As already mentioned in other threads I’m a big fan of undervolting ([1] [2]), but now I’m wondering if overclocking has any downsides if the VCore is not increased.
Basically, this would result in the same performance without anything (power, temperature) changing for the system. Assuming the overclock is stable, one could only gain from it.
Or is there anything I’m missing? (Maybe coil whine?)
Since we’re in the era of aggressive algorithm-driven boost for both CPUs and GPUs I’d say that there aren’t any downsides running an overclock with stock voltages (assuming it is perfectly stable, that is).
The true contributor to silicon degradation is voltage because high temperatures over an extended period of time make the potential barrier in transistors smaller and smaller (heat creates enough movement in the P and N junctions so much so that electrons stop flowing but just fill the voids in both junctions) causing the loss of functionality of a transistor
On Nvidia GPUs at least the colder you get the less juice you need, even more reason to keep them cool
Just check/benchmark actual performance before and after to see if it is even worth it.
In my experience with Vega64, RX480 (both reference) and my 6900XT (Powercolor Red devil) overclocking doesn’t win me much at all, even with raising power limit. Like… 1-2 percent or less. Well, maybe a bit more on the 6900, but that card doesn’t really need it yet, and its certainly not going to be the difference between playable at X level of quality vs. not.
Boost algorithms today are pretty good.
Now, if you give the card better cooling - ramp fans, liquid cool, etc. (or as recent buildzoid experiments show, better power filtering)… that will net you tangible gains.
In terms of drawbacks? Without taking noise, power consumption, etc. into account… Every crash you have… is it your overclock? Is it something else? Just how sufficient/comprehences was your stability testing? Did it include the exact workload you crashed in?
2c. In my view it is pointless unless you cool the card better.
edit:
where the 6900 would/does see gains is in memory overclocking. and the slider maxed out is some tiny overclock. lol.
As far as I’ve heard the true contributor to degradation are thermal cycles. Voltage comes only into consideration when it has been increased.
Yeah, that’s what I’m planning IF I’m overclocking it. At the moment, I wanted to have the discussion if there are any downsides.
True
So your reasoning is that even without increasing the voltage the card would run hotter and consume more power? Why exactly is that so?
Again, if the voltage is not increased additional cooling shouldn’t be a consideration.
Sure, thermal cycles do alter the structure of the crystals over time.
That’s what I meant to point out, but I kinda failed at it I guess haha Sure, stock voltage is gonna leave more life into a component on the long run.
I disagree with your whole post.
Overclocking might not be as usefull on reference cards but the RX480 was a very good overclocking gpu especially when tinkering with memory timing.
Source: my pretty popular bios https://forum.level1techs.com/t/fawkes-golden-rx480-580-upgrade-bios-xfx-gtr-s-v2/
Yes, the cards are already on the edge with boost at stock voltage. And heat isn’t necessarily the only limit to stability. And heat isn’t purely controlled by voltage - higher clocks (or higher sustained clocks) = more heat at same voltage.
That is not my experience. Recent cards: they’re already on the edge with boost algorithms. Which means if you want more, you’re better off just cooling the card better (and boost will work better) before trying too hard to overclock things. Or raising the power limit.
I’m not saying there are no gains to be had. I’m saying that when you have a crash (and you will, because drivers crash as well due to the nature of complexity vs. stability) - you simply won’t know if its a general driver crash or caused by your overclock.
There’s not enough headroom there to make it worth it. In my opinion. YMMV, but again, I’d suggest focusing on better cooling first would be my priority.
If you’ve already got a high end AIB partner card with an overkill cooler, go nuts - modify the fan curves to cool it better, and then play, etc. I’ve done that with my 6900XT Red devil and there’s not much to be had without messing with voltages or power limits. You’re talking barely out of margin of error on Firestrike or Time Spy vs. default behaviour.
But there’s not a heap to be had without increasing voltages or power limit and that means even more heat.
If you’re talking anything less than 10-15% I do not consider that worth the uncertainty over driver crashes.
Thanks, that’s news to me. I thought this was purely related to voltage and that from a temperature perspective there shouldn’t be a difference if it runs 1900MHz @1V or 2000MHz @1V. Do you know the reason for this?
- My experience is that the limiting factor was always the power limit. If the voltage for a clock speed was reduced a higher clockspeed could be hit, e.g. when previously 1900MHz @1V hit the power limit and I set 2000MHz @1V as a new curve the latter would be reached because of the decreased voltage.
Well, I get your point but one could argue that this are averages of averages which kind of falsifies the gameplay experience. What if the average influence the lower fps higher? Then the perceived performance increase would be worth it. I would even argue it might be always worth it because it is free performance.
However, you argued above that the temperature would also be increased which influences my reasoning a bit.
Anyway, out of curiosity I went ahead and increased the clock further which is detailed here:
I just found another thread in a different forum that agrees with your statement:
Well, the card is running harder (And drawing more power - check your metrics for power draw).
I suspect that the gpu spends more time with parts of it idle (and thus it shuts those parts shut down) if the clock is slower. i.e., not every part of the GPU is drawing power at the same time and thus, if part of the die is not powered it can cool down while idle.
Remember, voltage is not power. POWER is what generates heat, voltage is just how hard the current is pushed through the component, it’s only one component of the power consumption (simplification: Power and thus heat = volts x amps basically).
If you raise the clock, the time between “on” and off for everything on the GPU shortens and the heat has less time to drop off.
I know that Zen CPUs do this very aggressively, I very much suspect any modern processor (gpu or CPU) from the past decade does this to varying degrees.
This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.