GUIDE TO OVERCLOCKING GPUsTools: GPU-Z |
GPU Tweak |
NVIDIA Inspector |
MSI Afterburner Note: You will want to run GPU-Z and go to the "Sensors" tab and keep GPU-Z running throughout this process. GPU-Z can be used to log values from these sensors or for the sake of simplicity, finding either the max/min/avg values instead. The most important ones are "GPU Temperature", "GPU Load" and "VDDC" but also watching to see if your increased core/memory clocks are being applied too. To find the max/min/avg values from your sensors in GPU-Z, simply click on the current (real-time) values displayed for them.
Now, open up your favorite overclocking utility. I should add that if you choose to use MSI Afterburner and have a notebook NVIDIA GPU, you will not have any voltage control as you would on GPU Tweak, EVGA Precision X or NVIDIA Inspector despite still having a modified vBIOS,
unless you
enable voltage control for MSI Afterburner
here.TEMPERATURE: Typically, load temps on air cooling vary around 65-92°C (ambient temps play a factor also) depending on the form factor of the notebook chassis used. Some gaming notebook models like that of MSI, Razer and Gigabyte (even ASUS) use these slim
portable chassis which are great in theory, but often executed poorly thermally which cause their GPUs to under perform next to identical GPUs in thicker and bulkier chassis. So what's worse, louder fan noise for higher temps in a slimmer chassis or lower fan noise for lower temps in a thicker chassis? So how does this affect the GPU overall? The thermal interface material (TIM) or as we call it, thermal paste, can only do so much in terms of efficiency on a poorly designed heatsink assembly. This results in load temps to creep up slowly and ultimately triggering a throttling point. The core clock or boost clock will most likely throttle (underclock) because the temperature is reaching what's known as the preset
temp target value. Some GPUs in their form factor chassis leave little to no headroom in terms of overclocking and result in a lesser performing system due to the temperatures reaching implemented limits already.
Artifacts here are a bit more involved. If you had already gone through overvolting and overclocking until you found what was stable and what wasn't with your GPU, you'd have likely seen your share of artifacts associated with testing it. The artifacts that arise when the temperature is higher is all those that associated with you earlier but
now happen much sooner and at values
known to be stable for your GPU. This can lead to terrible FPS, overclock stability and will result in constant core clock throttling, indefinitely, until temperatures are under the preset temp target values again.
TDP/POWER LIMIT: This is the main root of evil that is core/boost clocks and why they're not able to hold their values. Your FPS can drop suddenly when the power limit is reached. How do you monitor what's happening? By using GPU-Z, specifically on the sensors tab. There you will see in the PerfCap field as "Pwr" which indicates reaching the set power limit values of a stock vBIOS. If modified however, then the power limits are raised to allow the core/boost clocks to go higher paired with adequately increased voltage (and hold) without being limited or underclocked based on the modified power limit values used now, but this also introduces another problem called
black screening.
Artifacts that result in a display driver to crash and become unable to send out a GUI error to the system for recovery. This is known as black screening and is often caused by an extremely high core/boost clock and amount of voltage unstable for the card at the currently set power limit. It's
more likely to happen as you raise your core clock and voltage beyond controllable temperatures, toward the point when sporadic stability begins occurring. As is the case with some cards, it can happen at a lower OV/OC ceiling rather than higher.
CORE/BOOST CLOCK: You first want to establish how high you can overclock your core before your display driver crashes,
without increasing the voltage yet.
I usually start at 50Mhz intervals until I start crashing, then I lower it by 25Mhz and fine tune it from there until it's stable. When you do begin crashing from an overclock (usually from a high GPU load game or while benchmarking) your overclocked values will be defaulted and the card will be running back to its stock clocks again. This tells you that the core clock you set was too high given the current voltage, so you must lower your core clock and rinse and repeat this process until you find what
is stable
. The reason we do this without just increasing voltage right away and just maxing the core/memory sliders, is A.) we want to set a
base for how high it
did go (from the results) and B.) you'd crash almost instantly and receive nothing beneficial to work with as all you know now is it
crashed, but which value(s) had caused it? Setting a base
first helps better determine what your temps, your ceiling and your overall stability in performance at the end is going to be at.
Artifacts that are stretched/protruding or is an assortment of multicolored textures or horizontal/vertical lines flashing which are scattered all over the screen, indicate too high of a core clock at the current voltage used or the GPU is overheating (reaching triple digits) because it's not throttling.
MEMORY CLOCK: Usually, the memory clock is increased somewhere from 200-1400Mhz depending on the memory ICs used (vRAM) as well as the model (and architecture) of the card. Increasing the memory clock allows for even higher bandwidth (increases fps slightly) for faster data transferring, but increasing the memory clock
too high can result in decreased performance as well as making the card unstable to even the lowest memory allocated program or game. When increasing the memory clock, it's best to start at 200Mhz intervals until you start crashing, then lower it by 50Mhz and fine tune it from there until it's stable.
Artifacts that are black/white or multicolored flashing shapes of various sizes (dominant one) which are scattered over the screen in patterns, indicate an unstable and too high memory clock. Higher memory allocated programs or games
can cause the memory ICs to overheat, which will also display these artifacts as well.
VOLTAGE: When increasing the voltage (accompanied by increasing the core clock) start by around 50mV (testing it) and increasing it by 25mV until 200mV (max on most mobile cards) is reached as this is usually enough to go up 250-600Mhz on the core clock (taking notice on the load temperatures each time you test). This is obviously dependent on the bin quality of your die, as your result from the OV (overvolt) with the overclocking headroom
will differ from that of another. It could be bad or it could be good. Searching what others get using the card you have can help you figure out if it's considered (by you with the research or by someone else) a
good overclocker!
Artifacts that are solid or flashing red, green or black vertical bars on the screen, indicate too much voltage is applied to the card. There has also been times where these particular artifacts mean there
isn't enough voltage applied
, but at the
edge of becoming unstable itself.