I think that it depends a lot on what one is doing.
So, a lot of games are bound on single-thread performance.
I have a Ryzen 9 7950X3D in my desktop. That’s the blingiest desktop AMD processor from 2023. That’s a six-year difference between release of those two processors and moving from a midrange to a top-end processor.
It benches at less than twice the single-thread performance of the Ryzen 5 1600. It’s faster. But it’s not really transformatively faster. It used to be, in, say, up until the early 2000s, that you’d double serial computation performance every 18 months. A 6-year difference between processors, as between those two, used to mean that the newer one would run pretty much everything about sixteen times faster, even setting aside differences in the processor bin.
Parallel processing has improved at a better clip, either via adding more cores to CPUs or the massively-parallel computation on GPUs. So if software can really utilize parallel computation effectively, then one might get larger gains over that period. And for some software — and games are an area where some entrants can do that — they can take advantage. But for a lot of software, hardware just isn’t changing as quickly as it once did.
And for games, it’s very common that the way in which they can take advantage of more parallel compute is nice-to-have but not really essential ways, like bumping resolution up or adding some extra visual effects. It’s not “the game becomes unplayable because the game logic can’t keep up” or something like that, the way it typically would have been in the 1990s.
There are definitely things that one can do where parallel compute makes a larger difference. If you’re a computer programmer compiling software and your particular environment can do parallel builds, then you can often get a pretty linear performance increase in the number of cores. If you do 3D rendering or video rendering, you’re probably bounded by the CPU, and software is often written to take advantage of parallelism there. But the vast majority of software is mostly-limited by serial compute. And serial compute performance just hasn’t been increasing very quickly for quite some years.
I think that it depends a lot on what one is doing.
So, a lot of games are bound on single-thread performance.
I have a Ryzen 9 7950X3D in my desktop. That’s the blingiest desktop AMD processor from 2023. That’s a six-year difference between release of those two processors and moving from a midrange to a top-end processor.
But despite all that.
https://www.cpubenchmark.net/compare/2984vs5234/AMD-Ryzen-5-1600-vs-AMD-Ryzen-9-7950X3D
It benches at less than twice the single-thread performance of the Ryzen 5 1600. It’s faster. But it’s not really transformatively faster. It used to be, in, say, up until the early 2000s, that you’d double serial computation performance every 18 months. A 6-year difference between processors, as between those two, used to mean that the newer one would run pretty much everything about sixteen times faster, even setting aside differences in the processor bin.
Parallel processing has improved at a better clip, either via adding more cores to CPUs or the massively-parallel computation on GPUs. So if software can really utilize parallel computation effectively, then one might get larger gains over that period. And for some software — and games are an area where some entrants can do that — they can take advantage. But for a lot of software, hardware just isn’t changing as quickly as it once did.
And for games, it’s very common that the way in which they can take advantage of more parallel compute is nice-to-have but not really essential ways, like bumping resolution up or adding some extra visual effects. It’s not “the game becomes unplayable because the game logic can’t keep up” or something like that, the way it typically would have been in the 1990s.
There are definitely things that one can do where parallel compute makes a larger difference. If you’re a computer programmer compiling software and your particular environment can do parallel builds, then you can often get a pretty linear performance increase in the number of cores. If you do 3D rendering or video rendering, you’re probably bounded by the CPU, and software is often written to take advantage of parallelism there. But the vast majority of software is mostly-limited by serial compute. And serial compute performance just hasn’t been increasing very quickly for quite some years.
And the GPU is generally more important than CPU for gaming.