IIRC they hit a wall of physics trying to increase the GHZ, so why didn’t they just keep increasing the number of cores?
Because there are practical limitations to how many cores you can throw at most tasks while seeing any meaningful benefit. Not everything needs that level of parallelism - particularly gaming - and the wider you try to make a piece of software the more time and effort needs to be devoted to reconciling it back into a single, coherent task to present to the user.
Consider a physics simulation. If you run a single, master physics simulation it can do all of its comparisons as it goes. If you run 16 different ones you now have to do 120 different comparisons to verify that something being moved by one simulation isn’t being affected by something handled in a different simulation. 64 different simulations at the same time is 4,160 comparisons. There are ways to separate this into areas to keep potential cross-interactions to a minimum, but that’s going to require additional pre-processing time for every additional core.
This is all before the physical constraints. More cores means a more expensive processor, means more cores competing for the same bandwidth (or a more expensive system to allow more bandwidth), more heat that needs to be dealt with, and reduced yields for the whole thing as complexity rises. Home users don’t tend to have server budgets to throw at hardware.
1 Like
Few games are designed to utilize more than 8 cores. A CPU like that would be best for virtual machines, rendering, and encoding. I use Handbrake to encode many of my videos to x265 and AV1 codecs. It uses however many cores you have. Games that truly could utilize more cores are TBS and RTS with massive maps and 50+ computer players to manage. I’ve played Galactic Civilizations III with such a configuration. Each turn involves waiting for 10+ minutes because the system is processing each computer player’s moves. You’d also need about 64GB of memory for smooth gameplay.
So the point is that few games have features that would use more cores.
And as the person above said. The more core support, the more complicated the programming becomes
1 Like
I mean, you can build yourself a Threadripper system, but it’s going to be incredibly expensive for not much benefit in gaming.
A friend of mine uses a Threadripper for his personal AI system
-Threadripper 9980X
-ASUS Pro WS TRX50-SAGE WIFI A
-3x GeForce RTX 5090 32GB GDDR7
Same reason as why we don’t have Petabytes of Storage yet.
Petabyte storage is being held back by the limitations of physics trying to get that sort of data density into the commonly accepted form factors. Using current technology, magnetic storage would need a bay about 28 times taller than a standard 3½” drive (or an 18½” form factor) to physically fit just 1PB, while an SSD of the same capacity would require 4,000 of the largest available NAND chips (465.75cm³, or nearly a pint in the “old measure”, of chips). And that’s just for the medium on which the data is stored - the corresponding circuity to access it and ensure integrity would also be significantly more sizeable.
1 Like