AMA: Computer Specialist

[Gelare]

Question: How can you judge where the bottleneck is in a computer? I'm sure it depends on usage, so let's assume for the moment that I'm designing a new computer for gaming. How can I judge how to balance processor speed, cores, RAM, etc?

[Crissa]

It's tough, but different sounds and pauses can be measured, if you know the default times. Slow program launch is associated with slower HD; delays in processing things like video once loaded is usually processor; lower game processing is processor bandwidth/bus speed.

-Crissa

[Murtak]

Question: How can you judge where the bottleneck is in a computer? I'm sure it depends on usage, so let's assume for the moment that I'm designing a new computer for gaming. How can I judge how to balance processor speed, cores, RAM, etc?

Pretty much the only way to be sure is to actually play games with one configuration, measure the performance, and then switch out parts and measure again. Given experience you can of course make an educated guess. Even when measuring though it is going to depend on a lot of variables.

Take the following with a truckload of salt, because I don't play many games on my PC right now:
- Most modern games place a lot of work on on the graphics card. This is going to be your major bottleneck for being able to play at high resolutions. As far as I know ATI is currently more value per dollar than Nvidia, and a 5750 is a nice middle ground between power, price and wattage.
- You only need enough RAM to hold everything you currently use. Not having enough RAM, even a little bit, will fuck up everything though, because swapping is orders of magnitude slower. 4 gigs should probably suffice, 2 is very iffy.
- Not too many many games utilize multiple cores, let alone more than 2. Dual core processors are probably fine (and much cheaper). Unless you actually want the best, AMD is currently much for bang for the buck. So AMD dual core, AMD triple core or quad core if you can find one cheap. Processor speed is even less of an issue. 2700 vs 2900 is going to be hard to measure, let alone notice while playing.
- HD speed should not matter much while gaming, unless you detest loading times. If so, a good solid state disk will set you back a couple of hundred bucks and absolutely blow any non-SSD out of the water.
- As for motherboards and how much impact they even have on gaming performance - no clue.

[Maj]

My understanding of part of why defagging helps is because...

[krainboltgreene]

Flipping bits is one side of defagging.

[CatharzGodfoot]

I've been led to understand that creating a partition on the outer edge of the disc and then sticking things that you want to load quickly (e.g. programs & OS files) there helps with the file sorting issue, and can make a very large but slow drive seem faster. It also has an organizational advantage for multi-boot systems, and makes recovering from infection a bit less painful.

[RandomCasualty2]

My understanding of part of why defagging helps is because

I knew it. The gays are responsible for slowing down my computer too!

Damn [EDITED] files.

[Cynic]

so when you defag your computer, you send them to one of those camps?

[Gelare]

- Not too many many games utilize multiple cores, let alone more than 2. Dual core processors are probably fine (and much cheaper). Unless you actually want the best, AMD is currently much for bang for the buck. So AMD dual core, AMD triple core or quad core if you can find one cheap. Processor speed is even less of an issue. 2700 vs 2900 is going to be hard to measure, let alone notice while playing.

How can processors with different numbers of cores be compared? How can you tell whether it's better to have a 2.7Ghz dual core or, like, a 2.3Ghz quad core? Or what? How fast do your dual cores have to be to equal a 3.0Ghz single core?

[Meikle641]

I thought the big thing with multi-core systems was that the processors are able to split the load, putting less stress on the system, which in turn makes it "go faster".

I had a 3 Ghz P4 a couple years ago, but I'm fairly certain my 1.5 Ghz dual-core is faster. With the same RAM it's able to run many games and programs better.

[Crissa]

It's like having more than one computer doing the work. Also, you have a bus speed that's much higher than it was years ago (which will affect single-thread programs more).

Different processors sort commands at different speeds; once it was uncommon to have math co-processors, but everyone has multiple of them now. The number of commands each processor knew was more limited (the number of 'bits' like 8-bit or 64-bit).

So between the same processor, speed is more important. Whereas between different processors, the other features are more important. Even if you operated a Core 2 Duo at 8 mhz, it would be 'faster' than a 80286 at 8 mhz. (Although an 80386 at 8mhz wasn't faster than an 80286, but it was only a small step.)

-Crissa

[zeruslord]

And the Pentium 4 was a low point in real work done per clock cycle, being worse than the Pentium 3, the AMD chips around at the same time, and the more recent Core 2s and Core I*s. This was mostly for marketing reasons; the features they sacrificed that made it less efficient also let them give it a significantly higher clock speed to put on the front of the box.

That aside, multi-core systems are only able to split the load in certain ways. Each processor can run one "thread" at a time, and if a program is written to only run in one thread, a multi-core system won't get it done any faster if nothing else is running. However, if you feel like running multiple programs, they will run faster if processor speed was the limiting factor. So a 1.5 ghz dual-core system won't run typical games as fast as a 2 ghz single core system, all else being equal, but it will handle a game, a music player, and a web browser all at the same time much better

[Meikle641]

That's probably it, yeah. I have lots of stuff on the go, usually.

[krainboltgreene]

Each core added only boosts performance for a majority of software (including the OS) by a minor amount. Programs simply aren't built with multiple cores in mind, and they wont be for a very long time. Only a few languages take multiple cores into consideration, more notably Erlang.

For every core you have your Erlang program runs N times faster.

[Draco_Argentum]

Question: How can you judge where the bottleneck is in a computer? I'm sure it depends on usage, so let's assume for the moment that I'm designing a new computer for gaming. How can I judge how to balance processor speed, cores, RAM, etc?

Its always the video card unless you buy the really expensive one. A cheap current gen processor and 4GB of RAM should keep an ATI 5870 fed with more data than it can process. That card would cost more than the cpu and ram put together. I like to read guru3d.com and their tests hardly ever show no performance improvement for a more powerful graphics card, leading me to believe that the cpu is not the issue.

Another thing to be careful of is high speed ram/system busses. Those sound nice, but often require you to over clock your CPU or buy a motherboard that lets you change multipliers to get any value. CPU frequency is a (fixed unless you by the very expensive version) multiple of bus speed. Running your motherboard at the 1800 its advertised as capable of will overclock your CPU for example.

[Surgo]

- Not too many many games utilize multiple cores, let alone more than 2. Dual core processors are probably fine (and much cheaper). Unless you actually want the best, AMD is currently much for bang for the buck. So AMD dual core, AMD triple core or quad core if you can find one cheap. Processor speed is even less of an issue. 2700 vs 2900 is going to be hard to measure, let alone notice while playing.

How can processors with different numbers of cores be compared? How can you tell whether it's better to have a 2.7Ghz dual core or, like, a 2.3Ghz quad core? Or what? How fast do your dual cores have to be to equal a 3.0Ghz single core?

It depends on the architecture.

The Core2 architecture, for example, was pretty awesome and a lower clock speed still meant it was better than the Pentium 4 or AMD architectures.

The only time you can really compare clock speed is within the same architecture, at which point bigger is better. Everyone else already talked about dual vs quad.

[Surgo]

Also, there are a few CPU-bound video games. Oblivion is CPU-bound in a lot of ways, though you can download various graphic enhancers to fill up your graphics card to its limit.

I'm told the Source engine is also CPU-bound, but I don't know why.

[RandomCasualty2]

Also, there are a few CPU-bound video games. Oblivion is CPU-bound in a lot of ways, though you can download various graphic enhancers to fill up your graphics card to its limit.

I'm told the Source engine is also CPU-bound, but I don't know why.

It probably has to do with how the physics are calculated is my guess.

[Murtak]

I thought the big thing with multi-core systems was that the processors are able to split the load, putting less stress on the system, which in turn makes it "go faster".

That is the theory. In practice, threaded code is hard. Some tasks can be trivially split, like video encoding. The best that can be done for most games, is to split off a couple of threads for specific tasks, say, physics, and regularly query them for results from the main thread. You may get a quad core to run at 1.5 or possibly even 2 times the FPS that way, but not even close to 4 times. And many games don't even do that much.

I had a 3 Ghz P4 a couple years ago, but I'm fairly certain my 1.5 Ghz dual-core is faster. With the same RAM it's able to run many games and programs better.

That might just be your graphics card. But apart from that, processor architecture has improved. As has already been said, GHz numbers can only be compared within the same architecture. It is entirely possible that even a single core 1.5 GHz processor is faster than an older 3 GHz CPU.

[RobbyPants]

My understanding of part of why defagging helps is because...

Whoops!


... What? Wait! No! You just increase your computer performance by taking the bundles of sticks out of it. That's what I meant! Really.

[Crissa]

SimCity is horribly CPU bound, as no threaded versions have been written ;-;

However, the intel duos were able to take code and run it more efficiently by even splitting a single threaded operation across more than one core - basically pretending that one core was better at math while the other was better at some other folding. Aside from knowing that existed; your OS is always doing stuff in the background, which seriously slowed down the single-threaded machines in the past, so the dual-cores can interleave much better.

Basically, the processor evolution was mostly 'more commands' or 'faster' until we got to the Pentium stage where 64-bit was possible, but not any more efficient. Then they began to look at 'more optimization' or 'more offloading'. A multicore is the ultimate in offloading - it's basically another processor right there. But having math, graphics, and drive controllers and caches were all types of offloading.

-Crissa

[Blasted]

However, the intel duos were able to take code and run it more efficiently by even splitting a single threaded operation across more than one core - basically pretending that one core was better at math while the other was better at some other folding. Aside from knowing that existed; your OS is always doing stuff in the background, which seriously slowed down the single-threaded machines in the past, so the dual-cores can interleave much better.

This actually applied to many single core P4s, Xeons and many other processors.
Most single core processors have several units able to do work, some floating point units, some integer units. 'Hyperthreading' (which is merely intel's trademark for multithreading) allows the cpu to process threads using different units at the same time.

Basically, the processor evolution was mostly 'more commands' or 'faster' until we got to the Pentium stage where 64-bit was possible, but not any more efficient. Then they began to look at 'more optimization' or 'more offloading'. A multicore is the ultimate in offloading - it's basically another processor right there. But having math, graphics, and drive controllers and caches were all types of offloading.

-Crissa

except this is quite wrong.
Processors for many generations have been taking on more and more functionality. My knowledge starts with the 486DX, which took on the maths-coprocessor, but then you go on with the Pentium MMX, which provided basic 64bit operations, Athlon, which brought the memory controller onboard (que DEC discussion elsewhere), 3DNow! and SSE multimedia (including graphics) extensions. The state of the art is currently the AMD fusion, which brings a GPU on as an additional core and the 'System on Chip', which moves everything onto a single chip.
While you could argue extra cores (including SoC extensions) are offloading, the extended instructions sets introduced with every new processor are examples of the ever expanding capability of single cores.

64bit operating is much more efficient and significantly faster, if you're doing operations where 64bit data is used. On a PC, this is generally where you're dealing with more than 4GB of RAM. 128bit operations are available, but limited to a few users. The SSE2 instructions allow 128bit operations and are mainly used for multimedia transforms (I understand - this is not my area of expertise).

[Crissa]

No, it's not 'wrong': It's overly simplified.

If you want to add more information, do so. But by saying 'you're wrong' when what I said was merely too simple for you to stand is just being pedantic.

64bit is very efficient now, yes. But 128bit and the first 64bit processors weren't very efficient for what they did. There is a balance of what is more efficient vs just doing simpler commands faster. This is why Alpha computing was so powerful for so long.

-Crissa

[Blasted]

Question: How can you judge where the bottleneck is in a computer?

I'll address this solely in relation to games:
The accepted method is to run the game with as much graphical niceties and at the highest resolution possible, then measure the frames per second(FPS)
take the resolution down, measure again. Every time you take the resolution down, the GPU is doing less work. Because less work needs to be done, frames can be drawn faster, so FPS go up. If at any time FPS don't increase, you've reached the point where the CPU is the bottle neck.

For modern games the GPU is almost always the bottle neck.
I haven't seen a situation where RAM is the bottleneck in years. I suspect 4GB is enough for the time being. Disk for most games really only counts for loading times, though on some games such as WoW, seek times can come into it and buying an SSD will result in less hiccups in gameplay.
Of course, for the most part, SSDs have removed the reason to defrag and if you're obsessing over moving files to the edge of the disk, I think they're very much worth the cash.

[Cynic]

now this is an off-the-wall computer question. Well it's vaguely related.

In Candelas, how bright is an average LCD monitor?