View Full Version : Another dual Opteron system
Water cooling noob doing the first post ....
I've gleaned a LOT of useful info by lurking around a bit.
In particular from here (http://www.xtremesystems.org/forums/showthread.php?t=58472&highlight=dual+opteronhere) as I'm planning on building a similar system for work/hobby.
But there are enough differences that some choices remain unresolved.
Perhaps the biggest difference is ambient temperature - it will likely reach 90 F during the summer. That is why I'm going with water cooling.
The case requires a bit of descrption so please be patient.
Next major difference is the case. It will be a double-wide generically based on this (http://www.spodesabode.com/content/article/cubecase) DIY article. This should provide enough room for dual independent cooling loops. Since I don't have a matched set of EATX cases I'll be scrounging around for some used ones.
The bolted together case will be divided internally by a clear Plexi, Lexan type panel with holes drilled for the tubing,1/2" Clearview or Tygon ) to pass through. Slicing through the holes horizontally will split the panel into sections. This will locate tube mounted rubber grommets used to avoid chaffing.
The cooling section will have fans front and back blowing air into the pump/rad area to supply the fans pulling air through shrouded side mounted rads/heater cores.
Will probably use Swiftech MCW6002 blocks with gpu and chipset TBD but probably Maze 4. Both of these are subject to change.
Sorry for being so windy.
First question - since each loop will be cooling one cpu and either a gpu or chipset block will something like a Black Ice II or weapon dual core ( maybe with 5/8" barbs ?? ) be adequate? Or should I stay with a pair of Black Ice III's or maybe a modded triple heater core if such a thing exists ?
Next question - considering the extra tubing length needed which pumps would be best?
Thanks for taking time to read and respond.
bfx
MaxxxRacer
05-31-2005, 12:45 PM
Well first off i suggest you dont get a chipset block. kinda pointless. especially with the mass of tubing ur already going to have in the loop.
6002 is a good choice. maze4 for the gpu is a good choice.
so you would be cooling 2 cpus and a gpu. in all honesty this can be handled with one cooling loop and should be done with one cooling loop.
Considering the heat load and how much space you will have you can go with the monstercore. it will cool bout as good as two Black Ice 3's and have less restriction. It is very large though. its the size of two 302's put side by side.
as to the pump.. the maze4 is low restriction and the mcw6002 is relatively low restriction and the mcw6002's dont need alot of flow to perform well. You can probably do it pretty well with a mcp655 or a 50Z.
Thanks for the reply.
I was comcerned about getting enough cooling on those 90 F - 100 F days since the ambient starts out where some systems end up for cpu temps.
I was considering going with the monster core if it could handle both cpus plus whatever. Even with it's size that will simplify matters quite a bit and probably not take up much more room than two weapom dual cores I'd considered using.
Since this will be mainly a work system I'm open to "cost-effective overkill" suggestions on any of the cooling components. Especially ones that will help during those times of high ambient temps.
It's not that I have a big budget by any means but I'd hate to see a mb and cpus get toasted because I skimped in a critical area.
Maybe as a noob I'm being overly cautious.
bkk
cartmanea
05-31-2005, 01:57 PM
I wouldn't worry about independant loops, it just increases your costs needing two pumps and two rads. I'd just get a high power DC pump like the Swiftech 650 and a big radiator like a dual or triple 120. It won't hurt running the two cpu blocks in series if you use low restriction cpu blocks like the Maze4 design. If you use higher restriction blocks than run them in parallel, but series will keep your loop simple and the cpu's will be within 2-3 degrees of eachother anyway. Go to your cpu's before your gpu also.
MaxxxRacer
05-31-2005, 03:21 PM
cartmanea, loop order doesnt matter. thats a myth.
you are being overly cautious as these cpus will run just fine in 90F ambient temps with stock heatsinks (that is if ur running the cpus at stock mhz and volts). they sure wont like 90F ambient but they wont crash on you.
but if u want to do any tweaking or anything of the sort its a good idea to improve the cooling as you are doing.
anyway the monstercore with shroud would really be an excelent economical solution matched up with the mcp655 and daul 6002's. All of this would be in series including the maze4. the flow would not be amazing but it would be adequate. If you are really obscessed with getting the flow up a iwaki would be an excelent choice and my recent tests show that it actually puts less heat into the system than the MCP655 or MCP650. appx 14.5watts. but since from what you said i dont gather that you are going to do anything crazy a large expensive pump is not something you should bother with.
cartmanea
05-31-2005, 04:29 PM
MaxxxRacer,
Loop order DOES matter if you aren't running a closed system (no res). But if you have a reservoir you will have pressure drop at each waterblock. If you are using waterblocks that have high restriction then it is good to run the more critical items first in the loop. If you are running a closed system though, none of that matters.
MaxxxRacer
05-31-2005, 05:12 PM
cartmanea your completely wrong on this one... it is a MYTH that most people believe when they first start into watercooling their PC's.
But explain to me why you think loop order does matter in a closed loop system without a res.
Whether there is a res or not there is a pressure drop at each block. I can prove this to you if u like. I have a pressure differential meter there i use on a regular basis. it shows the pressure drop.
I belive the reason you think that in a closed system the pressure is the same is becuase in physics that is what they teach. but if u read the fine print they say the pressure is the same everywhere just to make the calculations simpler. this is atleast what i have come across in my physics books.
When there is no flow the pressure is the same but when u introduce flow, the pressure changes. the further down the loop u go the lower the pressure and at the back of the pump the pressure is NEGATIVE.. yes u will actualy have pressure that is below ambient 14psi. this is due to the suction of the pump. but due note the less restricted the pump inlet is the less negative it becomes, and can be positive if u set the pump up right (have it standing on end with a large res force feeding the pump)
MaxxRacer -
Don't plan any overclocking. But when/if the prices of the 2.6 ghz Opteron comes down some I'll be really tempted to go for a cpu upgrade.
Probably should have also mentioned I'll be stuffing as much ram in as I can afford. Minimum of 2 gig per cpu
Also - while I'm setting up a run the cpus will be mainly loafing with some periods of 100% duty cycle lasting a couple of minutes to maybe an hour or two. But once the run is actually underway it may be several days of continous 100% duty cycle for both cpus (multithreaded apps)
I took a look at the the Iwaki prices a couple of weeks ago (I've been lurking here awhile). They ranged from GULP !! to just a couple of bucks more than I'd kind of thought I'd have to spend.
So - following the 'cost-effective overkill" mode - which Iwakis would you suggest? Or maybe a change in water blocks?
Thanks
bhx
MaxxxRacer
05-31-2005, 07:13 PM
id say keep those waterblocks either way. but if u wanted to upgrade the pump i would suggest the WMD-20RLT or RZT. im not really sure if the RZT would be a better choice compared ot the RLT as far as restriction goes. I say teh W as it is cheaper than the jap motor. but if u can go for the jap motor u might as well. its 190 vs 110 for the american.
My aplogies ....
Looks like I'll be swiching to dual Xeon Noconas.
I just found some new test results that indicate a 2.6 ghz Opteron equals a 3.0 Xeon Nocoma in performance when the app is Hyperthread enabled. Prevois data for apps seemed to indicate the margin was much narrower.
So I'll be going with dual Noconas in the 32./3.4 range.
Know the Noconas run noticeably hotter than the Opterons - I suspect more cooling may be needed.
I prefer to use dc pumps rather than ac.
Maybe going with two pumps in series? I know that would add more heat from the pumps - would the extra flow help offset it enough to give lower cpu temps than with one pump? Is even feasible to run pumps in series?
Thanks
bfx
MaxxxRacer
06-04-2005, 02:56 PM
would still go with the dual opteron but...
its very feasible to run pumps in series. dual pumps in series gives you about 125-150% of ur previous flow with one pump. so its nowhere near double. (200% being double)
the actual performance increase depends on how large ur radikator is (to disipate the extra heat of another pump) and where your blocks are on their flowrate to performance curves. considering ur gonna have 2 cpu blocks and a gpu block u should get some pretty decent gains. nothing spectacular but something noticeable.
saratoga
06-04-2005, 03:58 PM
MaxxxRacer,
Loop order DOES matter if you aren't running a closed system (no res). But if you have a reservoir you will have pressure drop at each waterblock. If you are using waterblocks that have high restriction then it is good to run the more critical items first in the loop. If you are running a closed system though, none of that matters.
That makes no sense. Loop order doesn't matter, and the pressure drop across each componet varies only with flow. Which it damn well better since pressure drop is a differential measurement, and not an absolute one.
I just found some new test results that indicate a 2.6 ghz Opteron equals a 3.0 Xeon Nocoma in performance when the app is Hyperthread enabled. Prevois data for apps seemed to indicate the margin was much narrower.
Some sort of video rendering app I'm guessing? I'd still take the Opteron. Its a much faster machine overall.
cartmanea
06-04-2005, 05:10 PM
MaxxxRacer, the argument you made earlier seems to have supported what I was saying. Since there is pressure drop at every block, then blocks that benefit more by higher pressure flows, like those with nozzles, should be earlier in the loop right? I know any differences in performance would be pretty much negligible, but isn't that true? If I'm wrong I'll accept it cause I haven't been in watercooling long, nor am I a Physicist or Fluids Engineer (I'm an Electrical Engineer). So this stuff isn't my forte, but if you have pressure guages throughout the loop and there is higher pressure at the beginning then higher restriction blocks that benefit by high pressure would benefit being placed at the beginning of the loop.
saratoga
06-04-2005, 05:56 PM
MaxxxRacer, the argument you made earlier seems to have supported what I was saying. Since there is pressure drop at every block, then blocks that benefit more by higher pressure flows, like those with nozzles, should be earlier in the loop right? I know any differences in performance would be pretty much negligible, but isn't that true? If I'm wrong I'll accept it cause I haven't been in watercooling long, nor am I a Physicist or Fluids Engineer (I'm an Electrical Engineer). So this stuff isn't my forte, but if you have pressure guages throughout the loop and there is higher pressure at the beginning then higher restriction blocks that benefit by high pressure would benefit being placed at the beginning of the loop.
Like I said above, its not absolute pressure, but relative pressure that matters. Thats why they say "pressure drop" and not just pressure. You're correct that the absolute pressure is higher at one end of the loop then the other, but since the absolute pressure doesn't matter, the order doesn't matter either.
To put it another way, if you have the pump push into an open air res, the entire system will be operating at negative pressure, yet the system still works. If absolute pressure was an issue, you'd have a big problem in this case. But its not, so the system works the same under positive and negative pressure.*
*So long as you don't cavitate the pump.
saratoga
06-04-2005, 05:58 PM
Wait, you're a EE?
Think batteries then. Hook a resistor and battery in a simple circuit. You can ground either the positive or negative terminal of battery, and the current flow is the same through the loop. You can even hook the ground up to 100V and it makes no difference. All that matters is the voltage (pressure) across the resistance, and not its absolute value. Same idea here, just change current to flow, and voltage to pressure.
cartmanea
06-04-2005, 08:00 PM
Yeah, I agree with that explanation. Though hopefully people would know that the pump is more efficient in pushing than pulling, so if your whole system was in negative pressure it would probably be less efficient.
Although that would take the worry out of small leaks, it would just suck air instead of getting your components wet.
Thanks again for the responses.
So looks like the Swiftech MCW6002-PX's, which seem to be Nocona compatible, and a Maze 4 for the gpu with, I hope, a monster core setup.
Would the MCP655 or MCP650 pumps work well in series? Or maybe some pump might be better for series use?
And, as saratiga surmised, it is video rendering apps.
http://www.gamepc.com/labs/view_content.asp?id=pciews&page=7
Max is probably heavily optimized toward Intel but still those are freakish results.
Thanks again and apologies for the cpu change.
bfx
MaxxxRacer
06-04-2005, 10:41 PM
as to the pressure. i used to think that the pressure mattered for impingement style blocks. i consulted a few experts on this and they asked me if i was seriously asking this and then i realized that i was most certainly wrong.
but in anycase. the flow is the same throughout the loop.. thats what the block cares about. flow not pressure. so theres the explanation of why it doesnt matter what the pressure is in the block.
the 655 would be what i would go with. its heat dump is higher but with the monster core it reallly will NOT matter so its what i would do. 2 of those in series would work really well. if u got an adjustable psu put them up to 12.5 volts to get some extra power out of them (they dont respond to extra volts past 12.8volts i dont think)
the block choice is good (those are the ones i picked out :p: )
cartmanea
06-05-2005, 12:20 AM
Where can you find these Monster Core's?
MaxxxRacer
06-05-2005, 12:40 AM
send "Weapon" a PM. he gets them from an autoparts supply and modifys them with barbs, puts them in an acid bath, cleans them and paints them. they are very nice.
Holst
06-05-2005, 02:20 AM
Just to clear up the pressure dispute.
As water is (practically) uncompressable, water at high pressure will be as dense as water at lower pressure so there will be no gains to be had from pressurising your system (or by putting CPU block first in the loop)
Now if this were aircooling then pressure would matter as you could force denser air through the heatsink, more mass/flow = better cooling.
MaxxxRacer
06-05-2005, 02:24 AM
hehe. that is a good point holst.. running air at a higher pressure would be effective.. u would just need an intercooler for it though as the air would get warm by compressing it....
the block choice is good (those are the ones i picked out :p: )
I'd kind of narrowed it either the 6002 or TDX but couldn't decide. Both seem to perform well but mainly they're readily available. Your input has been very helpful. It's appreciated.
Looks like I'll be using a pair of 6002 cpu blocks, a Maze 4 gpu block, dual mcp665 pumps in series, a monster core and a DIY res to fit wherever it'll fit.
Now I need to find a used matching pair of EATX cases so I can start the case nuke and reassenbly.
bxf
MaxxxRacer
06-05-2005, 10:17 PM
i say go with a mountain mods UFO-2 case. they are amazing.
Pinnacle
06-05-2005, 11:14 PM
i say go with a mountain mods UFO-2 case. they are amazing.
They are cool. They got so much fans in em they really could take of like a UFO
I have seriously looked at the UFO-2 and experienced bouts of excessive drooling.
But it just doesn't have the room I'll need. At least by the looks of it. Plus it looks as though it would be more difficult to have isolated air zones. But I keep going going back to it in hopes of finding a way to use it. The depth is a major problem - the main ps I plan to use is 8.5" long - then there's the aux supply for cooling power.
Add in an EATX card , monster core, two pumps, res, eide boot drive and, eventually, sata raid and scsi raid drives, and, well, I just don't see any way. Maybe it it was a 24'" cube.
Even with water cooking the other components still need good airflow. And I've never been the cram 4 cf of stuff in to 3 cf type - usually I go the other way and use oversize cases for both air flow and ease of assembly and maintenance. I'll leave the building of and living with cram packed systems to others.
Bur .... I really wish I could use the UFO-2 because it is a great case.
MaxxxRacer
06-06-2005, 01:24 PM
well based on what u just said i dont know what case u could use.. lol.. possibly the lian li pc75?
Actually that was the reason for two used cases. Take the sode panel, backplane etc. out of one and bolt them together to make a double wide.
At least one person here does that but I don't have the link handy so I'll use this (http://www.spodesabode.com/content/article/cubecase) as generic example.
I haven't given up on the UFO 2 - friend of mine that does a lot of radical - in the sense of moving internal stuff around to allow ofbeat combos to work - is looking into it as well.
bxf -
MaxxxRacer
06-06-2005, 02:10 PM
ok cool. sounds good. let us konw what happens.
Definitely will do.
It'll take some time - get monster/shroud, get case mods done around the monster core, finalize mb selection, check into possibilty of using a single ps with independant 'on' circuits for cooling and mb, decide whether to go with res/fillerport combo or fillerport only, check into custom LCD readouts with dedicated display for each cpu plus air temps into and out of the monster core, etc.
Then there's waiting for the anticipated huge price drop in Xeon prices while also monitoring developments on the AMD side.
All in all I'm anticipating October or November time frame. That'll also give me time to get a site going so I can post an article there about the whole thing, choices made and reasons and final result.
Plus keeping up with new stuff here as well as going back thru old posts - you never know what tidbit you might stumble accross that you'll find a use for later. I first saw the article about the DIY double wide case in early 2004 for example.
bfx
MaxxxRacer
06-07-2005, 02:07 PM
ok sounds cool.. u sure have a long time frame for this..
personally i say dual core opteron.. drool.
Yeah - I hear you.
But so does my bank account - and I hate when it starts crying. Makes whatever is in there all soggy. lol
I hope to be able to contact weapon by the end of this month. From other posts here I'm figuring a 3 to 4 week delivery time for the 'heatercorus rex' plus shroud - maybe more since he's busy.
By then the cpu price cut should be near - knowing it's coming I can wait. During that time the case can be built around the cooling system placement, etc.
When the price cut hits I'll order all the electronics at the same time. From that point things should move fairly fast.
I haven't given up on Opterons. I seem to be migrating away from 3ds max toward other apps with code more balanced for both Intel and Opteron cpus. So a Tyan K8WE mb with single core cpus is still a different option. Swap in dual cores after the price drops a bit.
nfx
MaxxxRacer
06-07-2005, 11:14 PM
teh swapping out dual core thing sounds like a good option. actually i know what u shoudl do. get the tyan dual board, buy one dual core cpu and later on get another dual core opteron. seems the most logical and least costly to me.
That would be interesting (and probably frustrating) to try. :)
Different stepping, different cpu speeds, etc.
Naw - I'm already in danger of going bald from pulling my hair while trying decide between AMD and Intel. :) :)
bfx
MaxxxRacer
06-08-2005, 12:21 PM
i dont think it would be frustrating.. in the time that u would upgrade the stepping would most likely not change and if it did it wouldnt be toa different architecture. and even if it did it would just mean a cheaper cpu for u if u bought the older stepping.. win win id say..
Think I'll pass on that. :)
bfx
MaxxxRacer
06-08-2005, 03:53 PM
lol.. hey i offered.
1 - Should I switch to dual swifty Storm 4 blocks instead of TDX?
2 - Since they don't some with mounting plates would somethinh lke this (http://www.servercase.com/miva/miva?/Merchant2/merchant.mv+Screen=PROD&Store_Code=SC&Product_Code=DualXeon-800+CPU+kit&Category_Code=Acce.)work?
2 - Are they too big to fit on a Supermicro serverboard?
MaxxxRacer
07-01-2005, 01:28 AM
it does come with mounting plates... check the site to make sure its fits ur socket.
id go with the swifty's over the TDX anyday.
Sorry - I meant swiftys. I'd been awake 22 hours when I posted.
I'd already checked the site - apparently it'll fit the Nocona - or at least a single Nocona - but doesn;t come with retention hardware.
Guess I'll have to wait and do final check on retention when order time comes.
The Storms sure seem like a good deal though. I'll send an email to find out future plans for retention devices. May have to start with the swiftys and convert later.
bfx
Just an update -
In a different thread I'd mentioned the possibility of using an automotive-type electric fan ro cool the monstercore.
After a bit more research into using a single fan I think I have a good solution.
Dayton A1259-MBT
CFM - 665
dBA - 52
Voltage - 115 AC
Hz - 50/60
RPM - 1600
Watts - 23
Amps AC - 0.23
Bearings - Ball
Depth (In.) - 3 1/2
Motor Type - Thermally Protected PSC
Housing Material - Cast Aluminum
Mounting Hole (In.) - 9 11/16"
Dia. (In.) - 10"
Wire Lead Size (In.) - 27" Leads
Widely available.
Cost range - $65 - $75
665 cfm @ 1600 rpm and 52dBA with 0.23 amp/115vac for $65 sounds pretty good to me.
Sunon also makes a A1259-MBT fan that's very similar that retails around $76 - $85 but it's specs are identical. And it's harder to find.
The only disadvantages are size and weight - ~ 4 lbs. But I'd already determined a UFO 2 case would be too small whatever I did and decided on a custom built case anyway. For shrouding/exhaust ducting an HVAC shop should be able to whip something up quickly and relatively cheaply.
Plus an AC fan and an AC Iwaki pump are natural mates. :)
bfx
lv_dicedealer
07-01-2005, 04:05 PM
52dB is acceptable ????
You might want to rethink that
Absolutely - compared to what the noise levels would be for four fans providing that much air flow.
And LOTS quieter than my systems that run 5 to 7 92mm van tors cranking full speed in the summertime.
Not to mention those two hurricane fans also running in the summertime for general house cooling.
I can sleep thru that so 52 dBA is nothing.
bfx
MaxxxRacer
07-01-2005, 09:17 PM
lol.. u can slee through taht?!?!?
I wouldnt be able to think with that.
I don't sleep in that room all the time - just when I need to switch apps during sleep time. I'll set an alarm clock - which I can easily hear over the background white noise - do the switch then back to sleep. Somehow I've always been able to tune out repetitve noises like that.
I've mounted the van tors inside boxes with a bit of sound insultion on both the inside and outside plus used 1/8" thick silicone pads to mount both the fans and the boxes. They're not that much quieter but it lowers the freq a bit. Makes the family dog and cats a lot happier. You should have seen their exit stage left routine when I fired up the first box w/u the insulation.
I'll do the same the Daytons. Should work even better considering it's lower rpm.
bfx
id say keep those waterblocks either way. but if u wanted to upgrade the pump i would suggest the WMD-20RLT or RZT. im not really sure if the RZT would be a better choice compared ot the RLT as far as restriction goes. I say teh W as it is cheaper than the jap motor. but if u can go for the jap motor u might as well. its 190 vs 110 for the american.
Recap -
Dual mw6002 cpu blocks - one maze 4 gpu block - monstercore - 1/2" tygon
Questions -
1 - Does the recent testing of a 30RZT tilt the choice in favor of a 20rz/20rlrzt? btw - does the 'L' signify 115/120 volt motor?
2 - wmd vs. md - any advantages to the md over wmd 20's? Maybe smaller motor size and less weight for the md ? Different size ports?
Thanks
bfx