Best CPU available: AMD Ryzen 9 7950X3D?

[NineNine]

Am I correct in assuming this is the best CPU available today to run Factorio?

From what I've been able to put together, it seems that the CPU cache is the bottleneck for Factorio, and the AMD Ryzen 9 7950X3D has the largest on-chip cache on the market right now:
Total L1 Cache: 1MB
Total L2 Cache: 16MB
Total L3 Cache: 128MB.


And it looks like the #2 is AMD Ryzen 9 7900X3D:
Total L1 Cache: 768KB
Total L2 Cache: 12MB
Total L3 Cache: 128MB

Am I correct here?

(I have a late model i7, but I'm at 30-40 UPS right now, so the factory cannot grow! The factory must grow!!!!)

[Loewchen]

Both of these processors have their cores on two chips, but only one chip gets the extra cache. For the performance to improve the OS' scheduler would have to run factorio on the chip with the extra cache, but windows seems to always run it on the cooler/higher frequency cores (unless amd/microsoft changed that recently) which will always be the cores without it.
I'd wait for the 7800X3D that costs less and has all cores on a single chip with extended cache, eliminating the problem altogether.

[shopt]

https://www.youtube.com/watch?v=DKt7fmQaGfQ&t=641s

Ultimately I agree with Loewchen. The 7950X3D is the best CPU you can currently get for Factorio, *if* you disable the CCD without the extra cache (which HUB did in that linked video). However the 7800X3D will be released soon and will gave practically the same performance as the 7950X3D at a much cheaper price, and without needing to do hacks like disabling half of your very expensive CPU.

[jodokus31]

And for AM4 folks, the 58003XD is still worth considering, although it's a bit slower than those new 5nm models

Checkout these benchmarks on factoriobox:
https://factoriobox.1au.us/results/cpus ... =1.0.0&vh=

[NineNine]

The 7800X3D releases in two weeks! Where does somebody buy a CPU (and a motherboard to go with it)? I've never had to build a computer before, but..... the Factory Must Grow!!!

[azesmbog]

Without detracting from the merits of the Ryzen 9 7950X3D, I suggest looking at the results of processors in heavy maps:
https://factoriobox.1au.us/results/cpus ... 9605e8d509
Who has 100 UPS??

[FuryoftheStars]

Without detracting from the merits of the Ryzen 9 7950X3D, I suggest looking at the results of processors in heavy maps:
https://factoriobox.1au.us/results/cpus ... 9605e8d509
Who has 100 UPS??

Actually seeing the results between those two, it makes me wonder: how is the 75th percentile calculated for the results?

[robot256]

Remember that while the vanilla Factorio engine is heavily optimized to minimize cache misses, the Lua interpreter that runs mod code is not. Large, heavily modded maps might not see any improvement from Vcache at all, so physical RAM latency is still important.

[posila]

while the vanilla Factorio engine is heavily optimized to minimize cache misses

Oh, that is very flattering, but also (as I strongly believe) very untrue statement

[azesmbog]

So we are talking about really large (not modified) maps.
The factory must grow!
When the map completely fits into the processor cache, there is an undeniable advantage for 3D AMD, but as soon as the map size is larger than the cache, then Intel slowly, but comes forward. Of course, the frequency of the processor and memory itself is also important.
So you have to choose - but carefully. but choose

[FuryoftheStars]

That should be fairly easy to do (exceed a proc's cache size). A simple 100x100 chunk area, in tile id alone (absolutely no other entities, not even decoratives), uses almost 10 mb of memory.

But really (unless my knowledge on how all of this works is wrong/incomplete, which it likely is), unless you found a way to play the game with absolutely nothing else vying for processor time, not even the OS, you wouldn't have the full proc cache available to you at all times, anyway.

[jodokus31]

Without detracting from the merits of the Ryzen 9 7950X3D, I suggest looking at the results of processors in heavy maps:
https://factoriobox.1au.us/results/cpus ... 9605e8d509
Who has 100 UPS??

But, the Intel processors draw a lot more power compared to the the AMDs according to some tech reviews. Additionally, I can undervolt my 58003XD by -0.075, without losing performance

[FuryoftheStars]

Without detracting from the merits of the Ryzen 9 7950X3D, I suggest looking at the results of processors in heavy maps:
https://factoriobox.1au.us/results/cpus ... 9605e8d509
Who has 100 UPS??

But, the Intel processors draw a lot more power compared to the the AMDs according to some tech reviews. Additionally, I can undervolt my 58003XD by -0.075, without losing performance

According to what I can find, the 13900k draws 125W base and 253W turbo, whereas the 7950X draws 170W base (+45W) and 230W PPT (-23W).

That said, if you're concerned about power, you can downscale them and it seems like the AMD will come out better.

https://www.anandtech.com/show/17641/li ... 900k-7950x

But I'm not sure if I'd want to pay for a top end processor to then downscale its performance...?

[jodokus31]

When the system is under full load, the AMDs seems to be a more energy efficient than Intel, at least according to those tests:

I had a german source, which is this:
https://www.computerbase.de/2023-02/amd ... 3d-test/3/

But this (english) is another under "Power Consumption"
https://www.techspot.com/review/2636-amd-ryzen-7950x3d/

Interestingly enough, last one also test Factorio and the result seems to differ slightly from factoriobox samples. They probably used a rather small map.

[FuryoftheStars]

Ah, the X3D specifically has a -50W base draw compared to the plain X, but this seems to put it as only slightly better than Intel in power.

That said, various benchmarks do show either of the AMDs as being better performance wise (though the Intel KS isn't that far off and is better than the K (and appears to be best in single thread performance)).
https://www.cpubenchmark.net/compare/50 ... -9-7950X3D

[NineNine]

So it looks like the answer is , "Yes, the AMD Ryzen 9 7950X3D should be the best CPU available today for Factorio in most situations."

Thank you fellow People of the Internet!

[shopt]

But this (english) is another under "Power Consumption"
https://www.techspot.com/review/2636-amd-ryzen-7950x3d/

Interestingly enough, last one also test Factorio and the result seems to differ slightly from factoriobox samples. They probably used a rather small map.

They use the "default" map from factoriobox, which IIRC is a 10k SPM base made by flame_sla.

The other thing to consider is that with professional benchmarkers, they are good at controlling for most things and giving an apples to apples comparison. Community benchmarks from factoriobox will be larger in quantity, but much noisier due to things like selection bias, different other non-CPU hardware, people just using different settings, etc.

[SoShootMe]

Both of these processors have their cores on two chips, but only one chip gets the extra cache. For the performance to improve the OS' scheduler would have to run factorio on the chip with the extra cache, but windows seems to always run it on the cooler/higher frequency cores (unless amd/microsoft changed that recently)

There is a huge difference (74% higher UPS) going by the Techspot Ryzen 9 7950X3D review mentioned by jodokus31 (first one under Gaming Benchmarks). From a quick look on FactorioBox this seems to be more than all the (impressive) gains since 1.0.0.

It's impossible for the OS to reliably figure out that the best thing to do is restrict a process/some thread(s) to the cores with extra cache available. The best case scenario is that OS specifically detects Factorio (obviously not a scalable approach) or that the scheduler implementation is changed in a way that happens to benefit Factorio, such as using additional hardware performance metrics (if they exist) or using them differently ("Hmm, cache misses are high; I'll try it over here for a bit and see if that helps"). The only guaranteed solution is direct intervention, ideally by Factorio itself, but:

I'm going to make an executive decision and say the operating system(s) are responsible for scheduling things to run on cores and not software (us).

Simply put, I think this is becoming an increasingly untenable position.

[FuryoftheStars]

https://www.techspot.com/review/2636-amd-ryzen-7950x3d/

Interestingly enough, last one also test Factorio and the result seems to differ slightly from factoriobox samples. They probably used a rather small map.

You know, as I'm checking out the results from this page, too, I'm actually a little surprised by them in the Intel 13900K vs KS results. Unless I missed something somewhere in the specs, I thought the KS was the same as the K but with a slightly higher per core clock frequency, yet while the KS does do slightly better than the K in the majority of their tests, the K actually outdid the KS on Factorio.

[quyxkh]

It's impossible for the OS to reliably figure out that the best thing to do is restrict a process/some thread(s) to the cores with extra cache available.

I wouldn't go that far. Maybe nobody does it, but I'm not sure even about that - - AIX would notice when pairs of processes interfered with each other and schedule them into different cache domains, it's not hard to imagine an OS being taught to set up the performance counters so as to track cache-miss-per-retirement stats and schedule processes where they get the biggest wins.