Isn’t most games still dominated by single thread performance?
(As mentioned another post here, for even more CPU-intensive games like Factorio, the difference is even starker.)
For general mixed workloads (assuming you need more threads), I think the 7950X3D is the way to go - 8 3D VCache cores for similar gaming, 20% cheaper than the 14900KS, generally neck and neck on development workloads but also has AVX512 support, and of course all while using significantly less power (also 120W TDP, 160W PPT - you can do even better w/ PBO+UV or Eco mode). Here's the OOTB power consumption comparison, it's a bit bonkers: https://www.techpowerup.com/review/intel-core-i9-14900ks/23....
It’s improved a lot, I still use Lasso but strictly speaking I don’t really need to.
Typically, those would be sequential algorithms with large memory needs and very random (think: hash table) memory accesses.
Examples: SAT solvers, anything relying on sparse linear algebra
Users might realize how many of their cores and cycles are being effectively wasted by limits of the memory / cache hierarchy, and stop thinking of their workloads as “CPU bound”.
I appreciate it’s a complicated and subjective measurement: Hyperthreading, superscalar, out-of-order all mean that a core can be operating at some fraction of its peak (and what does that mean, exactly?) due to memory stalls, vs. being completely idle. And reads meet the instruction pipeline in a totally different way than writes do.
But a synthesized approximation that could serve as the memory stall equivalent of -e cycles for perf would be a huge boon to performance analysis & optimization.
One thing to keep in mind is that while slightly higher idle power might cost you a bit more, high power when processing means that you will also need a much beefier cooling solution and have higher noise levels and exhaust more heat to boot. Again, from the TechPowerup review https://www.techpowerup.com/review/intel-core-i9-14900ks/22.... they show that for a multi-threaded blender workload, the 7950X3D sits at 140W, while the 14900KS hits 374W. I don't believe there's a single consumer air cooler that can handle that kind of load, so you'll be forced to liquid cool (or go even more exotic).
Different motherboards and settings are sort of a hidden factor in this in general it seems.
There is no such thing as idle in a modern desktop.
Intel Meteor Lake NUCs should perform similarly, but they tend to be a few hundred dollars more for basically the same performance (the Minisforum EliteMini UM780 XTX barebones is currently $440, the cheapest Core Ultra 7 155H minipc I could find was the ASRock Industrial NUC BOX-155H at $700). At this point though, personally, I'd wait for the upcoming Zen5/RDNA3.5 Strix Point APUs, which should be the next big jump up in terms of performance.
7950x3d is limited to ddr5 6000 while 14900 can do ddr5 8000.
The benchmarks you shared are using ddr5 6000. So by upgrading the ram, the 14900 should come out on top. The memory controller is a key part of the cpu. It makes sense to test them with the best components they are able to use. I know it probably doesn’t matter, but if you’re chasing those last 5% of performance fast ram is a better value than custom water cooling.
The user-mode video driver and kernel mode driver both use other cores as well