Radeon HD 7000:  HD 7790 |  HD 7850 |  HD 7870 |  HD 7870 XT |  HD 7950 |  HD 7970 |  HD 7990 |  Radeon HD 8000:  HD 8750 |  HD 8770 |  HD 8850 |  HD 8870 |  HD 8950 |  HD 8970 |  HD 8990 
GeForce 600:  GTX 650 Ti Boost  |  GTX 660  |  GTX 660 Ti  |  GTX 670  |  GTX 680 |  GTX 690 |  Titan: GTX TITAN |  GeForce 700:  GTX 760  |  GTX 760 Ti  |  GTX 770  |  GTX 780 |  GTX 790 

July 31st, 2012

GeForce GTX 660 Ti Performance And Specification Leaked

Thanks to TweakTown we have the first benchmark results of upcoming GeForce GTX 660 Ti graphics card. The GeForce GTX 660 Ti seems to be a great solution for mid-range segment, but not in all applications.

TweakTown is known for multiple breaking of NDA’s, as their relations with NVIDIA aren’t really that great. I’m really surprised that they released tthose performance charts right now. Since there are still two weeks before card is officially released. Does it mean that the launch date is set sooner than expected? Well that was neither confirmed nor denied so it seems that you will still have to wait those two weeks to grab one of those cards (TweakTown says though that the card will hit retail in mid August).

And the most importantly, final specification that nobody will call rumored any more. Card features Kepler GPU with 1344 CUDA Cores. It has lesser Raster Operating Units than GTX 670 and GTX 680 – 24 (so there are three Raster Engines with 8 ROPs each – meaning there are only three Graphics Processing Clusters). Therefore I’m guessing that GeForce GTX 660 Ti is not using GK104 silicon, but if it does, it looks very odd. That’s because every SMX has 192 CUDA cores, so there should be at least 7 of those, but in GK104 every Graphics Processing Cluster can only have 2 SMXs (there’s one additional SMX which would not fit). So NVIDIA has either removed one Raster Engine with whole GPC and one SMX or this is a completely new architecture (and that’s why TweakTown decided not to reveal the GPU-Z readings).

Card is also featuring 2048 MB of GDDR5 memory operating on 192-bit interface. Standard clocks are 915 MHz for base clock, 980 MHz for boost clock and 1502 MHz (6008 MHz effective) for memory.

In 3DMark11 the GTX 660 Ti is performing about 6.4% slower than GTX 670 and 14% better than HD 7950. Those were the most anticipated comparisons. Card is also faster than HD 7970, but this is only the 3DMark performance, and in higher resolutions it will be more noticeable that 2GB card with only 1152 cores will not be faster than 2048 stream processors and 3GB in HD 7970.

Even though this review does not include the most popular games is shows that GeForce GTX 660 Ti has comparable overall performance to Radeon HD 7950. However, AMD card is much better solution if you are planning to use it with high-resolution and anti-aliasing enabled.

GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti

GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti GeForce GTX 660 Ti Performance And Specification Leaked geforce gtx 660ti

3DMark11 ComparisonPerformance PresetExtreme PresetPerformance
NVIDIA GeForce GTX 680 2GB96063063112,4%
AMD Radeon HD 7970 GHz Edition92483114109,7%
NVIDIA GeForce GTX 670 2GB90062986106,4%
NVIDIA GeForce GTX 660 Ti 2GB85892677100,0%
AMD Radeon HD 79708240276497,7%
AMD Radeon HD 79507008229685,6%
NVIDIA GeForce GTX 580 1.5GB6884212280,0%
AMD Radeon HD 78706788207578,7%
AMD Radeon HD 7850 (OC)6045179669,6%
  • skr13

    I´m waiting for TT Preview, but they usually use old games, what isn´t that great.
    Anyway 660 Ti with OC can catch 670 stock, but still don´t know if this Ti is 256-bit/1152 or 192-bit/1344.

  • SoCalledMe

    Wow , that looks great , if it ends up being $300 it will fit my budget and it will be best bang for buck , looking forward to get one :D

  • http://videocardz.com/ VideoCardz.com

    If what OBR wrote is true, then it may be 1344 cores with 192-bit…

  • skr13

    Above 1080p the performance will be on pair or slower that 7950.
    That card or 660 non-Ti can be the next “8800GT” ;)

  • BestJinjo

    Looks good using relative 3dMark11 scores to its GTX670/680 cousins but using 3dMark11 to measure overall gaming performance is misleading as the author noted. GTX680 is just 50 points worse than 1200mhz HD7970 in 3dMark11 but loses easily in real world games against a 1200mhz HD7970:

    http://www.techspot.com/review/555-gigabyte-radeon-hd-7970-soc/page3.html

    Good to see more competition at the $250-300 price level. Hope it starts price wars in the mid-range sector. :)

  • BestJinjo

    I am going with 1152 SP + 256-bit / 2GB. The performance is too close for it to be a 192-bit card. The 192-bit is more likely for a 660 but the Ti version likely has the full-fledged memory bandwidth intact if it’s performance is so close to the 670.

  • skr13

    This graph(3dmark) is about Stock cards, this review from techspot is a OC card, and 7970GHZ is a desperate measure to be Faster single gpu on earth…
    Agree we need competition and price wars in mid-range.

  • http://videocardz.com/ VideoCardz.com

    I think that memory bandwidth and memory size has the smallest impact on 3Dmark performance. Look at Radeon HD 7970 3GB 384-bit get beaten by this mid-range card.

    EDIT So I was right :)

  • Pingback: Nvidia GeForce 6xx Series (Kepler) - Page 62

  • Pingback: Filtrado el bestial rendimiento de la Nvidia GeForce GTX 660 Ti | El Chapuzas Informático

  • skr13

    UPDATE: So this is the real deal:
    http://www.tweaktown.com/reviews/4869/nvidia_geforce_gtx_660_ti_2gb_reference_video_card_review/index.html
    192-bit ,24 rops, 2GB, 1344 Cuda Cores.

  • http://videocardz.com/ VideoCardz.com

    Added :)
    Looks like a great card!

  • skr13

    Thanks, don´t know if you had saw it ;)
    Yes good performance, but if AMD drop 7950 to less than $300, still a valid option for me.
    HA, yes you were right about ROP´s question in the other day :)

  • SalemSwift

    Argh, why wouldn’t they test Battlefield 3 like everyone else?

  • skr13

    “but in GK104 every Graphics Processing Cluster can only have 2 SMXs (there’s one additional SMX which would not fit).”
    But that´s happen in GTX670 1344 SMX, where is only 1 SMX in a GPC, instead of 2.
    Yes, but with 24 rops is not possible to do like 670….
    FORGET IT ;(

  • http://videocardz.com/ VideoCardz.com

    I’m still rewriting the whole post.
    I added a picture of GK104 to pixelate how extraordinary it looks.

    GTX 660 Ti beats HD 7950 in most games (in Metro 2033 this correlation is opposite). Card has worse performance in higher-resolutions (that’s not surprising) and with anti-aliasing enabled.

  • BestJinjo

    You aren’t understand what I am trying to say. 3dMark11 is not an indication of gaming performance. Take HD7970 @ 1080mhz for example. It’s slower than GTX680 in 3dMark but in the real world, it beats GTX680 in games:

    http://www.techpowerup.com/reviews/Gigabyte/HD_7970_SOC/28.html

    Do you play 3dMark11 over games?

    If you skip to the end of the TT review, GTX660Ti’s 24 ROPs and neutred memory bandwidth make it tank in real world games:
    http://www.tweaktown.com/reviews/4869/nvidia_geforce_gtx_660_ti_2gb_reference_video_card_review/index13.html

    Also, HD7970 GE is no desperate attempt. It’s a real card and for $460, you can get a Vapor-X with amazing cooler that’s both faster and cheaper than the GTX680, and with overclocking will beat an overclocked 680:
    http://www.superbiiz.com/detail.php?name=AT-7970GHZ

  • BestJinjo

    Ya, you were right on the specs but not on the performance. In that review once AA is turned on on Page 13, 660Ti’s performance falls off a cliff, badly losing to the 7950. Against a 1.1ghz 7950, this card has no chance.

  • BestJinjo

    Look at AA performance of the 660Ti. Very very poor. NO way this card can touch an overclocked 7950.

  • http://videocardz.com/ VideoCardz.com

    That depends on the game, we all know that Metro 2033 is one of AMD’s favourites…

  • francmic

    Lets have alittle wager then shall we, I’ll shave my head if a reference 7950oc beats a reference 660ti oc

  • BestJinjo

    Read Page 13 with AA/AF turned ON. Poor performance, losing to the GTX580.

  • skr13

    No, i play video-games, but i remember the time when AMD leads in 3dmark06 or Vantage, before DX11 version appear.
    GTX 680 is for competing with 7970 1st Tahiti, not months later GHZ Editon.
    I have to say again that 7970GHZ consumes more power than GTX680 thats a fact, if some care or not about it.
    And 7970GHZ still didn´t appear in market, don´t know why?
    Prices of AMD7000 launch so high and still in IMO.
    Nvidia can drop prices when they want, launch GTX685 to destroy 7970GHZ etc.

  • skr13

    YEah, i´m waiting for Battlefield 3 or Crysis 2 bench.
    Dirt 3….LOL
    Anyway 680 vs 7970 670 vs 7950 660Ti vs 7870 who wins xD
    Nothing more i can say.

  • http://videocardz.com/ VideoCardz.com

    They are waiting for Battlefield 4 :)

  • OBR

    wait for proper results, with proper drivers … :)

  • francmic

    and a proper cooler

  • Raghar

    A card can have smaller speed than a card from previous generation that was two classes above. Look at 550 vs 460. Considering NVidia wanted to improve power consumption, nobody sane expected massive (two classes difference) improvements.

    Also when I moved to 1920×1080 monitor, I stopped to see any point in AA.

  • http://twitter.com/AlesOfficial JoyLessUA

    Seems like we will finally get a cheaper 7850 and 7870 :)

  • BestJinjo

    If you look at TweakTown’s latest review where they overclocked GTX660Ti, it loses to a stock GTX670 in all games with AA on. That’s despite a 155mhz GPU overclock. The 24 ROPs / 192 bit bus is going to result in much worse overclocking scaling than GTX670 cards.

  • BestJinjo

    1) Who buys reference HD7950? You need a strong cooler to overclock HD7950 to 1.1ghz.
    2) How far are we talking overclock? Max GTX660Ti vs. HD7950 at 1.1-1.15ghz?

    You might as well start shaving your head now then because at just 1.02ghz, HD7950 is just 2 fps behind a GTX680 in Battlefield 3 (One of NV’s best games) and 4 fps ahead of the 670:
    http://www.techpowerup.com/reviews/Sapphire/HD_7950_Flex/31.html

    Tweaktown’s overclocked GTX660Ti couldn’t even touch a STOCK GTX670:
    http://www.tweaktown.com/articles/4873/nvidia_geforce_gtx_660_ti_2gb_reference_video_card_overclocked/index13.html

    It’s really very simple mathematics. GTX660Ti has 75% of ROPs and 75% of the memory bus width of the GTX670/680. HD7950 has 100% of the ROPs and 100% of the memory bus width of the 7970.

    HD7950 has 40-50% overclocking headroom (800mhz –> 1100-1200mhz).

    At 925mhz, HD7950 is only 5-6% slower than the HD7970:
    http://hexus.net/tech/reviews/graphics/34761-amd-hd-7950-vs-hd-7970-clocks/?page=3

    It’s physically impossible for an overclocked GTX660Ti to beat an overclocked HD7950 since 7950 is basically an underclocked 7970 chip with a slight reduction in shaders and Texture units, while GTX660Ti is a 25% gutted GK104.

    Look at this review and the summary chart that was done just yesterday:
    http://www.xbitlabs.com/articles/graphics/display/geforce-gtx-690-sli_6.html#sect3

    HD7970 GE 1050mhz is faster than GTX680 1137mhz with the latest drivers. HD7950 @ 1.15ghz is faster than HD7970 GE 1050mhz. I’ll let you absorb that information at to how you believe a 24 ROP / 192-bit GK104 will beat an overclocked 7950 that has no problem beating a stock GTX680. :)

  • BestJinjo

    AMD had the fastest single-GPU from January 9, 2012 to March 22, 2012 with HD7970 (2.5 months)
    NV had the fastest single-GPU from March 22, 2012 to June 22, 2012 with GTX680 (3 months)
    AMD reclaimed the performance title starting June 22, 2012 with HD7970 GE and Catalyst 12.7 Beta.

    There are at least 2-3 1050-1100mhz HD7970 cards easily available in US and Europe after 5 min of Google search:
    - PowerColor Vortex II 1100mhz on Newegg for $450
    http://www.newegg.com/Product/Product.aspx?Item=N82E16814131471

    - Sapphire Vapor-X HD7970 GE for $458:
    http://www.superbiiz.com/detail.php?name=AT-7970GHZ

    - ScanFX Radeon HD7970 X-Edition (1050mhz) for
    £324
    http://www.scan.co.uk/products/3gb-scanfx-radeon-hd-7970-x-edition-5700mhz-gddr5-1050mhz-dvi-i-hdmi-2-x-mini-display-port

    There is also the 1070mhz MSI HD7970 Lightning

    vs.

    NV’s Asus GTX680 DirectCUII TOP 2GB that retails for way over $500 in the US and £478:
    http://www.scan.co.uk/products/2gb-asus-gtx-680-directcu-ii-top-28nm-pcie-30-(x16)-6008mhz-gddr5-gpu-1137mhz-boost-1201mhz-cores-15

    A 1050mhz HD7970 beats the Asus GTX680 Direct CUII Top 2GB:
    http://www.xbitlabs.com/picture/?src=/images/graphics/geforce-gtx-690-sli/zfulltable.png

  • BestJinjo

    So you can’t see jagged edges/stair-case anti-aliasing and crazy texture shimmering on leaves/foliage on a 1080P monitor in bright games like SKYRIM, Crysis, Mass Effect, Deus Ex, Max Payne 3, Battlefield 3? If you don’t use AA at all, $300 GPUs don’t make a lot of sense. You’d be better served with a $170 GTX560Ti or $220 HD7850. $300+ GPUs are there to max out settings. Not even a 2560×1600 monitor can fully mask aliasing. The aliasing on cables, building structures/vertical walls, vehicles is especially terrible in FPS games.

  • Notafanboy

    Do we know how many watts this thing consumes?

  • MOSLER

    So, I paid 150$ more for GTX 670 with 6,4% better performance compare to GTX 660 Ti…SHI~!

  • Pingback: The GTX-660TI is looking to be a Winner - Hardware Canucks