Ok, so I've been thinking about a rant on this topic for a while anyway, but a good friend's predicament made me decide to go ahead and do it. For the sake of this post, we'll call him "Thomas", to avoid any potential embarrassment.
If you've ever seen a high end Enterprise workstation, or an older PC, you might remember something being vaguely different from a lot of PC's these days. If you can guess what I'm talking about, raise your hand. Then feel like a moron, since nobody can see your hand. The thing I'm talking about, of course, is the fact that those computers don't move enough air to achieve escape velocity. Holy shit, crazy talk, I know. No turbofan sitting next to your desk? How dare I blaspheme the pantheon of PC case manufacturers that way?
Don't get me wrong, there's absolutely nothing wrong with a case with a decent amount of airflow. But a decent amount of airflow and a decent amount of moving air that doesn't do a damn thing are entirely different animals. My friend "Thomas" is a great example of this. He has some cheap-ass gamer case that Ali-Baba and the CyberPower Thieves stuck him with. It has two front 120mm intakes, top-mounted isolated PSU intake, optional (in this case installed) 140mm top exhaust, the radiator for his (ugh) Corsair H50 mounted as an exhaust in the back, and one or two side 120mm fans. That's a metric asston of air moving around, in case you hadn't noticed. Somehow, though, he can't keep SLI GTX460's very cool. Granted, that's not always easy, but it's not insanely difficult, either, and he's hitting dangerous temps under load.
So. why am I bringing this up? Well, airflow is complicated, and I'm not going to claim to know everything (or even close) about aerodynamics. What I do know, though, is that a lot of things people think they know about PC cooling are very very wrong. A popular myth: "Get a bigger case to keep things cooler." Horribly wrong. Get a case that's got room for your components. Large amounts of dead airspace with no components create a path of least resistance for airflow that doesn't serve a purpose.
There's a couple of (general) rules to follow when trying to figure out your airflow.
1: Know your components. Nothing more important than knowing your configuration's needs. Granted, these can change later on with upgrades and whatnot, but it's still good to know.
2: Keep dead air space to a minimum within reason. Don't cramp your components, trapping heat is bad. But once your components fit reasonably, too much more space makes for bad airflow to where you need it.
3: No competing airflow. Avoid cases where your PSU is going to be fighting your CPU cooler or GPU airflow.
Here's the thing. Air follows the path of least resistance. Hot air rises, and heat is bad stuff. Because of this, if you can stand the slightly higher noise, External Exhaust graphics cards are good stuff, since they blow the hot air out of the case, unlike radial coolers that just get the heat away from the GPU and let it rise up through the CPU cooler. Negative pressure cases are ok for systems with limited restriction on airflow, but if there's a lot of stuff in there, actively directing the airflow with a positive pressure system is frequently better.
Don't forget to prioritize cooling. RAM doesn't need cooling as much as your CPU, so don't stick some retarded fan over your memory that thrashes airflow for your CPU. If you have a side panel, push air into your GPU if it's external exhaust, or pull air away before it can rise if it's radial cooling. If you're using a closed loop liquid cooler, I don't care how much it helps your CPU temps to use the radiator as an intake, just say no to pulling heat into your case.
Above all, apply common sense. The more directly cool air gets to your components, and hot air gets away from them, the better. Always. Avoid dead air, and avoid competing airflow. Avoid pointless restriction.
Remember, a lot of moving air doesn't mean good cooling. A space shuttle launch moves a shitload of air, and that's not cold at all.
Saturday, October 1, 2011
Monday, September 26, 2011
Benchmarking for Fun (Or for Info.)
So if you frequent tech forums, you might hear something along the lines of "...only matters in synthetic benchmarks." or "You can't tell the difference without a benchmark." And, if you like the idea of having a slightly overpowered system, you may just want to know how to go about those benchmarks. Even if that's not the case, you may be wondering what will help your performance in a particular game the most, or wanting to find out just how effective your CPU or GPU is for a certain game.
If you read a lot of reviews, you've probably seen all the methodology, all the different names of software, and seen tons of numbers with pretty bars and graphs. Well, I hate to tell you this, but generally, the pretty bars and graphs don't just come with the software. Luckily, it's the numbers that matter anyway.
So, for starters, lets discuss benching for fun. The most commonly used benches for gaming performance are Futuremark's 3DMark series. The "standard" comparison points are the default settings, which you can get in the free trial version. Lets you get a better database of scores going.
If you're going to bench for fun, the most important thing to remember is that you need a consistent set of benching processes. The best way to do this is to have a secondary account on your PC with the absolute minimum of automatic services, to make sure memory use is consistent. Keep drivers updated, although you usually won't want to use Beta drivers unless you need a specific feature. (Always test beta drivers prior to sustained use, some graphics drivers have been known to cause thermal issues.)
So, now that you have your consistent stuff, you get a baseline. That's just at your normal settings, how does it work. After that, you can try tweaking things to see if they gain you performance. Whatever you decide to tweak, make sure you test with the same processes as before. It really can make a huge difference in scores, or enough to skew data to the point of unreliability.
Generally, the GPU is the easiest thing to tweak, since there's simple GUI based OC utilities that work reasonably well, like MSI Afterburner, or (to a point) EVGA Precision. Remember to push it up a little at a time, and test for stability before starting the benchmark. If you raise your GPU OC and suddenly lose performance, you either need to raise your GPU voltage (at your own risk), or lower the clock back down a bit. Generally, in graphics cards, RAM clock will be less performance for the voltage than core clock and shaders, so you generally won't want to bother with it.
The CPU, is, of course, one of those things people really think about overclocking, and a lot of people are scared of it. If you do your homework, have the right components, and do it carefully, it can be a perfectly safe way to gain performance. You only void your warranty if you do damage that can definitely be attributed to overclocking, so usually you're safe if you keep the OC within the CPU manufacturers specified voltage range.
This isn't an overclocking guide, but I will say that in recent i7's, it's usually best to raise clock to the desired level first, then see if you can squeeze hyperthreading back on safely. Hyperthreading will raise your score in 3DMark, but less than most clock speed gains. RAM speed and timings can also affect it, as can IMC clock, and basically all the usual culprits.
But now, I'm sure, you're wondering about the benching for information thing. After all, E-peenery is fun, but there's only so much you can say about it. If you want to learn something through benchmarking, you're going to need a few things.
1: Consistent Monitoring. Be it FRAPS for FPS/Frametimes, HWMonitor for temps, or something else, you must use software. Eyeballing it is worse than useless.
2: Consistent Playthrough. You need a replay, or a specific save file, or something that you always use for benchmarking. If you're getting an FPS recording of different things, it doesn't tell you much.
3: Objectivity. This is the key. If you aren't objective, you're useless. You need to be willing to get results you didn't want or expect.
Well, as for methodology, it depends a bit, based on what exactly you're wanting to test, but essentially you need to isolate a component. For CPU/RAM testing, graphics settings not requiring the CPU should be turned down, to include resolution. This ensures that the limiting factor will be your CPU. For RAM, you just get a baseline at a "Typical" setting, like CL9 1333 for DDR3, and work from there, using the same CPU clock.
If you want to isolate GPU, you crank the desired graphics settings, turn down anything you can involving the CPU, and hope you aren't trying to isolate the graphics card on an RTS. Won't happen unless you have a really screwy rig.
If you want to monitor temps, you need a consistent ambient temperature, and consistency in everything except what you're testing. Including what you use to heat up the PC, and how long you run it prior to measuring.
I know this hasn't really been long on details, but it's really more to get you thinking the right way.
If you read a lot of reviews, you've probably seen all the methodology, all the different names of software, and seen tons of numbers with pretty bars and graphs. Well, I hate to tell you this, but generally, the pretty bars and graphs don't just come with the software. Luckily, it's the numbers that matter anyway.
So, for starters, lets discuss benching for fun. The most commonly used benches for gaming performance are Futuremark's 3DMark series. The "standard" comparison points are the default settings, which you can get in the free trial version. Lets you get a better database of scores going.
If you're going to bench for fun, the most important thing to remember is that you need a consistent set of benching processes. The best way to do this is to have a secondary account on your PC with the absolute minimum of automatic services, to make sure memory use is consistent. Keep drivers updated, although you usually won't want to use Beta drivers unless you need a specific feature. (Always test beta drivers prior to sustained use, some graphics drivers have been known to cause thermal issues.)
So, now that you have your consistent stuff, you get a baseline. That's just at your normal settings, how does it work. After that, you can try tweaking things to see if they gain you performance. Whatever you decide to tweak, make sure you test with the same processes as before. It really can make a huge difference in scores, or enough to skew data to the point of unreliability.
Generally, the GPU is the easiest thing to tweak, since there's simple GUI based OC utilities that work reasonably well, like MSI Afterburner, or (to a point) EVGA Precision. Remember to push it up a little at a time, and test for stability before starting the benchmark. If you raise your GPU OC and suddenly lose performance, you either need to raise your GPU voltage (at your own risk), or lower the clock back down a bit. Generally, in graphics cards, RAM clock will be less performance for the voltage than core clock and shaders, so you generally won't want to bother with it.
The CPU, is, of course, one of those things people really think about overclocking, and a lot of people are scared of it. If you do your homework, have the right components, and do it carefully, it can be a perfectly safe way to gain performance. You only void your warranty if you do damage that can definitely be attributed to overclocking, so usually you're safe if you keep the OC within the CPU manufacturers specified voltage range.
This isn't an overclocking guide, but I will say that in recent i7's, it's usually best to raise clock to the desired level first, then see if you can squeeze hyperthreading back on safely. Hyperthreading will raise your score in 3DMark, but less than most clock speed gains. RAM speed and timings can also affect it, as can IMC clock, and basically all the usual culprits.
But now, I'm sure, you're wondering about the benching for information thing. After all, E-peenery is fun, but there's only so much you can say about it. If you want to learn something through benchmarking, you're going to need a few things.
1: Consistent Monitoring. Be it FRAPS for FPS/Frametimes, HWMonitor for temps, or something else, you must use software. Eyeballing it is worse than useless.
2: Consistent Playthrough. You need a replay, or a specific save file, or something that you always use for benchmarking. If you're getting an FPS recording of different things, it doesn't tell you much.
3: Objectivity. This is the key. If you aren't objective, you're useless. You need to be willing to get results you didn't want or expect.
Well, as for methodology, it depends a bit, based on what exactly you're wanting to test, but essentially you need to isolate a component. For CPU/RAM testing, graphics settings not requiring the CPU should be turned down, to include resolution. This ensures that the limiting factor will be your CPU. For RAM, you just get a baseline at a "Typical" setting, like CL9 1333 for DDR3, and work from there, using the same CPU clock.
If you want to isolate GPU, you crank the desired graphics settings, turn down anything you can involving the CPU, and hope you aren't trying to isolate the graphics card on an RTS. Won't happen unless you have a really screwy rig.
If you want to monitor temps, you need a consistent ambient temperature, and consistency in everything except what you're testing. Including what you use to heat up the PC, and how long you run it prior to measuring.
I know this hasn't really been long on details, but it's really more to get you thinking the right way.
Subscribe to:
Posts (Atom)