GPU vs. CPU Upgrade: Extensive Tests

Overall Performance and Price Comparison

The overall results serve as a means of orientation and there is no weighting between benchmarks. The frame rates are simply added together; as a result, games with high fps results may balance out weaker resolutions.

Overall performance GPU-CPU fps
Geforce 8800 GTS OC (512 MB) Q6600@3.2 2830.7
Geforce 9800 GTX (512 MB) Q6600@3.2 2823.3
Geforce 8800 GTS OC (512 MB) X6800EE@2.94 2772.3
Geforce 9800 GTX (512 MB) X6800EE@2.94 2746.1
Geforce 8800 GTS OC (512 MB) E6750@2.67 2731.2
Geforce 9800 GTX (512 MB) E6750@2.67 2723.8
Geforce 8800 GT OC (512 MB) Q6600@3.2 2685.1
Geforce 8800 GT OC (512 MB) X6800EE@2.94 2640.8
Geforce 8800 GTS OC (512 MB) Q6600@2.4 2632.8
Geforce 9800 GTX (512 MB) Q6600@2.4 2623.6
Geforce 8800 GT OC (512 MB) E6750@2.67 2620.6
Geforce 8800 GT OC (512 MB) Q6600@2.4 2526.9
Geforce 9600 GT OC (1024 MB) Q6600@3.2 2393.7
Geforce 9800 GTX (512 MB) E2160@2.41 2371.8
Geforce 9600 GT OC (1024 MB) X6800EE@2.94 2363.4
Geforce 8800 GTS OC (512 MB) E2160@2.41 2346.4
Geforce 9600 GT OC (1024 MB) E6750@2.67 2336.9
Geforce 9600 GT OC (1024 MB) Q6600@2.4 2262.6
Geforce 8800 GT OC (512 MB) E2160@2.41 2225.6
Geforce 9600 GT OC (1024 MB) E2160@2.41 2039.0
Geforce 9800 GTX (512 MB) E2160@1.8 1975.0
Geforce 8800 GTS OC (512 MB) E2160@1.8 1944.0
Geforce 8800 GT OC (512 MB) E2160@1.8 1910.2
Geforce 9600 GT OC (1024 MB) E2160@1.8 1753.0
Geforce 7950 GT (512 MB) X6800EE@2.94 1112.7
Geforce 7950 GT (512 MB) E6750@2.67 1110.2
Geforce 7950 GT (512 MB) Q6600@3.2 1106.3
Geforce 7950 GT (512 MB) Q6600@2.4 1098.4
Geforce 7950 GT (512 MB) E2160@2.41 1073.0
Geforce 7950 GT (512 MB) E2160@1.8 1000.3
Geforce 6800 GT (256 MB) E6750@2.67 463.2
Geforce 6800 GT (256 MB) X6800EE@2.94 462.2
Geforce 6800 GT (256 MB) E2160@2.41 456.6
Geforce 6800 GT (256 MB) E2160@1.8 453.7
Geforce 6800 GT (256 MB) Q6600@3.2 445.1
Geforce 6800 GT (256 MB) Q6600@2.4 439.0

The upgrade table shows the performance increase when changing from one combination to another. The individual benchmark results have been converted to percentages, but there is no weighting for extremely high fps rates.

Here is how to read the table: the vertical combinations in the column on the left are the starting points "from", while the upper combinations of graphics cards and CPUs are the upgrade "to". Where the two lines meet, you will find the value that represents the performance gained or lost, in percent.

gpu cpu nvidia

The loss of performance can be at most -99.9%; otherwise the frame rate would be 0. The performance gain can be well over 100%, if for example, a frame rate rises from 1 to 10 or from 40 to 100 fps.

For between $464 and $541 (300 and 350 Euros), you can purchase a very fast combination of CPU and graphics card. Since the Q6600 at 3200 MHz is less than suitable for the boxed cooler, the price comparison has been calculated using a G0 stepping and Xigmatek HDT-S1283 costing $46 (30 Euros).

Price comparison of GPU and CPU GPU Euro CPU Euro GPU+CPU Euro
Geforce 7950 GT (512 MB) X6800EE@2.94 253 802 1055
Geforce 9800 GTX (512 MB) X6800EE@2.94 229 802 1031
Geforce 8800 GTS OC (512 MB) X6800EE@2.94 210 802 1012
Geforce 6800 GT (256 MB) X6800EE@2.94 189 802 991
Geforce 9600 GT OC (1024 MB) X6800EE@2.94 183 802 985
Geforce 8800 GT OC (512 MB) X6800EE@2.94 156 802 958
Geforce 7950 GT (512 MB) Q6600@3.2 253 198 451
Geforce 7950 GT (512 MB) Q6600@2.4 253 174 427
Geforce 9800 GTX (512 MB) Q6600@3.2 229 198 427
Geforce 8800 GTS OC (512 MB) Q6600@3.2 210 198 408
Geforce 9800 GTX (512 MB) Q6600@2.4 229 174 403
Geforce 6800 GT (256 MB) Q6600@3.2 189 198 387
Geforce 7950 GT (512 MB) E6750@2.67 253 132 385
Geforce 8800 GTS OC (512 MB) Q6600@2.4 210 174 384
Geforce 9600 GT OC (1024 MB) Q6600@3.2 183 198 381
Geforce 6800 GT (256 MB) Q6600@2.4 189 174 363
Geforce 9800 GTX (512 MB) E6750@2.67 229 132 361
Geforce 9600 GT OC (1024 MB) Q6600@2.4 183 174 357
Geforce 8800 GT OC (512 MB) Q6600@3.2 156 198 354
Geforce 8800 GTS OC (512 MB) E6750@2.67 210 132 342
Geforce 8800 GT OC (512 MB) Q6600@2.4 156 174 330
Geforce 6800 GT (256 MB) E6750@2.67 189 132 321
Geforce 9600 GT OC (1024 MB) E6750@2.67 183 132 315
Geforce 7950 GT (512 MB) E2160@1.8 253 52 305
Geforce 7950 GT (512 MB) E2160@2.41 253 52 305
Geforce 8800 GT OC (512 MB) E6750@2.67 156 132 288
Geforce 9800 GTX (512 MB) E2160@1.8 229 52 281
Geforce 9800 GTX (512 MB) E2160@2.41 229 52 281
Geforce 8800 GTS OC (512 MB) E2160@1.8 210 52 262
Geforce 8800 GTS OC (512 MB) E2160@2.41 210 52 262
Geforce 6800 GT (256 MB) E2160@1.8 189 52 241
Geforce 6800 GT (256 MB) E2160@2.41 189 52 241
Geforce 9600 GT OC (1024 MB) E2160@1.8 183 52 235
Geforce 9600 GT OC (1024 MB) E2160@2.41 183 52 235
Geforce 8800 GT OC (512 MB) E2160@1.8 156 52 208
Geforce 8800 GT OC (512 MB) E2160@2.41 156 52 208

The E2160 and E6750 with the Geforce 8800 GT or 8800 GTS 512 is the best option, purely from a price point of view.

Price comparison GPU and CPU Euros per frame
Geforce 6800 GT (256 MB) X6800EE@2.94 2.144
Geforce 7950 GT (512 MB) X6800EE@2.94 0.948
Geforce 6800 GT (256 MB) Q6600@3.2 0.869
Geforce 6800 GT (256 MB) Q6600@2.4 0.827
Geforce 6800 GT (256 MB) E6750@2.67 0.693
Geforce 6800 GT (256 MB) E2160@1.8 0.531
Geforce 6800 GT (256 MB) E2160@2.41 0.528
Geforce 9600 GT OC (1024 MB) X6800EE@2.94 0.417
Geforce 7950 GT (512 MB) Q6600@3.2 0.408
Geforce 7950 GT (512 MB) Q6600@2.4 0.389
Geforce 9800 GTX (512 MB) X6800EE@2.94 0.375
Geforce 8800 GTS OC (512 MB) X6800EE@2.94 0.365
Geforce 8800 GT OC (512 MB) X6800EE@2.94 0.363
Geforce 7950 GT (512 MB) E6750@2.67 0.347
Geforce 7950 GT (512 MB) E2160@1.8 0.305
Geforce 7950 GT (512 MB) E2160@2.41 0.284
Geforce 9600 GT OC (1024 MB) Q6600@3.2 0.159
Geforce 9600 GT OC (1024 MB) Q6600@2.4 0.158
Geforce 9800 GTX (512 MB) Q6600@2.4 0.154
Geforce 9800 GTX (512 MB) Q6600@3.2 0.151
Geforce 8800 GTS OC (512 MB) Q6600@2.4 0.146
Geforce 8800 GTS OC (512 MB) Q6600@3.2 0.144
Geforce 9800 GTX (512 MB) E2160@1.8 0.142
Geforce 9600 GT OC (1024 MB) E6750@2.67 0.135
Geforce 8800 GTS OC (512 MB) E2160@1.8 0.135
Geforce 9600 GT OC (1024 MB) E2160@1.8 0.134
Geforce 9800 GTX (512 MB) E6750@2.67 0.133
Geforce 8800 GT OC (512 MB) Q6600@3.2 0.132
Geforce 8800 GT OC (512 MB) Q6600@2.4 0.131
Geforce 8800 GTS OC (512 MB) E6750@2.67 0.125
Geforce 9800 GTX (512 MB) E2160@2.41 0.118
Geforce 9600 GT OC (1024 MB) E2160@2.41 0.115
Geforce 8800 GTS OC (512 MB) E2160@2.41 0.112
Geforce 8800 GT OC (512 MB) E6750@2.67 0.110
Geforce 8800 GT OC (512 MB) E2160@1.8 0.109
Geforce 8800 GT OC (512 MB) E2160@2.41 0.093

The extra cost for the Geforce 8800 GTS 512 is often acceptable, due to its better two-slot cooler. The Geforce 9600 GT with 512 MB is a better value for money, and very popular for those working with a budget of up to $186 (120 Euros).

Price comparison for graphics cards Euros per frame
Geforce 6800 GT (256 MB) 0.0695
Geforce 7950 GT (512 MB) 0.0389
Geforce 9800 GTX (512 MB) 0.0150
Geforce 9600 GT OC (1024 MB) 0.0139
Geforce 8800 GTS OC (512 MB) 0.0138
Geforce 8800 GT OC (512 MB) 0.0107

Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
75 comments
    Your comment
  • DjEaZy
    will there be a AMD/ATI roundup???
    2
  • randomizer
    That would simply consume more time without really proving much. I think sticking with a single manufacturer is fine, because you see the generation differences of cards and the performance gains compared to geting a new processor. You will see the same thing with ATI cards. Pop in an X800 and watch it crumble in the wake of a HD3870. There is no need to inlude ATI cards for the sake of this article.
    5
  • randomizer
    This has been a long needed article IMO. Now we can post links instead of coming up with simple explanations :D
    5
  • yadge
    I didn't realize the new gpus were actually that powerful. According to Toms charts, there is no gpu that can give me double the performance over my x1950 pro. But here, the 9600gt was getting 3 times the frames as the 7950gt(which is better than mine) on Call of Duty 4.

    Maybe there's something wrong with the charts. I don't know. But this makes me even more excited for when I upgrade in the near future.
    0
  • Anonymous
    This article is biased from the beginning by using a reference graphics card from 2004 (6800GT) to a reference CPU from 2007 (E2140).

    Go back and use a Pentium 4 Prescott (2004) and then the basis of these percentage values on page 3 will actually mean something.
    0
  • randomizer
    yadgeI didn't realize the new gpus were actually that powerful. According to Toms charts, there is no gpu that can give me double the performance over my x1950 pro. But here, the 9600gt was getting 3 times the frames as the 7950gt(which is better than mine) on Call of Duty 4. Maybe there's something wrong with the charts. I don't know. But this makes me even more excited for when I upgrade in the near future.

    I upgraded my X1950 pro to a 9600GT. It was a fantastic upgrade.
    0
  • wh3resmycar
    scyThis article is biased from the beginning by using a reference graphics card from 2004 (6800GT) to a reference CPU from 2007 (E2140).


    maybe it is. but its relevant especially with those people who are stuck with those prescotts/6800gt. this article reveals an upgrade path nonetheless
    0
  • randomizer
    If they had used P4s there would be o many variables in this article that there would be no direction and that would make it pointless.
    1
  • JAYDEEJOHN
    Great article!!! It clears up many things. It finally shows proof that the best upgrade a gamer can make is a newer card. About the P4's, just take the clock rate and cut it in half, then compare (ok add 10%) heheh
    2
  • justjc
    I know randomizer thinks we would get the same results, but would it be possible to see just a small article showing if the same result is true for AMD processors and ATi graphics.
    Firstly we know that ATi and nVidia graphics doesn't calculate graphics in the same way, who knows perhaps an ATi card requiers more or less processorpower to work at full load, and if you look at Can you run it? for Crysis(only one I recall using) you will see the minimum needed AMD processor is slover than the minimum needed Core2, even in processor speed.
    So any chance of a small, or full scale, article throwing some ATi and AMD power into the mix?
    0
  • randomizer
    In the case of processors, throwing in some AMD chips would be a good idea as they are often a fair bit slower than a similarly priced C2D. However, I don't think having ATI cards in the mix would show up anything really different than what we have now. The higher end cards will be bottlenecked on a slow processor while the slower cards won't be bottlenecked as badly. A 3870X2 will need a more powerful CPU to reach maximum potential just like a 9800GX2. Of course, the amount of CPU power needed will almost definitely be different, but the overall conclusion is that buying a next-gen (or current-gen rather) card is going to benefit you more than a new CPU unless it's real old. That is all the article is trying to prove, not which CPU/video card is best.
    0
  • LuxZg
    I've got say I agree about no need to add ATI cards, as they are easily comparable with the selection of nVidia cards shown in article.

    But it lacks comparision with single core CPU's. In the times of 6800GT's popularity we've had single core Athlons like Barton, and perhaps early Athlons/Semprons with 754 socket. As adding AGP system would complicate things way too much, I think at least Athlon64 socket 939 could be used, as those were very popular in between/during 6800GT and 7xxx series.

    At very least we should have one set of benchmarks on low/old CPU like that, so we can see if buying faster card is of any use at all, or will we reach a bottom of performance the same way that we see 6800GT unable to use additional power from new CPUs.

    Other than this, it's a great article!

    PLEASE - make one more roundup like this once GT200/4800 cards are out!

    P.S. And nice to see that 9800GTX, and overclocked Q6600 are still right on 300W consumption, meaning any quality 400/450W supply is more than enough for them!
    1
  • Reynod
    Onya ... thanks for pinching the idea from our thread on the subject buddy !!! Tou guys must troll our topics looking for good ideas eh??

    Seriously though ... cheers and thanks !!

    Hey ... update the article with a couple of AMD cpu's thrown into the mix ... perhaps a 3800, 5000. and a 6400 and a couple of 50 series Phenoms.

    Thanks again...
    2
  • cah027
    This is a really cool article. I think it should be updated with every new generation of new parts (ex: new ATi and Nvidia and Nahelem and Fusion)

    I wonder if Nehalem will be such a big boost if it will show gains across the board of GPUs ?
    0
  • danatmason
    I've been waiting for an article like this! Great to have figures to back up the idea of a cpu bottleneck. I tried to pair an 8800GT with an Athlon x2 4000+ at stock speed and I was HUGELY disappointed. But with a quick OC of the processor, I'm sitting happy! So I imagine those results scale similarly with AMD processors and the same idea - clock speed matters more for games, not cores - will still hold true.
    0
  • royalcrown
    there's no need to throw a bunch of AMD processors in here for one reason, well two...

    1. You can see on the cpu charts where the AMD you have compares with an intel...so it would perform about the same with the same card.

    2. cpu's in the same FAMILY on the same architecture will scale the same between them relative to the way the c2 duo's scale to each other.
    0
  • ubbadubba
    the E8xxx and E7xxx are not mentioned. Please either include in next edition (along with a few AMD CPUs) or comment on which parts they would mimic. Like, does the E8400 @ 3GHZ behave like a E2160 @3GHz, or is the E8400 still a little bit better by X%. E7200 and E8200 are fairly cheap for new CPU models, so that would be nice for the next go round.

    A game that's missing is UT3 -- it favorably benefits from X3 AMD CPUs as a cost-effective solution, whereas the X3 may not be as good in other games. But maybe that would throw off overall results if 1 game did really well and did not follow the same trends as other games.

    and yes we need AMD CPUs to see the scaling effects across the different GPUs. The interaction of weak CPUs with strong GPUs or vice versa is not represented in the CPU charts.

    speaking of which: when are the GPU charts going to get updated with modern GPUs and games?
    0
  • Anonymous
    What an incredible article! My jaw dropped to the floor at the cost per frame chart. Really nice work! Now, the holy grail is cross-linking article data like this to the CPU and Vid charts. I am very impressed.
    0
  • a 6pack in
    DjEaZywill there be a AMD/ATI roundup???

    that is something that i am really curious in. I've been bummed out that i killed my quad and back to dual. I guess its not really noticable..
    0
  • hcforde
    SCY,(15/05/2008@10:10) most people may have a greater tendancy to upgrade their GPU over their CPU. Maybe that is why it was done. Us techies may not see it that way but the general market may.
    0