Print

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

[G-Flex]

That's not what I said.

I said that it IS theoretically possible to spread the load from a single process across multiple CPU/cores, but that it'll just alternate between them rather than running it on both at the exact same time.

[KaelGotDwarves]

That's not what I said.

I said that it IS theoretically possible to spread the load from a single process across multiple CPU/cores, but that it'll just alternate between them rather than running it on both at the exact same time.

The point I was trying to make is that DF is not run on multiple cores. Which is true and swapping DF from core to core doesn't in fact make DF run faster unless you're attempting to run many other programs at the same time as DF.

[G-Flex]

Yeah, and I'm sure there's some level of overhead involved.

If possible, it's probably best to just make sure a single core is available for DF and DF only. I guess.

[Nadaka]

It spreads it out, but NOT in the sense of executing it in parallel! It's just distributing the CPU load for reasons of heat and whatever else.


In other words: It's not making DF run on two cores at once, technically. It's running some on one core, then some on another, then some on the first, etc. So it's the equivalent of lifting boxes or filing papers and alternating which hand you use, as opposed to filing two papers or lifting two boxes at once (one with each hand).

yes, this. Most modern multiprocessor/multicore systems will perform load balancing on their cores to prevent errors or defects caused by the concentration of heat that could occure if one core was pegged all the time while the others were idle.

[decius]

Also, there is some load that can be offloaded onto a video card. I noticed a significant jump in performance going from no video card to having a mid-grade one.

[Quatch]

the 40d# series has some openGL speedups.

RAM is pointless as a limiting factor, DF takes almost none.
CPU on one core is the limit. an openGL capable GPU can take some, but not (or even close) to all of this.

[nitnatsnoc]

After a certain point, RAM has very limited returns. Provided you have enough RAM for your OS and all the other current applications running then DF will probably not need more than 200-300mb except under the most taxing circumstances.

Unless you are using a computer with less than 1gb of RAM, a faster CPU is, under almost all circumstances, a better choice.

A basic RAM guideline for the average Microsoft based PC is as follows:

Windows XP/2000 - 1gb
Windows Vista/7 - 2gb

I have no idea for Mac & Linux but I imagine you could get away with a bit less. That said RAM is so very cheap for the most part, and even the most basic of new PC's come with 2gb or more it seems these days.

For a CPU, I guess you want to maximize overall clock speed. So for Intel this would be the Extreme Edition CPU's or for AMD the Black Edition CPU's. Overclocking could likely get you another few FPS, but if you don't know what your doing then I don't recommend it.

[G-Flex]

One thing people are forgetting here is that RAM gets piss-easy to upgrade.


If you have 2GB taking up two slots, and have two slots free, you can just go grab another 2GB later. And even if you need to REPLACE what you have, usually it's not super-expensive.

[Jay]

I have never seen DF hit anything more than a gig of RAM, ever.
If you have 2GB of RAM on XP or 3-4 on Vista/7, you're fine on RAM.

CPU cycles are by far the most strained resource for DF.
And they're also not getting very much more abundant as we advance anymore; everybody's adding more cores, not more raw processing power per core, and DF can only use one core.

[G-Flex]

New architectures are still, for the most part, better. The megahertz myth is, well, a myth.

[Thief^]

Megahertz is like engine litres. A higher number is normally better, but other efficiencies make a lot of difference. e.g. in a car these are things like fuel injector, turbo, superchargers, multiple valves on cylinders, cylinder layout, even the fuel used and accuracy of piston fit to the cylinder. On a cpus it's things like branch prediction accuracy, branch prediction failure stall time, cache size and speed, pipeline length, out-of-order execution ability, extension support (SSEx), number of hardware threads (aka hyperthreading), speed of bus, speed of memory controller, and dynamic frequency scaling response time (to name just a few). These things only quantify in obscure stats. In engines' case, you get horsepower / torque curves. In cpus' case you get MIPS or GFLOPS. Even then the actual performance depends on what you're doing (a high horsepower/torque engine won't win you any races if it is in a massive truck).

The pentium 4 is a good example of a cpu which went for all-out megahertz speed at the expense of everything else, most specifically pipeline length (far too long) and branch prediction failure stall time (also far too long), which resulted in it getting outperformed by cpus with 2/3rds the megahertz speed (AMD Athlon XP).

These days it's a closer race, but megahertz is getting more and more useless as a performance measurement. Pretty much every cpu is around 2.5GHz these days, and some really suck. Look at benchmarks.

[Nadaka]

New architectures are still, for the most part, better. The megahertz myth is, well, a myth.

For a given architecture, higher mhz will be faster for dwarf fortress. Its not a myth. An i7 920 will be slower than an i7 975. An Athlon II x2 420 will be faster (for dwarf fortress) than an Athlon II x3 435.

[G-Flex]

New architectures are still, for the most part, better. The megahertz myth is, well, a myth.

For a given architecture, higher mhz will be faster for dwarf fortress. Its not a myth.

Well, yeah, obviously if all other things are equal, higher clock speed is better.

The myth is "higher clock speed = better CPU performance" as a general statement, which, as Thief^ elaborated upon quite well, is not always true.

[CobaltKobold]

RAM is usually present in sufficient quantities- if you're doing a steep 16x16, you'll need the gigabytes. Sitesize is more likely to eat your RAM than anything else.

CPU speed is more important- creatures running around.

[KaelGotDwarves]

New architectures are still, for the most part, better. The megahertz myth is, well, a myth.

 An i7 920 will be slower than an i7 975.

Except intel gimped the die on the new i7s and you can overclock 920s that much more than the 975s. They got sick of everyone making a $200 chip faster than their $1000 EXTREMEEEEEEEEEEEEEE edition. limited 8x pciE on sli/crossfire too but that doesn't really affect DF