Ive recently been looking at a new graphics card for my gaming computer, and I see a lot of statistics like “1650 boost clock speed ,” But I don’t understand why the manufacturer wouldn’t just set the performance comparative to begin with?
In: Technology
There is such a thing as a so-called “silicon lottery”. No two chips are identical, and some are able to run faster safely, some aren’t. Temperature permitting the GPU will voluntarily overclock itself to boost clock speeds – and sometimes beyond – as long as it can take it. The base clock is the guaranteed speed.
CPUs can usually do it better since they have many cores which can be idle or busy. When most cores are idle there’s enough power (and heat) capacity available to let the busy cores speed up significantly while the idle cores basically shut down, keeping the overall power utilization within the limits.
It’s less clear cut with GPUs since you don’t have that kind of control over them, but as long as it’s safe they’re willing to speed up a bit.
The boost speed is the max frequency the chip will run by itself.
The problem is higher frequency will result in more heat production. It can also require higher voltage so more energy usage for the same work.
So normally you run it a t a lower base clock speed.
There can be another limit and that is the amount of heat that can be transported away. For a CPU the boos frequency depends on the number of cores you run so more boost if 1 core is used to the max but less if 2 and even less for 8.
The limitation than to be the amount of heat that can be removed from the CPU.
Exactly how to boost speed work for GPUs is not clear to me but is could be so that it too is limited by heat. So the clock might be faster for some load that does not use all part of the GPU.
In the CPU case it’s usually a power constraint. There’s a certain amount of power and dissipated heat a CPU can go through, and if a multi-core chip is mostly idle but has one program (technically, one thread) doing lots of simple instructions like adding two numbers together it can ramp up the speed while maintaining the same power limit. But if all cores are firing, or if some cores are doing complex instructions like adding sets of 16 numbers at a time (using technology like AVX) the power limit can be reached at lower clocks, so the boost clock can neither be guaranteed nor sustained. It’s basically opportunistic overclocking.
Latest Answers