For much of the 2010s, we were stuck with mainly dual-core and quad-core CPUs in PCs. However the arrival of Ryzen shook the PC industry, causing a rapid increase in core counts. At the time, there was fervent discussion on this matter, with many questioning if more cores were worth it, and how many cores are more than enough?
So how do things stand today? The latest Intel and AMD consumer processors top out at 24 and 16 cores respectively. What extent of modern software can take advantage of all those cores? What modern workloads are still bottlenecked by single threaded performance?
People keep saying this, but does it tho? I’ve seen multiple benchmarks where 8 slower cores are faster than 6 high speed cores. I would say cache is more important that single threaded performance in 2023.
That just isn’t true. I don’t know what benchmarks are you watching but you won’t find a single game where Ryzen 7 3700X is faster than Ryzen 5 5600(X).
Same goes for 5800X vs 7600X. On intel’s side it’s harder to compare because higher core count CPUs have more cache.
https://www.youtube.com/watch?v=cKgDrW5H5go&t=475s
Really? Gaming benchmarks for Ryzen 7000 seem to be pretty much flat across the board, with minimal gains upwards from the 6-core 7600. Which games are you thinking of?
Main Threads in games are still limited by single core performance.
Main Threads in games are still limited by single core performance.
Cache improves single threaded performance in gaming though.
One thing that gets constantly overlooked in these scenarios is the fact that 8 core CPU’s have more L3 cache than 6 core CPU’s. so if a game uses 6 threads, an 8 core CPU of the same architecture with the same clock speed will potentially perform better than it’s 6 core counterpart.
* only on Intel, which has the L3 made out of slices attached to each P-core or E-core cluster (x4).
AMD segregates its L3 at the CCX level, so every part made from the same die set has the same L3. There’s a bit of a complication with the 12 and 16 core, because if all the threads are working on the same data the L3 is effectively 1-CCD-sized, but if they’re working on different data (like with
make -j, VMs, or some batch jobs), you get the benefit of both CCD’s worth of L3.That’s only true for intel, who disable part of the L3 along with the cores. AMD however has the full L3 enabled on their 6-cores.
The marketing folks love to add up the L2 cache as well, but since that is not shared cache, each core still has the exact same amount of cache available to it, in spite of having a lower number on the spec sheet.