Page 2 - All Folders: Please Read!

Count me in. Just lemme know what you need from me.

Introduction

A lot of people are AFRAID of to fold. Yeah, you heard it. It’s true. There are several reasons, why people think that folding is bad, and they simply don’t fold. Which are these reasons? Well, one of them is that F@H is doing intense mathematical calculations, so for sure it will affect overall system performance, it will slow down their systems, games think that they’d lose precious fps’, and such. The fact is that THESE people do NOT understand that Folding@Home only uses the processors’ spare cycles. So if another software or game is running, then simply your operating system will give much more priority to that application than to F@H. If it’s configured ”as it’s meant to be”, then Folding@Home’s (and every other distributed computing application’s) priority is set to idle (or in some cases ‘low’).
Whenever you start an application it’s priority will be set as normal by operating system- normal is higher than idle/low, so that’s the reason why folding uses only spare cycles and system resources, so generally speaking it’s working (calculating) when your system is idle or it’s not running at it’s full potential.

The other reason why people don’t fold, is that they think that running a system at 100% load at 24/7 it will reduce it’s lifespan, and it will just die in a “few” months. Well, think about how computer processors are made, and that usually they’re rated for 20-22 years. Running your system at higher load will reduce the life span of your components, mostly due to greater heat produced, even if you’re overclocked, let’s say – you’d lose around max ~5 years (that was really exaggerated), what would that mean? You’d still have more than 16 years to run your precious processor.

Purpose of this article

In this article, I’ll try to “show” you the real truth, how folding affects system performance, with lot’s of benchmarks, charts, and numbers.

The F@H client what I was using for these tests, benchmarks is the 5.04(beta) ‘console’ version.

My system specs are the following: Pentium 4 630 HT 3GHz@3.6GHz, ASUS P5P800, 2x512MB DDR400 (dual channel), 120GB SATA2 Seagate Barracuda 7200rpm HDD. Operating system is Microsoft ® Windows 2003 ‘Server’.

The benchmarks, and comparisions were not made with the purpose to compete with somebody, or that I’ll show you here my scores, and how my system performs, their one and only purpose is to compare my system performance “without folding” vs. “folding with one client” vs “folding with two clients due to HT”.


I’ve used the following versions of applications for the tests:

1. SuperPI / mod1.5 XS
2. SiSoftware Sandra 2007 –lite–
3. MadOnion’s 3DMark01SE
4. MadOnion’s PCMark’02
5. FutureMark’s PCMark’04
6. Prime95 23.8.1.

During the testing, the display resolution was 1024x768@75Hz, with 32-bit color.

I didn’t changed priorities, configurations, options at any of those testing applications, just started them, and run them, in different situations. Also when running same tests one after each other, I quickly rebooted, because of the possible “data” already in cache and such. It’d hurt the benchmarks, and here like I said- I don’t want to demonstrate how my system performs, only that “how” distributed computing like Folding@Home affects system performance, and fps, and such for gamers. Read one, we’re heading to the test section now. I didn’t “manipulated” any of the results/benchmarks, if you’ve got any questions regarding my testing way and methodology do not hesitate to PM or e-mail me (you can find information in my forum profile).

TESTING nr#1 – gaming performance – 3dMark01SE

Gamers are always feared of the possible FPS drop/loss when in extreme situation, because that’d result in an in-game ‘death’ or ‘defeat’. That’s not good.

Why I’ve used 3DMark01SE? Because this it’s one of the 3DMark version what is really depending on processor speed and it’s using heavy our processor. Newer version are more heavily on video card, and that’s why they stress the GPU instead of CPU.

Let’s move on.

Here’s the result screenshot, without folding running:



And here’s the screenshot, when it’s actually folding one client in the background:



Yeah, I know, you’ll stare and scream that there’s so called HUGE performance loss, my points from 6984 dropped to 6740. But don’t forget, that benchmark results can and will fluctuate like with around 5%. That’d be like a fluctuating with around ~350 pts. In this case it’s lower. Also for your confidence, check the actual FPS drops. First game test: from 76,3 fps dropped to 76; second game test: from 96 dropped to 95,6; third game test: from 129,4 dropped to 125,5 – and here’s the last one, the only test what depends on Pixel Shader, and it puts huge load/stresses your video card, how much is the FPS drop? It’s zero. Today’s games are more dependant on GPU, and they stress more the video card, than the CPU. So actually this test is showing the real result and can be compared with games nowadays. As you can see, zero performance drop.

Let’s see how are the results when dual folding, using my processor’s Hyper Threading capability:



Hehe, yeah it’s actually better than with one client running. But, do you remember what I wrote in my last paragraph about score fluctuating? Well here’s only around ~100 pts fluctuation, and the allowed is should be around 350. So don’t make problems, check out the FPS drops, I won’t enumerate again them here.

It's clear enough, how folding affects gaming performance, let's see more.

TESTING nr#2 – overall system performance – SiS. Sandra 2007 ‘’lite‘’

The next categories of people are those, who’re afraid of performance loss, while folding.

I’ve done the following tests with Sandra 2007: “CPU Arithmetic”, “CPU Multimedia”, “Memory and Cache Performance”.

Let’s see the results!

A. CPU Arithmetic



See the results? See the differences? You’re already convinced, right? Whoa, that was FAST.
I won’t explain again the theory of score fluctuations when benchmarking, figure out yourself, and remember it, because it applies to every application.

The difference between folding and not is around 23MIPS. That difference is so ‘small’ that it can be and must be negligible-- even if you’re using HT and running 2 folding clients.

B. CPU Multimedia



Nah, here’s the multimedia synthetic benchmark. Results are clear enough, there is no real margin of difference, you can’t even notice the difference while watching only the charts, so don’t get too excited. Folding@Home demonstrated that it doesn’t affects multimedia performance.

C. Memory and Cache Performance







And here’s one of the charts, which contains all three of them:



First of all, notice the difference between “without folding” and “one folding client”. What’s the difference between “GREEN” and “PINK”? Almost nothing… it can be again negligible.

But… here is also another so-called issue. As you can see at this test, there is a big difference between running one client and two clients on the same. Why is this? Probably, and I’m almost sure that it’s, because of that so-called hyper threading technology. It’s alright, and okay, but memory and cache performance drops. The difference is around 2000MB/s it’s not so huge, but still. Also don’t forget that running on HT 2 clients, doesn’t gives you a “double” ppd boost, it’ll only give you around +150% boost, not 200%. The same reason why is here. Running on a really dual core processor would be totally different. If you’ve got one, then test it for yourself, and if you feel so.

Conclusion for Sandra would be, that actually there is no performance difference, between folding or not. It doesn’t affects system overall performance.

Let’s move on to then next testing.

TESTING nr#3 – pure mathemathical performance – Prime95

Okay, let’s see how folding affects our processor’s performance, when it’s computing, heavy algorithm and such. Prime95 stresses your CPU resulting in higher temperatures, and if instability problems occur, it’d inform you. I’ve used it’s built-in “Benchmarking…” section.

Here are the results:







The performance difference between these tests are so “slight” so that you need to focus on the numbers to realize them. There is no performance gap, some of the results fluctuating several milliseconds, but hey, when you’ll notice them in real life? Never.

TESTING nr#4 – calculation of PI – SuperPI

We’ve already seen how Prime95 benchmarks and performs when there is the well-known Lucas Lehmer algorithm. But now, let’s see how is performance affected when calculating PI.

A. SuperPI 1MB







Actually here is a “little” difference between the results. Keep in mind that when calculating PI on 1MB, and not higher it ONLY uses pure CPU processing power, and it barely touches our RAM (memory), because the data fits in our cache (usually).

There is no difference between running one client or two folding clients.

So, what bottom-line, this is the first benchmark where our performance is affected by distributed computing. We can’t forget that SuperPI is REALLY vulnerable, and sensitive about background applications. For example let’s say- running a simple Winamp and a FireFox (or IE/Opera) can degrade more than 1sec performance our so-called pi performance. SuperPI is sensitive, and usually there is no software/application/game where this difference should matter, we aren’t calculating PI everyday.

Whenever you want to benchmark your newly overclocked system, it can be done by “stop”-ing the folding client(s), and then it won’t affect your precious score. 

B. SuperPI 4MB







SuperPI being one of the most sensitive benchmarking application, here is also a little difference. But since in real world and overall system usage you don't calculate PI 24/7, right? Even if folding affects your system performance, that'd be around max ~1-2 sec, and hardly you notice that.

TESTING nr#5 – ‘old’ system benchmarker – PcMark2002

Remember? When there was no FutureMark, only MadOnion. I’ve tested with the PcMark2002 free version. But it's kind of oldschool...

Nah, let’s see the results and comparisions:







Well, unfortunately here is also a 'little' difference.

See for yourself… also this software is a little old, and maybe it’s not handling the priorities well, like it gives much resources to folding at home… and it didn’t uses it for itself when benchmarking.

Even if this would be true, it won’t effect today’s performance, the tests which were for performance effectuated in PCMark2002 are just old school.

Nevermind, let’s move on.

... reserved for possible future updates or benchmark additions ...

CONCLUSIONS

The results, as you can see, most of them are clear enough, how folding@home affects system performance.

Performance differences are negligible or even zero. It does NOT slows down your computer. You won't lose precious FPS' when highend gaming. Also, when rendering, compiling, encoding, designing and such, the difference would be at a long term calculations, a maximum of one or two seconds, and I think- that you can live with that. For gamers, well probably that you're gonna also wait a few miliseconds longer while loading the next map/level. But after that, it won't "hurt" your frame-per-seconds.

So, I think that you're convinced now. Please, join our folding@home team, for more information, see this link. Head over to Stanford's website, download the latest CONSOLE client (at the moment it's the 5.04beta) for your Operating System. Install it as a service. Read the following linky, about how to "tweak" F@H. Use the flags, as you want them, and happy folding (during these tests, I've run with the following flags: "-local -advmethods -forceasm" - and bigwu's enabled).

If you've got any more informations, questions, additions, opinions, feel free to contact me via PM/E-mail/IM.

FOLD TO CURE, FOLD FOR DEVH!

Lol

Excellent, benchmarking hyena. Great work, very thorough, at least what is up so far. Gosh people isn't that enough to convince you to fold, yes there are some areas where you lose some performance, but that is why there is the close button on F@H, you waste so many cpu cycles, when just casually using your computer.

FOLD TO CURE!

Once again excellent job madhyena.

Glad, you like it.

Actually the performance difference can only be noticed, with SuperPI and the oldschool PCMark2002- and this 'performance' gap is max a couple of seconds...

When you're benchmarking your newly overclocked system to hit the world record at SuperPI 1MB, then close damn'it F@H for those minutes, but after that, why waste your precious ownage processors' spare cycles.

As a note for overclockers, Folding@Home client is an excellent way to burnin, and stability test your system. It's not the perfect way to stability test (see my guide in sig, which is the perfect way), but still, folding can and will notice unstable systems, and will give you errors, will finish WU's early, not like they're meant to be. Then you'll know, hey- I've got a problem.

Keep the results coming, I'm curious. If you can use the benchmarking softwares or at least some of them, what I've used, all of them are freeware versions, so no need to pay money to use them. One thing is clear, with dual folding at a Hyper Threading system, it affects cache & memory performance. But I can live with that. But even Stanford Univ. stated that running 2 clients with HT rigs, it's not adviced, not for you, not for them. Because you won't get +200% ppd boost (double), only like +150% (one and half). But the results will be delayed for them, and they want the results, as fas as possible, to examine them.

So, if you aren't a "ppd-maniac" or hard-core folder, who's always checking his/her rank in the team, then I'd suggest running only one client. It's better for Stanford, for the cause, and for yourself, as performance. Then the performance difference will be zero or negligible.

Good luck, and fold to cure.

Come'on, where are your results? benchies? opinions? ideas?

BUMP! Gotta ressurect this thread.

I'm in.

i can also do a comparison of running two units on a HT system if you like.

Of course, see my comparisons as a "sample". Let's see the results.

i was thinking really that we wanted as many people as possible with different CPUs.

and i missed your whole section out until i'd posted hyena, soz.