Does extra L3 cache in CPU affects performance in Paradox games?

[Furleppe]

In the near future I'm probably buying a new Ryzen 5800X3D, and it got me wondering. Does extra cache (main feature of this CPU) affects performance in any meaningful way in PDX games? Or maybe it doesn't matter as much as single-threaded performance? I don't know much about this topic, and it would be interesting if someone who does, had an answer for this question.

 

[BlackCobra]

my best guess is that should have a good impact on performance. There is a short benchmark run of a 5800x3d with factorio and there it really shines, +50% over 5800x and 10% over 12900k.
Techpower may be doing a benchmark soon with Stellaris (https://www.techpowerup.com/forums/...le-would-like-in-the-next-cpu-rebench.293862/)
But since today should be the lifting of reviews embargo, there should be coming out new reviews of it, hopefully also with games which may actually benefit of it.

 

[rubert]

If anyone gets 5800X3D and runs some benchmarks on Paradox games I definitely would be interested to hear about them, especially about Stellaris how fast days progress in the late game. I could upgrade my 3800X to it but the prices seem pretty expensive least here.

 

[AndrewT]

Our hardware genius @jpd might have some useful input on this!

 

[jpd]

I can only speak in generic terms, of course, not having access to either CPU so I don't have first hand data.

A (CPU) cache only provides it's benefits if the code and/or data actually resides inside the cache. In other words, if the algorithms you run are small enough to fit inside the cache, either in whole or, at the very least, partial.

If you run a procedure (a piece of code) from start to finish that's so large it doesn't fit inside the cache, then rerunning that procedure for the second time will run as if it wasn't cached at all. Because, by the time you reach the end of the procedure, the code at the start will have been cleared from the cache to make room for the bit in the middle and the end. And running the bit at the start will then have pushed out the bit in the middle (because that, now, is the oldest), and running the bit in the middle will have pushed out the bit at the end (because that, again, is now the oldest). Effectively, running such a procedure for a second time continuously needs access to the next bit of code that just has been cleared from the cache to make room for the bit that's running now. It's chasing it's own tail, so to speak.

The same applies to data. If the code you run manipulates a pool of data that's too large to fit in the cache, and it does it repeatedly on that data in the same order, then the data manipulated first will have been pushed out of the cache by the time your second loop starts.

I'm not a coder for Paradox, so I don't know how exactly the game's algorithms run, or how (much) data is manipulated in tight loops (where the benefits of a cache really shine). So you probably have some benefits from a larger cache. But I can't say for certain, nor can I say by how much.

 

[grandsteed]

At minimum, Europa Universalis II benefits greatly from an L3 cache, running the game on a (Mac but windows is the same in this regard, just easier example) G4 at 733 MHz without an L3 cache is very choppy and stutters when many units or constructions are in play on the map, but a G4 at 733 with 2MB L3 is smooth and no stutter. Eu2 is 20 years old, and is almost exclusively reliant on CPU and RAM, but I do believe anything requiring cached calculations will be much better with more L3. Again, this is not the most relevant example given the age of the Europa engine, but it should be an indication that this style of game is affected by cache. Cache and RAM.

 

[Furleppe]

Thanks for the answers, in the end I did get the 5800X3D, CK3 and Stellaris do feel smoother, but that's purely anecdotal, and I have no knowledge on how to do proper testing.

But to those interested, I found this thread on Reddit, where user tested late-game Stellaris and the gains (compared to 2700X) seem to be significant:
5800X3D Stellaris benchmark

Update: And there, comparison with 5900X:
https://github.com/xxEzri/Vermeer/blob/main/Guide.md#stellaris

 

[rubert]

Interesting, that is quite noticeable so I think even if upgrading from 3000 series CPU the gains would be pretty big as 3700X compared to 2700X does not have that kind of performance difference in single-threaded applications.