How does temperature effect stability

Discussion in 'Overclocking & Hardware' started by jebusv20, Sep 20, 2012.

  1. jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    hey guys, so I'm just interested in the physics of OC.
    Does temperature in any way effect stability?

    example: a CPU is unstable at frequency F, temperature T and voltage V

    would simply dropping the Temperature of the cores increase stability?

    If not, is stability a pure function of frequency and voltage , where extreme cooling is only useful for situations where the temperature would be dangerous for a voltage required to maintain stability at a high frequency?
     
  2. Liighthead

    Liighthead New Member

    Joined:
    Jul 22, 2010
    Messages:
    212
    Location:
    Melbourne
    i would say yes. alot

    look at rigs that are watercooled to air cooled and air cooled to stock coolers..
    can OC a lot more just by upgrading the heatsink ( eg my 2600k could do 4ghz on stock cooler ( ik its nothing but i like my things cool ) now with a D14, sits at 4.6 24/7 folding... and can go higher lol )

    also with benching ect you couldnt get ( without something going boom ) a cpu upto like 7ghz lol but it can be done with dry ice/ Ln2 ect ect ect

    all in all Y3$ it does help :) :rolleyes:
     
  3. looktall

    looktall Working Class Doughnut

    Joined:
    Sep 17, 2001
    Messages:
    26,532
    my basic understanding of it is that electrical signals flow better at lower temperatures.
    increased temp reduces electrical flow, which in turn causes instability.
     
  4. OP
    OP
    jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    I think you will find that goung from stock to big air to water only really changes the overclock by allowing more thermal output without hitting a wall. What i mean to say is, changing the cooling mechanism has (from my personal experience) only served to allow for greater volts under a specific temp. But i have never gained frequency without
    a voltage bump.
     
  5. Unframed

    Unframed Member

    Joined:
    Mar 30, 2010
    Messages:
    9,158
    Location:
    Hella south west
    Sort of true. The only major issue comes with heat that will damage the electrical device. The issue is that heat is a bi-product of electrical flow.

    As for OC's in my experience your stability issues with heat will only be limiting your clock with thermal throttling/shutdown. A CPU at 75% of its thermal maximum won't be any less stable than one at 20%
     
  6. OP
    OP
    jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    My own personal experience agrees with this notion.
    How about in the case of extreme subambient cooling?
    Would an OC that was partially stable under air/water become stable under DICE or LN2?
    Or will the volts still have to rise?
     
  7. Unframed

    Unframed Member

    Joined:
    Mar 30, 2010
    Messages:
    9,158
    Location:
    Hella south west
    I've never worked with LN2 or anything more than kit water. However, to my knowledge LN2 can create something known as the "cold bug" where temperatures get so low that stability is jeopardized. When I was reading about this though it mentioned in the article that Phenom II CPUs seemed immune to the cold bug. More to the point of your question, volts would still need to rise and as the volts rise so does the temperature hence the need for such extreme cooling methods.
     
  8. The Wolf

    The Wolf Member

    Joined:
    Sep 9, 2011
    Messages:
    449
    CMOS switching speed is reduced as temperature increases. If you try to run the frequency of the system constant, but up the temperature to a point where the switching time is more than 1/frequency, you start to introduce BIT errors.
     
  9. Liighthead

    Liighthead New Member

    Joined:
    Jul 22, 2010
    Messages:
    212
    Location:
    Melbourne
    to a point yes would be more stable just changing the cooler..
    but then as you bring up the mhz, you need more voltage...

    so kinder need both lol to go faster :)
     
  10. chook

    chook Member

    Joined:
    Apr 9, 2002
    Messages:
    3,789
  11. OP
    OP
    jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    thanks for the link :)

    So if Voltage is the only thing that directly effects stability, why use extreme cooling if safe temperatures could be maintained with more conventional tech.
     
  12. Unframed

    Unframed Member

    Joined:
    Mar 30, 2010
    Messages:
    9,158
    Location:
    Hella south west
    You would not use high end cooling on stock clocks or a moderate OC
     
  13. Nerb

    Nerb Member

    Joined:
    Mar 2, 2006
    Messages:
    1,259
    Answering the OP, increasing temperature, by any means, increases electrical resistance in the circuits. The chips require current to operate (without the correct current, it will be unstable), so as temperature increases, resistance increases meaning you need higher voltage to maintain current. A higher voltage increases the temperature and around we go. So the cooler you can keep the chips, the less voltage you need. Ohms law.

    An electrician can tell me if im wrong.
     
  14. the3coopers

    the3coopers Member

    Joined:
    Jan 7, 2012
    Messages:
    1,685
    Location:
    Sydney 2151
    Let's say the machine was rated at 4.0GHz and you got it to 4.8GHz on air cooling at stock voltages. But when you tried 4.81GHz, it becomes flaky.

    Next you try to increase CPU Core voltage to get more stability, but it doesn't help - you've hit the wall!

    The overclocked CPU is getting hotter because of the higher frequency (it's working heaps harder than stock). So adding more voltage doesn't help, because the increased heat makes it LESS stable!

    But if you could drop the temps somehow, then you could pump more voltage into it. And with more voltage, your CPU is more stable.

    I think you may be thinking of it the wrong way around. The "extreme" cooling is not there to maintain "safe temperatures". It's about getting maximum frequency out of the device.

    The 4 things are linked. Frequency <> Voltage <> Temperature <> Stability

    a) The hotter it gets, the less stable it becomes (once it pases the instability threshhold).

    b) The faster it goes, the hotter it gets. Go back to step A.

    c) You can make it more stable if you give it more voltage.

    d) The more voltage you give it, the hotter it gets. Go back to step A.

    etc, etc

    The key to making it go fast is to make it STAY COOL.
     
    Last edited: Sep 21, 2012
  15. OP
    OP
    jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    so how would one tell if a clock is unstable because it needs more volts, or because its too hot?

    example: my 3930k (under water) can pass 3DM11 @ 5.1ghz using 1.5V, it freezes @ 5.2 running 1.55+, AFAIK the temps don't hit 65. Do I need more volts, or is it time to go DICE?
     
  16. the3coopers

    the3coopers Member

    Joined:
    Jan 7, 2012
    Messages:
    1,685
    Location:
    Sydney 2151
    You still seem to be not understanding that these things are linked.

    Your CPU may not be capable of doing 5.2 at ANY temperature or voltage.
    - Or it might do 5.2 on 1.6V
    - Or it might do 5.2 at 60°C, but not the 65°C that you're getting
    - Or it might need 1.75V to get to 5.2, but you cannot keep it under 90°C no matter how much cooling you chuck on it, so you blow it up :lol:

    Anyway, unless you do some "scientific" modelling, you will never know the answer.

    What I used to do was to make an Excel spreadsheet, and use graphing to determine what I needed to know to get max overclocks. Here is the methodology, you can design your own method around it...

    CHART 1. First, for a certain Speed, plot the Voltage vs Temp... see what the relationship is (it's not necessarily linear). You need to be able to extrapolate the effect of Voltage on CPU temps.

    You might find that for every 5% increase in Voltage, your CPU temp rises by 10%. Use percentages, it's easier because it scales better.

    Also note - unless you can temperature control your room then you need to use relative temps, in other words if Room Temp was 20°C and CPU was 55°C, then your charts should record that as 55-20 = 35°C. You need to work on the rise above ambient. Or normalise it somehow (outside the scope of this discussion) - whatever, you need to make sure that any instability wasn't the result of ROOM temps increasing, get it?

    CHART 3. Next, for a given Voltage, plot Speed vs Temp... see what the relationship is. You need to have enough data to extrapolate what effect frequency has on temps.

    You might find that a 5% increase in speed increases the temps by 5%, so that is linear (it might not be, this is just an example). Doesn't matter... the important thing is to KNOW what happens, rather than just guessing or assuming.

    CHART 3. Now, set the Speed pretty fast, fairly close to the edge (let's say 4.90GHz) and then start throttling your cooling fans (in a controlled way) to raise the temps. Monitor the temps... at some point, the CPU will get unstable - record that temp, because that is you max temp at 4.90GHz.

    Next, bump the CPU to 4.95GHz and repeat the same thing. Find out what the max temp is before the CPU gets flaky at 4.95GHz.

    Next, bump the CPU to 5.0GHz and repeat the same thing. Find out what the max temp is before the CPU gets flaky at 5.0GHz.

    And again at 5.05 and again at 5.1GHz. The more sample points the better.

    OK... now from that, and extrapolating of the graphs you made, you can determine what temp is your absolute max for 5.15GHz, and for 5.2GHz and for 5.25GHz. Call these these your "threshold temps"

    And if the extrapolated temp for 5.2GHz is less than the 65ºC that you're reading, then your problem isn't temps.

    So, therefore you need more volts... You already know that you won't get 5.2 on 1.55V, so work out how much the temps will increase by if you bump to 1.6V. Plug that back into the spreadsheet and see if your voltage is going to push you over the temperature threshold.

    Now, what I am focussing on here is voltage, because it's using the cooling that you already have and you can bump the voltages in BIOS>

    And now that you have all this data, you can then use it "backwards" to determine what effect making it COOLER will have. So, if you drop the CPU temp by 10% from 65ºC to 60°C, how much would you be able to drop the voltage and stay stable?

    It's all about getting the data ina mthodical way, so then you can crunch it and model the behaviours and how they interact.

    Aften you plot enough, what you'll fin dis that the graphs aren't straight lines, but are curves. And that the laws of diminishing returns apply - there is a point where no matter how many volts you ram into the thing (and provide cooling to stop it self destructing), it simply won't go any bloody faster.

    I hope that all makes sense... it's clear in my mind, but trying to write it down in a structured way in a casual forum post is not particularly easy. I'll see if I can find an article online that is better written, and maybe has some sample data. :)
     
  17. OP
    OP
    jebusv20

    jebusv20 Member

    Joined:
    Apr 11, 2011
    Messages:
    444
    Location:
    Brisbane 4074
    got yah. :)

    work out if temps are problem through extrapolation.
    If not, volt bump.

    Can anybody explain why so many people use extreme cooling (DICE / LN2) if it has such a small effect?

    I am starting to guess that it is because it makes high volts (1.6+) safer for longer periods of time, rather than a mark-able increase in instability in and of itself.
     

Share This Page

Advertisement: