limited amount of days.

    Although I'm not very satisfied (coz lack of ATI card in your possession), I thank you for the review with Fallout, Left 4 Dead and Last Remnant with DX9. Yup I'm still using XP coz the bog-down symptom with Vista is too noticeable for my rig.

    1920x1200 as minimum threshold? Cool, as my 23" is limted to 1920x1080 anyway :P
    0
  • JeanLuc
    I think you guys should cut Tino Kreiss some slack this I believe his first publication? Saying things like "this is the worst article I've read in a long time" doesn't actually help. You can blame the choice of benchmarks suites on sites manager/editor not the author as he only does what he is told to write. So with that in mind............Cangelini your fired.

    I am curious though, HAWX is a game sponsored by ATI so why is the HD4890 getting it's backside tanned by the GTX275? It's not just a few FPS behind either the difference is quite remarkable and yes I do realise the BFG GTX275 is overclocked but it's not overclocked by a lot.
    1
  • cangelini
    JeanLucCangelini your fired.


    Perhaps you'll hire me as a copy editor for your posts instead? ;-)

    In all seriousness, Tino has been with Tom's German office for a long time. I've asked the staff responsible for testing there to drop in and provide some feedback on the products and benchmarks used here.

    Best,
    Chris
    0
  • linaaslt
    How lame this article is... i was always wondering why they don't use full potential of gpu, if ATI is capable of using DX10.1 (and game uses that technology), why not use it. it might not be fair for nvidia but ffs, i believe that this kind of review should show all potential of products.
    Shame for TH!
    9