Memory Overclocking

Data availability is important in the performance of GPUs, and AMD and NVIDIA pack huge amounts of bandwidth into their designs in order to accommodate this need. While AMD's high end parts have moved over to the newer, less tested, GDDR5, NVIDIA will stick with GDDR3 until at least their next architecture revision (though it is still unclear exactly what memory technologies NVIDIA will support beyond the current generation). This does mean that NVIDIA needs twice the number of pins to achieve the same bandwidth (at the same clock speed), but this isn't a huge problem for the already monolithic G80 and GT100 based GPUs.

With the 448-bit wide connection to GDDR3 memory, NVIDIA's GTX 275 needs to run it's RAM at a higher clock speed in order to achieve the same data rate the Radeon 4890 can hit with it's 256-bit GDDR5 bus. Certainly fast GDDR3 has had time to mature and is highly available. This and the fact that demand is still much higher for GDDR3 mean that NVIDIA is saving some money on competitive memory subsystems. But needing a higher baseline clock speed to compete with AMD's solution could mean less overclockability overall.

We were able to get a greater than 23% clock speed increase out of our 4890, but the best we could manage between a couple of GTX 275 samples was a little more than 14%. Starting out with very nearly the same memory bandwidth, our overclocked AMD part comes out ahead in absolute terms.

It is important to remember, however, that absolute bandwidth doesn't matter as much as how well the bandwidth matches the demand of the GPU. This isn't something we can easily ascertain, but our look at the impact of only overclocking memory certainly shows that the bandwidth NVIDIA chose for the GTX 275 is a good match for the core and shader clock speeds with which it is paired.

1680x1050    1920x1200    2560x1600

We will be digging deeper into how memory speed impacts performance after we look at the rest of our scaling tests, but without any other assistance, just overclocking memory is not going to gain a lot for the GTX 275.

Index Core Overclocking


View All Comments

  • serrias - Monday, June 08, 2009 - link

    Sorry if its been posted, But the math is wrong :S...

    Thats like saying.. you have 3 people (each one represents either memory,core or shader) they are all making wallets, and can each do 10 in an hour.
    That means 3 people x 10 wallets per hour = 30 wallets a hour total.

    Then you tell them each to work 10% faster...... Therefore..
    Wallet maker one does 11 wallets a hour (1.1x10=11)
    Wallet maker two also does 11 wallets a hour (1.1x10=11)
    And Wallet maker Three also does 11 wallets a hour (1.1x10=11).

    This equates to 33 wallets an hour (for a 10% increase from each wallet maker)

    This is still overall just a 10% increase in performance.

    Now onto the artical, It is impossible to get a performance increase above the amount overclocked... Right?.... The highest overclock here was about 17% on the shader if i remember right
    (So by my math even if the core and memory were also 17% overclocks it would still just be 17% faster overall... From my example above)
    .. This all makes it definatly impossible to get a 30%+ increase.

    Sorry, But this artical is screwed up.
  • Tigashark - Monday, June 08, 2009 - link

    Serias you are making the assumption that the core , shader and mem are all performing the same job "Making wallets" and therefore the performance improvement should increase in a linnear fashion

    Using your wallet maker example , changing the layout of their workspace by "10%" might lead to a 20% increase in efficiency and corresponding 20% increase in wallets per hour

    the shader and core clocks obviously have a multiplicive relationship , not an additive one and the core and shader clocks are obviously helping eachother do their job more efficiently in some manner .

    Im sure an Nvidia engineer would be able to answer the "why" of this ;)

    Science is based on theory -> experiment -> observation ->conclusion (and maybe another round of experiments if the results dont fit your theory)

    In this case your trying to say "but my theory must be right so the observed results are wrong"

    Thats like saying "the world is flat so something else must be happening when people circumnavigate the earth , cos it couldnt possibly be that its not flat"

    Bottom line you can throw as much theory at this as you like , the

    Observed results speak for themselves .

    Great article btw :)
  • Tigashark - Monday, June 08, 2009 - link

    Meant to say

    Science is based on theory -> experiment -> observation ->conclusion and if need be - > REVISION
    (if the results dont fit the theory , you revise the theory untill it explains the results , THEN retest)

  • serrias - Tuesday, June 09, 2009 - link

    You are correct in how the core, shader and memory are seperate, and how those seperate things are unlinked and affect the framerates in very differnt proportions, I did originally think about this... But I just to make it more understandable... I used a simplified example to show it. It still is incorrect though.

    The basic principal is the same, You cannot achieve a larger performance increase the the percentage you originally increased from.
    The wallet making example was suitable for simple understanding , Using that example you are correct, If you did for increase workspace by 10% you may get a 20% increase in wallets per hour, the only case where this could happen is where the workspace is the limiting factor, and the increase is because perviously he could only fit say 90% of the needed materials on the desk, whereas now he can fit 100% and work at a much faster rate (seeing as he dosnt need to go back and get more materials half way through.)

    But, This does not apply to a GPU efficiency because its just a differnt enviroment.

    Im sorry but no matter what, It is definatly impossible to overclock three components by 17% 10% and 10% (or whatever the review said) and achieve a theoretical performance improvement of 50%, with a actual of up to 32%... Perhaps by freak occourance may happen to around 19-20%, But that would be driver issues and inperfections in architecture, No way would 32% be possible!!!
  • serrias - Tuesday, June 09, 2009 - link

    Oh yeah, finally,
    Its obviously ridiculous if the clocks have a multiplying increase
    (xxx*xxx*xxx = total percentage is not right.)

    Let me put this into simpler terms so you can understand....

    That is the same as using your analogy...
    if you decrease the clocks by 50% on each one you will come out with
    So thats like saying if you run 50% of the speed youll get 12.5% of the performance

    Or.. even more stupid...
    If you run 20% of the original core speed you get 0.8% of the original performance

    Or the other way round... a 50% clock increase on all components (actually possible on some GPU's) would be
    1.5*1.5*1.5 = 3.375
    So a 50% increase in clocks means a thoretical increase of 337.5% ..... :S
    So can you see now how it is clear that you cannot multiply the percentage increase to determin theoretical MAXIMUM performance... (Maximum reffering to the pervious running at 20% of clock speeds = 0.8% MAXIMUM performance.... (Have to follow same rules for all equations)
  • tenq - Monday, July 26, 2010 - link

    China wholesale electronics
    China wholesale from
  • mypapayaone - Monday, June 08, 2009 - link

    China wholesale electronics
    China wholesale from papayaone">
  • helldrell666 - Saturday, June 06, 2009 - link

    Extremely useful Data Mr. Derek.I just love your articles.Your one of the great guys at anandtech who keep anandtech unique in every single way.

  • KhadgarTWN - Friday, June 05, 2009 - link

    A great review, but the problem is still there,
    4890 perform far worse in reality than these canned benchmark

    My 4890 has no change on par with GTX 275 at 1920x1200 on most of game.... I have them both on E8500
  • SiliconDoc - Saturday, June 06, 2009 - link

    If you notice the games also have some customized settings and the 4890 they used was a special for their eyes only manufacturer direct channel non retail version ! LOL
    Yes of course, the red rooster BS here is so thick it's PATHETIC.">
    "We absolutely must caution our readers once again that these are not off-the-shelf retail parts. These are parts sent directly to us from manufacturers and could very likely have a higher overclocking potential than retail parts."
    Yes, caution us way down into page 3 of the 4890 extravaganza review - but just COPY AND PASTE those special card numbers here and NEVER MENTION IT - and put a sappy 703 core on the GTX275 and call it "extravaganza!"
    Yes, your 275 whomps the 4890 in real life - that's why they have to go through maxxing it out for 4890 here to get what they got - a specially OC'able card from manufacturer - specialized custom game settings and ati favorable games in the majority - a great oc on ati not even achievable retail, and a crappy oc on nvidia that RETAIL BEATS !">
    Yeah, just really grewat and valuable data as hell666 says the next page... because he doesn't have a clue, he's bleeding out red rooster all over.

Log in

Don't have an account? Sign up now