Finding The Right Overclock And Testing Stability
Finding The Right Overclock
The first thing to note about overclocking a graphics card is that you should always have a rough idea of what that particular graphics card is capable of. The easiest way to do that if to read online reviews about the specific model of graphics card you have, for example the Gigabyte R9 270 OC. If you cannot find reviews of the specific model, then find reviews of the generic model: so the AMD R9 270. From reading reviews you should make a mental note of three or four different clock speed configurations reviewers achieved so you know where to be aiming for your overclock. With the particular graphics card we have I was not able to find any online reviews, but I managed to find plenty of R9 270 reviews. Based on reading those online reviews my expectation was a GPU clock speed between 1075MHz and 1175MHz and a memory frequency of 1475MHz-1575MHz. Of course you should consider that your graphics card may be amazingly good or bad and fall outside that expected range: this does happen. However, having a general expectation is still useful when you’re overclocking.
The next thing to consider is how far you want to push your graphics card: it is safe to run the overclock 24/7? In most cases the answer will be yes unless you’re over-volting by a lot (e.g. more than 100mv/0.1v). In most scenarios my recommendation would be to overclock your to its maximum stable clock on stock voltages, then roll back 25MHz or so just for a safe margin. This is mainly because if you’re right on the borderline of instability some applications may be unstable while others might not, you also might have pushed your graphics card too far which could result in a performance decline. I would recommend graphics card overclocking with stock voltage and stock fan profiles, those extra voltage tweaking options are not available on most graphics cards and generally speaking most graphics cards have smart fan profiles.
So when you begin overclocking you should be doing this simultaneously with stress-testing/stability-testing. The main reason for this is that most graphics cards will be able to handle very high clock speeds at the desktop but as soon as you put them under heavy 3D load they can become unstable and crash. My best advice would be to start with the graphics card core clock. Keep adding increments of 25MHz onto it until it becomes unstable, for this constant stability testing I allow Unigine Heaven to loop in the background with all the settings maxed out. I wait 2 minutes after raising each 25MHz increment before deciding if it is safe to move to the next increment. Keep doing this until your system freezes, blue screens or until the display driver resets and then you know you have pushed it past its maximum. That means your maximum stable overclock was the clock speed before the system crashed, so if it crashed at 1150MHz then 1125MHz should be your final overclock. During this I would recommend upping the AMD or Nvidia power limits to their maximums and I would leave the memory alone.
Next you should head on over to Furmark to further test the stability of the core clock. This should only be done for the core clock because Furmark is useless at testing the memory from my experiences. If your core clock is still stable under Furmark then bingo it should stay stable under most other things. If it isn’t stable and crashes then consider rolling back another 25MHz until the system is stable.
Next we can work on the memory overclocking. For this you should reset the graphics card core clock to the default value. Then start overclocking the memory with similar 25MHz increments but this time leave the stability test for a little longer, about 5 minutes, before upping the next increment. This is because memory problems are less critical and take longer to become apparent, memory problems are normally observable with things like artifacts, screen tearing (not related to frame-rates) and other random blotchy effects that wouldn’t be there at stock. The maximum memory overclock will depend on the memory type: SK Hynix memory modules overclock better than Elpida or Samsung chips. You can find out what brand of memory your graphics card uses by downloading TechPowerUp’s GPU-Z utility. The same principle applies with memory, keep going until you hit instability. Then roll back to the last stable speed. A few extra hints and tips to note with the memory are that sometimes a memory clock will be stable under 3D Load (games for example) but not 2D load (desktop environment), that will come in useful if you’re seeing artifacting on the desktop. You should also not check the “apply overclocking at system startup” option until you’ve found your final overclock, this applies for core and memory, because you could end up stuck in a system boot-loop of booting in at unstable clock speeds. If you find yourself at that position then you should boot into Windows at Safe Mode and uninstall or reset MSI Afterburner. Also note that pushing the memory “too far” can actually lower performance so try and benchmark the graphics card to ensure that your memory overclock is actually adding to performance, not making it worse. Your memory may be stable at 1600MHz for example but it might perform better at 1525MHz.
After all our overclocking efforts the final overclock we settled on was 1150MHz on the core clock and 1525MHz on the memory clock. This is a great outcome for our Gigabyte R9 270 OC which came with stock speeds of 975MHz core and 1400MHz memory. On the next page we will assess the performance gains this overclocking brings to the table. In percentage terms we achieved a 18% boost in the core clock and a 9% boost in the memory clock.