Building the Ultimate Gaming Desktop

Alan writes "FiringSquad has just posted my Ultimate Gaming Desktop system building guide in which we take a no-budget but don't-waste-money approach. We even use an Athlon FX-57 in here. This is in fact only day one of a five-day series that will total over 32,000 words..." From the article: "Today's games aren't multithreaded. So, when designing a gaming system only one CPU core is needed. Therefore, the fastest individual core is going to be what's important for having the fastest frame rates and the fastest benchmarks. In real-life, when you're playing a game, your CPU still needs to spend time managing memory, the swap file, all while keeping your real-time anti-virus file scanner and firewall active. Everyone claims to run a clean system, but how many of us have been dropped out of a LAN game because we received an instant message?"

Comments

Re:Hmmm

[uberwidow]

Yeah, the typical reviewers don't seem to understand that benefits of dual core and dual CPU systems. In Widows XP you simply right click one of your CPUs and assign affinity to whatever application you want dedicated to that CPU. With the OS and the other overhead apps affinitized to the other CPU you have the potential of a full opteron 275 fully dedicated to your game.

However, it is standard media reviewer dogma to poopoo dual core and dual CPU systems. It doesn't make any sense because there are several games in development now that make use of multiple threads and World of Warcraft makes use of sycronous loading which allows multiple graphical loading requests to be made at once. Hyperthreading helps with this but dual core or dual CPUs would help much more.

Premature optimization is the root of all evil

[Moraelin]

Yes, that would be good and fine, if it wasn't utterly false.

Let's take those in the paragraph you quote, because two are synchronous tasks, and one is a whole 0% to 1% CPU time.

1. The swap file. Do you understand how that works? It a process wants to read 1 byte from the memory location X, that process can't possibly proceed until that byte has been fetched. If that's in a page currently swapped out, it can't possibly proceed before it's finished loading back into RAM.

So offloading swap management solves... what? You wait for that page anyway, and wait exactly as much time anyway, because it's the HDD that's the bottleneck there. So offloading that to another CPU will bring exactly _zero_ benefit.

2. Your real time virus scanner. Another synchronous task: if your game is waiting (e.g., at a loading screen) for a block to be loaded and scanned by the real time AV scanner, that's it. That thread is stopped and waiting until the scanner is done with that block.

So, again, a second CPU will bring exactly _zero_ benefit there,

3. Your real time firewall. Hit CTRL-ALT-DEL, look at the CPU usage for that one. Oops it's at most 1%, most of the time less. Yeah, it sooo makes sense to buy a slower dual-CPU for that.

Here's just some simple maths: if you have a 2.8 GHz CPU and lose 1% of that to the firewall, it leaves you with some 2.77 GHz worth of power for your game. If you get a 2.4 GHz dual CPU so the second one can take care of the firewall, you're left with a 2.4 GHz CPU for your game. Ooops, so dualies are still a losing proposition after all.

So, no offense, all I see there is one aspect of why premature optimization (in this case, of hardware) based on false assumptions and lack of measurement is bad. That's just the problem: you end up dumping time and/or money and more often than not end up with something actually _slower_ than the straightforward solution. In this case you dump a bunch of cash on a l33t dual-core solution, based on false assumptions about what those processes do and how, and actually end up _slower_ than a cheaper one-core solution. Was it worth it?