PDA

View Full Version : One more 820


Finken
12-30-05, 12:47 AM
The following is intended to be:
A favor for pscout ;)
Inspiring
A guide to overclocking
Promotion for Folding @ Home
And hopefully a nice little project for me

I know a 820 may not be the smartest choice right now, with the 65 nm chips around the corner and so on...

I managed to fry my A64 3000+ (which did 2.7 (9x 300). I could have gotten a new one with a newer core. I could even have gotten a Opteron. And since I was really interested in getting a dual core I might as well have gotten myself a X2 seeing as the old one had to be replaced anyway.
But I had already been looking at the Pentium D for a couple of weeks because of the price.
Anyway, to be honest I just wanted some change! :)

On the A64 I was using a Swiftech MCW478-UTH (with a 226W TEC) and one BIX2. At stock speeds I was getting around -15°C (which caused hangups with my winchester). And at 2.7 GHz with a Vcore of 1.7 or whatever I was using, it was still below ambient at around 15°C.

According to my estemated calculations, my Pentium D should produce around 150-200 W of heat, depending of course on how high I can clock it. The calculations are based on a Vcore of 1.5 V.
I don't think the 226W TEC is going to be able to handle that kind of a load very well, and I'd also have to find a way to mount the block on the new 775 socket. So I decided to get a better waterblock (the MCW478 just doesn't have enough surface area IMO) and an extra BIX2.

The complete cooling rig will consist of a Danger Den TDX 775, 2x BIX2 with 6x 120mm fans giving around 40 cfm of airflow each, and a MCP655 pump delivering 20L/h.
What kind of temperature I'll get will be very interesting becaus this is actually my first normal H2O-rig. I've only used phase change and H2O+TEC before.

When I get my parts (hopefully today) I'll try to OC it on air first to see what it can do on the stock cooler.
This is in part due to the request from pscout to show how these 820 fold. And I want to show what they can do without fancy and expensive cooling.
But also I don't actually have all the parts for the cooling yet, so I can't start off with that.

The cooling should be educate to reach pretty good clocks, but what might be holding me back would be the RAM (TwinMOS PC5400 with crappy timings!) or possibly the PSU.
If the RAM is supposed to do 666 MHz that would be an FSB of 333 when using a ratio of 1:1. And with a multi of 14 that's a clockspeed of 4666 MHz. I doubt I will get that high, so apart from the timings everything should work.

My 550 W Fortron Source is supposed to deliver 180 W at 12 V, so I think I sould be OK, but I can always throw in another PSU to power fans, drives, and my pump.

Right now all I can do is wait for my stuff so I can start tweaking :)
But I'll keep posting all the progress, and I hope at least someone finds it a little interesting.... if not at least I'll have fun doing it :)
I will also try to get you some pics of the setup. I don't have a usable camera, but I've asked a friend to come over and take a few pics for me. So maybe you'll at least get some pretty pictures :D

Max0r
12-30-05, 01:10 AM
That's a neat lookin CPU I want to see what kind of #'s it can push.

pscout
12-30-05, 02:19 AM
Thanks Finken ... subscribed! :thup:

Not sure what the stock hsf is like but it will be interesting to see if you can fold QMD's much even at stock. Maybe with the stock 820 thermals. In the cooling section there is a thread on cooling and 830. (updated with some air cooling success with sonic tower) http://www.ocforums.com/showthread.php?t=429954&page=1&pp=30

I never had a stock intel hsf for my 830's but ihsetdr's was almost throttling at idle!

I wonder if intel gets encouragement from aftermarket hsf companies to make poor hsf's? And what companies like dell, etc do for cooling on these dual cores.

Sorry for the rant ... maybe you will find the stock 820 can do something more than idle without overheating on the intel hsf!.

Finken
12-30-05, 02:45 AM
Haha don't be so sure about it. I've heard these things get crazy hot! :)
Oh and thanks for the link.. I'll check it out. Do some research on what to expect.

What kind of temps do you get with your swiftech?

Btw, I'm kind of a noob when it comes to folding. I've been folding for a few years, on several machines, but I've never been thinking much about settings and flags or different work units. I might need some advice later on...

Oh and btw, that's an interesting point about hsf companies. I'm sure they're pretty happy :) But I don't think they can talk Intel into making hotter CPUs :)
As for Dell, I think Dell users don't generally notice if the chip would be throttling. But it would still be interesting to know if they have to give these PCs special care because of the heat...

EDIT: About the waiting... Looks like there'll be more of it :( Didn't get the parts today, so I'll have to wait till monday :(

pscout
12-30-05, 08:10 PM
The swiftech is running thru a dual heater core. I have the rad in a separate case in an adjacent room with 4 cheap fans ... right now its at 32C according to asus probe but it is getting outside toronto air.

The highest i have seen it was in the summer ... normaly around 41, sometimes up to 43.

It has about 4' of hose outside the cases and thru a wall so i can move them to get inside.

Been running it like this on 1.5v 724 for nearly a year folding, which on QMD's, for last 6 mos, is like running prime95 724.

Temps go up a bit when i am ripping/encoding video while still folding.

Folding is simple for me ... i am virtually all 800fsb p4's ... all setup for folding QMD's.
just set bigpackets=yes in client.cfg (using notepad, NOT wordpad) and set a flag -advmethods in properties on shortcut or in registry if running the console as a service. Like this you will get QMD's 100% of the time as long as it is a P4 and has at least 512MB. I can give you details if you want.

On HT p4's just run 1 folding instance ... but you can run 2 on the dual cores.
There are various ways to setup and configure them depending on what suits you best.

AMD guys have more of a challenge managing their farms ... WU mix changes over time.

Avoid the GUI client! ,,, open gl stuff makes it unstable.

Finken
12-30-05, 09:27 PM
Thanks for the details, I was wondering how to enable bigpackets. About the registry I'm clear, but the two instances... I have to have two sepparate clients in two different folders right? Like c:\program files\FAH-client1 and c:\program files\FAH-client2

Because I didn't get my stuff before the weekend I can't give you any numbers until about a week from now :( as I'm going on a LAN.
Well, I suppose I'll be home at least once so I'll at least try to give you stock temps, but all the interesting stuff will have to wait :(

Happy new year everyone :bday:

hUMANbEATbOX
12-30-05, 10:09 PM
i've seen a couple of sigs around here with 820's at 4.2ghz on water. you should have a good shot at at least 4ghz.

Finken
12-30-05, 10:23 PM
Ok, I'll do my best.. then we'll se what happens :)
4 GHz is what I'm shooting for, and if I get close to it I'll still be happy, but if I get more I'll be happier :D


BTW, why the Intel logo avatar? I'm curious.

I started using the logo for whatever CPU I was running as avatar. And then I thought it would be fun to leave the P4-logo even when I was running an AXP-M and later an A64. And it worked.. a lot of people got confused and thought I was a fanboy or something :D

Now it's more of a way to say that both companies make good CPUs, and even if I'm running one from AMD it doesn't mean I don't like Intel or anything... Guess I'll soon have to switch to a A64-logo :D

But seriously... Everyone is proud of their new rig with great parts and a lot of time and effort invested to OC and make it great. That's the reason I used those logos as avatar, because I was proud. But you can't go too far and start making AMD vs Intel a personal battle! That's kind of what my avatar is about ;)

EDIT: Oh btw, about the ratios... I think I might need to clarify what I was trying to say.
666 MHz (PC5400) is DDR, so the actual speed is 333 MHz. And I probably won't be running any other ratio then 1:1. So the only way to hit that kind of speed would be if I also get the FSB up to 333 MHz (1333 QDR). I don't know if that makes more sense then before?

pscout
12-30-05, 10:36 PM
Thanks for the details, I was wondering how to enable bigpackets. About the registry I'm clear, but the two instances... I have to have two sepparate clients in two different folders right? Like c:\program files\FAH-client1 and c:\program files\FAH-client2

Because I didn't get my stuff before the weekend I can't give you any numbers until about a week from now :( as I'm going on a LAN.
Well, I suppose I'll be home at least once so I'll at least try to give you stock temps, but all the interesting stuff will have to wait :(

Happy new year everyone :bday:

For 2 instances, you need to have 2 directories, and you need to run with the -local switch. Did you do the wedo 1 click? or install as a service manually?

I run all mine from shortcuts - makes it easier to manage and tinker with but once all is stable and unchanging i will likely switch most of them to services. I have 15 rigs to manage :D


... and a happy new year to you too!

Finken
12-30-05, 10:44 PM
Haha, with that ammount of rigs to manage it pays off with a good system for it :)
I'm using the wedo one-click.

Finken
01-02-06, 03:23 PM
Got the parts today, so now I'll get started... my buddy with the camera chickened out, so there'll be no pics on the building of the system :(
But I'll try to get some on the complete rig after installation.

Now I have to mod, build, and clock as myck as I can during one night :) Tomorrow I'm off to that LAN...

Finken
01-02-06, 11:52 PM
I've almost got everything set up. Now I just have to find that plate for mounting a second PSU. I noticed I didn't have enough 4-pin connectors to power everything.

The system looks pretty nice. I have a lot of cabling, and I could probably have done a better job, but it took more then enough time as it is!

When I started filling up the watercooling to test it for leakage during the night I noticed leaks around the rad. In a few hours I'll try to get some tools that I need, and I'll try again...

The pump is suspended from the cage-like structure surrounding the upper PSU (no pics yet sorry, try finding some pics of a Stacker if you don't know what i mean) so it's hanging down. It's actually up-side down, so we'll se how well that works.

I thing I'll sleep about 3 hours, then I'm off to the nearest hardware store. Need to be able to test for leaks at least a few hours before I go to that lan...
Too bad I'll hardly get any time for overclocking :(

EDIT: About 33°C idle in BIOS. Hard to tell because of fluctuations.
*on my way to LAN party*

Finken
01-09-06, 02:07 AM
I'm back now, and I have some bad news.
My worries about the PSU were justified.
I got a ****load of stability issues. After a while I clocked the CPU down from 2.8 to 2.45 and lowered the Vcore.
I don't know if my motherboard is overvolting, or if it just reads the Vcore wrong, but it's always around 0.05 - 0.1 higher then it should be.

Seems a ~100W CPU + 3 harddrives + 11 fans + pump and a X800XL were a little too much for a 180W 12V rail :(

I do have a few PSUs that I could throw in and power drives, fans and pump. And the stacker is perfect for this as it has a spot reserved for a second PSU. Now if I could only find the plate that you're supposed to mount it on! :(

If I don't find it I'll probably just throw it in there :) ..just can't really move it with a PSU tumbling around in the case :)

But I have some good news aswell... Granted I'm running the CPU under spes, but at full load I'm no higher then 35°C. So at least the cooling is working :) And btw, the load temps aren't a whole lot higher then idle, so I'm hoping they're not going to climb too high once I start clocking.

Finken
01-11-06, 11:02 AM
I've added the second PSU. Now my CPU almost has the full 180 W of the 12 V rail.

I jumped right in at 3.5 GHz (250 MHz FSB) at default voltage. Because I want to be using my computer while I check for stability I'm mostly just running some games. I'll probably up the FSB again before I go to bed and run some stress tests during the night.

For checking stability in the background I find sandra burn in pretty good. You can set the priority, and how much you want to push your CPU (CPU usage), and you'll probably want it at 100%. If you set the priority low you can do whatever you want with your PC and you will always use 100% of your CPU.

I've heard that the best way to detect an unstable overclock is with super pi (but prime95 has also been very popular). I usually use 3DMark 2001. It relies heavily on CPU and memory, and I detected several bad configurations when working on my last rig. If your CPU is unstable 3DM2001 will freeze up in notime ;)
I'll probably use super pi, prime and 3DM before I settle for the final max speed.

If there are any newbies reading this maybe I should explain the procedure with stresstesting.

You're probably going to overclock several komponents in your system, and not only the CPU.
With an AMD CPU you also get to max out the FSB pretty good because you can lower the multi and remove the CPU as bottleneck.
And these days we don't have to worry about PCI and AGP speeds anymore as most motherboards (at least the ones you would consider for overclocking anyway..) offer the ability to lock these settings.

But you still have the FSB, CPU, and memory to deal with. And if you really want the most out of your rig you'll probably OC your video cart aswell.

If you've OCed your CPU, and start working on the video card you will sooner or later run into stability issues. Doesn't matter if you're hitting a wall during testing, or if you don't even test and your system craps out on you while using it. Now where's the problem, the CPU or video card?

You have to make checkpoints as you go along. If I fire up a good app for stressing my CPU and go to bed, and everything is still running in the morning I can probably call this speed stable (to be absolutly sure I'd rather run it one more night, maybe even a different app). If I increase the speed more and run into problems I know my highest stable speed has to be somewhere my latest checkpoint and the first speed that gave me trouble.

So it's always clock, test, clock, test, clock test...

The more time you take to make sure everything is stable, the more you will thank yourself later.


BTW, pscout, what exactly was it you wanted me to check for you?

pscout
01-12-06, 07:43 PM
I was interested in what kind of temps you are getting at max stable oc running dual qmd's, and what kindof ppd it can produce. Was initially more interested in results on air, since that is what i am running my 830's on atm. But i plan to build a second water system to try on them as soon as I get through some other priorities. so water results are interesting too.

Also. on stability testing, I like to use prime95 for a little while 30min-2 hours depending on my confidence with the OC, then turn it off and fold qmd's disconnected from the Internet ... If you can produce a qmd or 2 with no EUE's this is probably the best test. And by backing up the wu before you start it folding, and disabling lan/internet access, you won't turn in a bad wu due to stability testing. If it fails, you can just restore the failed wu after adjusting your fsb down a bit, and repeat the folding test. If it completes successfully, you get folding points for your testing time.

Experience of the folding masters (eg ChasR) on the OCF Team32 folding team is that folding will find instabilities that long term p95 etc tests will not! I figure my folding farm rigs are stable when they can fold for days and weeks on end 24 hours a day without interruption essentially running something as good or better than prime95. And the folding is contributing to good science that benefits all of us and our kids.

Just don't use folding as a simple burn in test without regard to creating bad results through failed WU's. Do some initial stability testing with other tools first. I like prime95 since it will fail very fast if i have pushed too far. I just back off 5 fsb at a time until it passes the first and second tests of the blended torture test. After that, as long as temps don't rise, my rig will probably pass the 'folding' test. If it doesn't pass in ital folding test, back off 5 more fsb and is will pass without fail.

This is most suitable for P4's since they can get a guaranteed supply of QMD's to fold currently. With AMD's, you need to get lucky and get some big workunits to stress them well.

Here is a link a folding forum thread on the subject : http://www.ocforums.com/showthread.php?t=434609