Print

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

[Grax]

we'll see a dynamic-scaling CPU capable of overclocking one core to above 8GHz at the cost of shutting down others, in about a year

Not more than 5Ghz with air cooling.
Overclocking over 3Ghz technologically impels exponentially higher energy consuming and heat dissipating.
So in my practice Core2Duo at about 3Ghz dissipates 65W as standard, overclocking to 4Ghz means 130W.
Maximum reached at 4.5Ghz and 170W.
5Ghz will theoretically devour and dissipate about 250W. This power is impossible to sustain by common motherboard and power units.

Intel processors' roadmap shows us that cores will be multiplying in future with frequency is almost staying within the nowadays boundary.

[Niseg]

I haven't had time to measure  what affects the game the most but my guess is that it's link list driven memory processor which means it kinda kill cache performance( I could be wrong). I think your best bet is a big cache(might not help at all) and high bandwidth RAM with a predictive data prefetch on cpu (if such thing exists) . High number of operations per clock along with high clock speed may help.

Maybe sandy bridge i7 2500k or 2600k with dual channel or get one of the older one with triple channel . I do admit I didn't  look at CPU performance  for years since most games are more GPU centered .  I was generally satisfied with my 3 year old q6600 +9800gtx  until I started playing DF. Now I'm "backtracking" to Overclocking subjects. My next plan is ddr2@800Mhz fsb@400Mhz  and core@3.2GHz (using an 8X multiplier) to see .I'm currently using ddr@667 fsb@333 and core@3.3Ghz which isn't optimal(gave up memory bandwidth for FSB and cpu). I don't think my CPU can do 3.6GHz without some work and in the summer heat it would probably won't be stable.

Workstations based on server processors might be good for DF but they are usually more expensive than mainstream desktops. I am also clueless about DF performance on an AMD but I think Intel took the lead in performance per clock with their core architecture.

[Grax]

I haven't had time to measure  what affects the game the most but my guess is that it's link list driven memory processor which means it kinda kill cache performance( I could be wrong). I think your best bet is a big cache(might not help at all) and high bandwidth RAM with a predictive data prefetch on cpu (if such thing exists) . High number of operations per clock along with high clock speed may help.

Maybe sandy bridge i7 2500k or 2600k with dual channel or get one of the older one with triple channel . I do admit I didn't  look at CPU performance  for years since most games are more GPU centered .  I was generally satisfied with my 3 year old q6600 +9800gtx  until I started playing DF. Now I'm "backtracking" to Overclocking subjects. My next plan is ddr2@800Mhz fsb@400Mhz  and core@3.2GHz (using an 8X multiplier) to see .I'm currently using ddr@667 fsb@333 and core@3.3Ghz which isn't optimal(gave up memory bandwidth for FSB and cpu). I don't think my CPU can do 3.6GHz without some work and in the summer heat it would probably won't be stable.

Workstations based on server processors might be good for DF but they are usually more expensive than mainstream desktops. I am also clueless about DF performance on an AMD but I think Intel took the lead in performance per clock with their core architecture.

Having 2.6Ghz processor overclocked to 3.2Ghz means you get 20% improvement.
Equal to FPS increase from 20 to 25.
Not a great deal for the price of the whole system stability. I'd say its not perceptible at all.

[Sean Mirrsen]

I often wondered, is optical logic possible? Kinda like the transparent crystalline "processor units" you see in every second sci-fi setting, except, y'know, not sci-fi. Maybe some kind of nanocrystalline macrostructure with intrinsic polarizing/piezoelectric qualities that allow the individual crystals to alter their transparency in planes nonparallel to the initial coherent beam passing through them... or something.

Okay, that sounded exceedingly Star Trek. But really, have there been any developments in altering the basic principle behind CPUs rather than refining existing technology?

[Grax]

I often wondered, is optical logic possible? Kinda like the transparent crystalline "processor units" you see in every second sci-fi setting, except, y'know, not sci-fi. Maybe some kind of nanocrystalline macrostructure with intrinsic polarizing/piezoelectric qualities that allow the individual crystals to alter their transparency in planes nonparallel to the initial coherent beam passing through them... or something.

Okay, that sounded exceedingly Star Trek. But really, have there been any developments in altering the basic principle behind CPUs rather than refining existing technology?

DF. The only game where users can argue about deep earth core geology, animal breeding and execution, surgery and microprocessor elaboration.

[Sean Mirrsen]

Well, it's not like it's rocket surgery. I mean, brain science. Or, whatever.

[Daetrin]

I often wondered, is optical logic possible? Kinda like the transparent crystalline "processor units" you see in every second sci-fi setting, except, y'know, not sci-fi. Maybe some kind of nanocrystalline macrostructure with intrinsic polarizing/piezoelectric qualities that allow the individual crystals to alter their transparency in planes nonparallel to the initial coherent beam passing through them... or something.

Okay, that sounded exceedingly Star Trek. But really, have there been any developments in altering the basic principle behind CPUs rather than refining existing technology?

Yep.  Birefringent materials have been looking like the best bet for optical computing switches, but the 'problem' with optical computers is that photons just don't act like electrons.  It requires really clever optical mechanisms to actually reproduce the AND/OR/NOT gates, and then there's the fact that you might as well use three value logic instead of two.  Oh, and you have to turn the optical output into an electrical one in order to send it to a monitor or a wifi card or etc.  Most of the bits for optical computing have been figured out but nobody's yet built a functional one.  Nobody wants to rewrite Assembler for three values, and it's just too expensive and finnicky to put together a working machine at the moment. Silicon technology is just so far along any other computing technology has a long way to go to catch up.

In a lot of ways quantum computing is further ahead, but I don't think anyone is qualified to write quantum-logic Assembler.

[Niseg]

Having 2.6Ghz processor overclocked to 3.2Ghz means you get 20% improvement.
Equal to FPS increase from 20 to 25.
Not a great deal for the price of the whole system stability. I'd say its not perceptible at all.

My Q6600 is a 2.4Ghz processor stock  and running at 3.2GHz is 33.3% . I'm hoping that pushing up memory and fsb bandwidth would give me more gains than just raw CPU speed. It's not considered hard to OC a Q6600 to 3.6GHz on air cooling.  I'm not sure my gains are only from  raw cpu speed.

I did the math and found that my current memory bandwidth (fsb synced to ddr) is 10.656GB/s and pushing to 400mhz fsb (staying in sync with RAM) it would be at 12.8GB/s which is about  20% gain. I don't think going any farther would help because my DDR can only go  up to 800MHz.

I often wondered, is optical logic possible?

yes , an optical transistor kind of exist in some way or form (not sure about the specifics). You are better off getting IBM to make you a  Silicon Germanium x86 processor . IBM made a processor that can run at 350GHz at room temperature and they designed the processors for XBOX,PS3 and Wii but those are all PowerPCs (sort of an extended RISC) . There is still an issue on what you should use for RAM. DDR-sdram is a  cheap technology. You can ask IBM to make you a 1024 bit bus for DDR (instead of dual channel hex channel) and at DDR3-2000Mhz it would run at 256GB/s.
 

[Nooodz]

Quickly returning to the original topic: Besides CPU and cache size/speed, wouldn't RAM frequency also be an important factor for DF? Its very easy to make the DF process take more than 1 GB of memory, so the difference between DDR2 and DDR3 should be noticeable. I guess this is especially the case in Windows, who just LOVES paging virtual memory.

[Thief^]

RAM latency is more important I think.

[Niseg]

Quickly returning to the original topic: Besides CPU and cache size/speed, wouldn't RAM frequency also be an important factor for DF? Its very easy to make the DF process take more than 1 GB of memory, so the difference between DDR2 and DDR3 should be noticeable.

I think getting the fastest memory or highest bandwidth would be a good idea provided that your CPU supports it ( I saw some of Intel cpu support upto 1333 and if that's true you'll gain nothing from plugging a 1600mhz or faster ram). I'm not sure about the internals of DF but if it's a cache thrashing program that uses a lot of pointers it may benefit from a higher  memory bandwidth and low latency would also be good but as far as i remember the effect of latency on other program is marginal (but they are not dwarf fortress).

DDR3 has a lower (half) internal clock speed as DDR2 so it saves a lot of energy but it's not necessarily faster(you thinking lower clock faster?!  think in parallel  ). Due to it's lower internal clock its interface can be much faster . The latency of ddr3 may seem high but it's measured in clocks ( real clocks which are half the rating) which are usually higher in ddr3. A ddr3 1600Mhz with cas latency of 10 has the same latancy of ddr2 800mhz with cas latency of 5  and the ddr3 has lower latency overall (shorter data stage due to higher clock).