Print

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

[Tiny]

Intel i5 2500k, as the intel i7 2700k has hyperthreading, which actually makes performance worse on games that don't utilize multiple cores.

You can turn hyperthreading off in the bios, It's not a huge difference but it helps a little.

For maximum DF, I'd go for one of the old high clocked single core AMD chips, if you can still find them and the motherboard to go with it, then spend what you save in some very dwarven overclocking (I don't think you can cool a CPU with magama but there is still engineering to be done).

[Putnam]

...Why does it seem like this thread is so heavily under the megahertz myth?

The thought that a 3.0 GHz Pentium 4 is any faster than a 2.0 Ghz i5 is laughable.

[janglur]

Well, older isn't necessarily better just because it has a higher frequency.  Again, it's IPS (Instructions Per Second), not sheer frequency, that matters.  A newer CPU (not budget or mobile) will have higher IPS.  If this weren't the case, we'd all still be running 486's, except at insane frequencies.  But in truth, FAR more improvements have been made beyond core speed to ultimately result in better performance.

Optimally you want a newer-gen technology, idealy 32nm or less (if they're out yet).  Second most important is the frequency, higher is better.  Gimmicks like Intel's Turbocore mean nothing if they don't engage, so read up carefully on what is required for them to engage, or you'll be wasting a LOT of money on a feature you can't utilize.  Nextly, go for the highest cache size you can-  this can make a big difference on older forts with a huge amount of data to track, namely item clutter.  And finally, dual core or better helps:  Dedicate DF to one core, and force everything else to the other core.  This should give a good 5-30% boost, depending on how much garbage you have running in the background.

Outside of the CPU, go for the highest frequency and especially lowest-timing memory available.  You don't need a lot, but fast memory greatly reduces the amount of time DF takes for certain tasks, especially invasions, caravans, generating content, weather, flows, and stocks screens.

Finally, make sure the HDD DF is on isn't utterly overburdoned, because DF can trash the HDD a bit while saving or loading.  If you don't mind the load/save lags, you can skip this step, as it really doesn't matter during any other time.

Also, avoid integrated video, as this robs your CPU and especially RAM of resources, which can impact DF.  It's only major in a few isolated cases, but every little drop helps when you breach HFS with 250+ dwarves on a 6x6 map with temperature and weather on.

Everything beyond these is either superficial (too little to notice or measure) or too circumstantial to even consider.



Edit:

Oh, also, avoid hyperthreading.  It's not very useful in any situation, and is actually detrimental to single-threaded apps like DF.  Long story short, this is because it attempts to process more threads in the same core space, which is only really useful in the case of many very small thread functions that require little cache or RAM support.  In other words, very little in real use.

[Tiny]

Well, older isn't necessarily better just because it has a higher frequency. 

Recent advances have mostly been in adding cores, the generation of chips from just before multicore took off, had quite similar IPS (per core) but could hit higher clock speeds and could often be pushed quite a bit further with good cooling. Single thread performance hasn't changed much in the last few years, some but not alot.

[lordnincompoop]

What you'll want is most likely either the CPU that can offer the most bang-for-the-buck disregarding cores, or the most performance disregarding the same. Off the top of my head, I'd recommend the i7 2500k overclocked as much as possible, paired with a good set of RAM, but others more in-tune with the details of CPU specs may be able to tell you more.

And what does Toady use?

Toady uses a Lynnfield something-or-other from what I remember. Nothing optimal.

...Why does it seem like this thread is so heavily under the megahertz myth?

The thought that a 3.0 GHz Pentium 4 is any faster than a 2.0 Ghz i5 is laughable.

Agreed.

Well, older isn't necessarily better just because it has a higher frequency. 

Recent advances have mostly been in adding cores, the generation of chips from just before multicore took off, had quite similar IPS (per core) but could hit higher clock speeds and could often be pushed quite a bit further with good cooling. Single thread performance hasn't changed much in the last few years, some but not alot.

Your researching skills are a little rusty, aren't they?

Sandy Bridge, for example, offers a 17 percent performance advantage clock-for-clock than a last-generation Lynnfield. Ivy Bridge (link in "Sandy Bridge") will offer a 20 percent increase over Sandy. Going through several generations and taking into account that a GHz is 10^9Hz (each represents a single clock cycle, though I'm a little fuzzy on that), the difference in performance will add up.

[janglur]

Well, older isn't necessarily better just because it has a higher frequency. 

Recent advances have mostly been in adding cores, the generation of chips from just before multicore took off, had quite similar IPS (per core) but could hit higher clock speeds and could often be pushed quite a bit further with good cooling. Single thread performance hasn't changed much in the last few years, some but not alot.

No, but a 20% improvement in single-threaded performance (such as Athlon64 to Phenom II) is still a significant enough amount to warrant spending $160 on a new CPU rather than $90 on an older one that will perform drastically more poorly overall.



Edit:
What that guy said.  =D

Also, cycles/second (Hz) is not necessarily the same as IPS (Instructions Per Second).  Most CPUs require 2-3 cycles minimum per instruction, and exponentially more for more complex instructions.  When you get past single-instruction executions, this adds up A HELL of a lot.

it's the difference between adding 2+2 a billion times a second, compared to mapping a floating point grid and making trajectory calculations 1,000 times a second, and the end result of work:time ratio for each.

[Tiny]

Well, older isn't necessarily better just because it has a higher frequency. 

Recent advances have mostly been in adding cores, the generation of chips from just before multicore took off, had quite similar IPS (per core) but could hit higher clock speeds and could often be pushed quite a bit further with good cooling. Single thread performance hasn't changed much in the last few years, some but not alot.

Your researching skills are a little rusty, aren't they?

Sandy Bridge, for example, offers a 17 percent performance advantage clock-for-clock than a last-generation Lynnfield. Ivy Bridge (link in "Sandy Bridge") will offer a 20 percent increase over Sandy. Going through several generations and taking into account that a GHz is 10^9Hz (each represents a single clock cycle, though I'm a little fuzzy on that), the difference in performance will add up.

I'd say in this case my reaserch skills were totaly unused, not rusty, I was just blindly passing on bad information :facepalm:.

I've got a 2.6GHz i7 OCed to 3.6 that works nicely for DF

[helf]

This view of hyperthreading being detrimental is left over from the netburst era. Hyperthreading on Intels new chips actually works and works well.

Turning it off will remove some chatter on the memory bus, which MIGHT help some if you don't have fast ram or ram with lower latencies. It will probably have a negligible impact, though.

[xordae]

...Why does it seem like this thread is so heavily under the megahertz myth?

The thought that a 3.0 GHz Pentium 4 is any faster than a 2.0 Ghz i5 is laughable.

This, pretty much.

Just get a recent CPU which has good performance-per-core. I really would not buy a CPU to cater to Dwarf Fortress. It has yet to be thoroughly optimized, and it is not representative of lots of other games which can utilize 2 or 4 or more cores. It feels like 4 is the sweet spot at the moment. You'll run into a FPS wall either way, depending on what you do. And I'm still playing comfortably around 30 FPS with 230 dwarves and weather and temperature active, on a stock Q9550. I don't mind that pace at all.

[janglur]

Furthermore, performance between versions for even minor updates are very drastic.  So what may be optimal for 31.25 may be utterly terrible for 32.15.  So, it's highly reccomended you just get an all-around good CPU and deal with not being in the top 1%.