Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Vadanarika]

The AMD Ryzen 7 9800X3D is releasing soon this month, and I am around the same time planning an upgrade, and wondering whether or not I can expect better performance from the latest chip, or if I should just go with an older chip.

I remember a while back it being speculated that the primary performance problem with Dwarf Fortress could be made better with large fast CPU cache, which is what the X3D series of CPUs has. I have seen a few anecdotal reports from people on the forums and subreddit that the X3D chips do give better performance, so I am planning to go with one of them, but I am unsure if the latest one will be a noticeable difference versus say the Ryzen 9 7950X3D or the Ryzen 7 7800X3D. There are some people and reviews that have benchmarked Dwarf Fortress world generation, and there doesnt really seem to be that big of a difference (though the difference in factorio is quite big) between the three. However, I also recall putnam a while back once saying that world generation is a useless benchmark due to some anomalies in the music and language generation process, so I don't know if I should take those benchmarks into consideration or not?

I like to make large megaforts spanning many Z-levels with big embarks in multiple biomes, mess with water stuff as well, if that is relevant. Big population is nice too. So DF performance is certainly relevant to me.

[jecowa]

I don't have either, but I wish I did. I doubt that the 9800X3D is all that much better than the 7800X3D for Dwarf Fortress, but the 9800X3D is probably the best Dwarf Fortress CPU in the world, and when it's in-stock, it's about the same price as the 7800X3D, so seems like might as well get the new one if it's available. 7800X3D is build on TSMC's N5, 9800X3D is built on N4. They both have the same amount of cache. The new one has higher RAM speed. Maybe in a couple generations, AMD will move on to N2, and hopefully the backside power delivery will allow AMD to stack more V-cache on the CPU without it overheating.

I'm curious about Putnam's thought on an ideal benchmark. I think world generation is bad just because it's not representative of the majority of play. Who cares how long a world takes to generate unless you plan on doing a lot of world generation. Instead, I would find a good fortress save that causes performance issues, but not due to door problems. Then turn off auto-save, any announcements that open dialogues or pause the game, turn on the pause-on-save-load, load the save, then unpause and time how long it takes for a certain amount of in-game time to pass.

Then do something similar for adventure mode, but at a site that has a high population, since that's probably the biggest issue an adventure mode player will run into.

[jecowa]

This Factorio benchmark is showing 20% improvement in 9800X3D performance over 7800X3D, so maybe it is a big improvement. https://factoriobox.1au.us/results/cpus?map=4c5f65003d84370f16d6950f639be1d6f92984f24c0240de6335d3e161705504&vl=1.0.0&vh=

Factorio is another simulation game with low graphical load.

[Putnam]

7800X3D is very, very good for Dwarf Fortress, to the point that I've actually been a little hamstrung by just how good it can be. I got sent a save that was chugging at 20 FPS and found, to my chagrin, that it was running at 120 FPS on my computer, though that's also because their own CPU was over a decade old. I can run 400-dwarf fortresses at 40-50 pretty easily.

High-population site in adventure mode is the ideal benchmark for "endgame performance". Adventure mode sites tend to be a worst-case scenario for performance on that front. I've seen a city that runs at something like 0.05 FPS just because there were a couple thousand units, presumably all stuck into the same room, somehow. So, like, it doesn't fix all the potential issues. Really, that counts as a bug.

[MaxDemone]

This is only barely related but I just picked up a Ryzen 9 7900x and woof does it kill it at Dwarf Fortress compared to my old i7-7700k 

[Mungrul]

This is only barely related but I just picked up a Ryzen 9 7900x and woof does it kill it at Dwarf Fortress compared to my old i7-7700k 

Coincidentally, my upgrade last year was from and to the same chip, i7 7700 to 7900X3D, and I love it .
It really extends time to FPS death.

In gamer circles, people get a bit sniffy about the 7900, claiming the 7800 is better for gaming, but the differences are negligible, and I feel for games like DF, the 7900 is probably better.

[sodafoutain]

In gamer circles, people get a bit sniffy about the 7900, claiming the 7800 is better for gaming, but the differences are negligible, and I feel for games like DF, the 7900 is probably better.

The 7900X3D is strictly speaking a worse chip than the 7800X3D, because the 7800 version has on-par performance in most games and is cheaper. However, your intuition is right. For a simulation game like DF, you'd be looking to get a chip with higher core count and higher base clock. Has anyone ever conducted any !!SCIENCE!! on overclocking and it's affect on performance?

I watch quite a bit of GamersNexus, so I'd be interested in figuring out a benchmarking method for this game.

[A_Curious_Cat]

I watch quite a bit of GamersNexus, so I'd be interested in figuring out a benchmarking method for this game.

Well, for starters you could take a look at Putnam’s post above…

[Putnam]

In gamer circles, people get a bit sniffy about the 7900, claiming the 7800 is better for gaming, but the differences are negligible, and I feel for games like DF, the 7900 is probably better.

The 7900X3D is strictly speaking a worse chip than the 7800X3D, because the 7800 version has on-par performance in most games and is cheaper. However, your intuition is right. For a simulation game like DF, you'd be looking to get a chip with higher core count and higher base clock. Has anyone ever conducted any !!SCIENCE!! on overclocking and it's affect on performance?

Higher core count might help, but the only multithreaded thing in the game is unit line-of-sight processing, which involves grabbing a lot of 2.5 KB structures and checking bits of them, which can pretty easily cause cache misses, which will throttle any gains from threading given time; and of course, while it is the slowest "guaranteed" part of the game that's multithreaded, it is just that part, which means Amdahl's law rears its head real quick. There's also the whole "differently-powered cores" thing going on that might cause weirdness in perf, which I got a 7800X3D in particular just to avoid, haha. Cache is probably the strongest way to get more performance out of the game due to just how much of it is big array access stuff, though.

Higher clock speed only really matters for apples-to-apples comparisons, i.e. within the same CPU class, as it has for the last 20 years.