Testing on a budget

[Etacovda]

If you were testing on a low budget, and had to do your testing on a rig, how would you prefer to do it?

At the moment, i have a duron 1100 with an old iwill board, and a 2500+ with an Abit board. The duron is probably a silly idea, i gather - the bios will go up to 1.85v setting, with a reading of slightly higher than that at 1.89v. Is it worth testing the duron?

I'm going to buy a multimeter with a k-type thermocouple; there also a cheaper option that reads temps to +/- 1.2%, at half the price. Being as illiterate as I am with electronic gear, heres a link

http://tinyurl.com/2a348 (tinyurl is very handy)

The basic idea was to use the meter in conjunction with two sensors, if possible. If not, theres a k-type meter i was looking at for about $50NZ. One measuring the water temp at the inlet, using a t line piece, with the sensor siliconed in place.

The other probe, I'd probably be gluing/siliconing in place against the cpu core, or alternatively going onto the back of the core and doing a socket mod. I know that Phaetus uses a soldering trick to get directly to the cpu diode - I'd prefer not to do that to the 2500, and AFAIK the duron doesnt have an onboard diode?

I realise that im not going to get neccessarily an accurate result - what im hoping for is some form of repeatability. I have some waterblocks to test - i have the best one to test (lucky enough to borrow a cascade). I'm going to look to get a maze4 or similar for a 'worst' testing piece, so I can get a decent 'feel' for where everything is at, and how my set up would be calibrated... as such. I'll be sticking with the one mounting kit through out the testing... is this wise?

For flow rate, Im going to look at a valve of some description - the best i can do cheaply at the moment is a ball valve, but im sure i could source something else. Keeping below say $100NZ would be a great thing, but i realise it will more than likely get more expensive than this very quickly.

I have an airconned room for temp control that is breeze free.

Advice? (yes, im waiting for the flames... this is a personal thing, so i can get a 'feel' for where im at, and it sure beats reading a motherboard diode and taking it as gospel, im sure)

Thanks in advance.

[Groth]

A gate valve (cheap) or a needle valve (not cheap) will do better for flow modulation. Balls are hard to control, they want go all out or do nothing at all. :D

What kind of board is the 2500 in? Does it read the internal diode? If not, maybe a minor mod would allow it to. Multiple temperatures reading (thermocouple, diode, socket thermistor) would allow you to spot oddness easier. And yeah, I think your Duron is from before AMD started putting thermal diodes in all their chips.

T thermocouple are a bit more accurate than the K flavor. More expensive too. Just a matter what you consider worth it. I didn't check the specs for the meter you linked to, tinyurl is blocked at the firewall (too many bad people disguise their links with it).

[gone_fishin]

Balls are hard to control, they want go all out or do nothing at all. :D



Get your test rigs Dr. Ruth certified :D

[Etacovda]

The board is an NF7-S; unfortunatly, unlike my old soltek, it doesnt read the diode temp but i have been looking around for a 'hack', which im yet to find.

After seeing that the range is only in one degree C increments, and not really the resolution i wanted, I guess its pretty pointless as i have a 3 sensor thermometer here which will do the same job, i guess. Once you start getting into the 0.1's it gets expensive quickly, I'd imagine!

As far as gate and needle goes, I'll do a search tomorrow and talk to my engineering shop and see what they have :) flow control for this is in the loosest sense anyway - its basically going to be a 'bucket count' style thing. A ball would create back pressure though, from what ive seen of it, not a healthy restriction at all... Now, i realise that im not going to get within a great margin of error here, but as long as its good to 1(even 2) deg +/- then im happy; this is just to test my blocks, I'm not after test data that proves anything to anyone else ;)

g_f : lol!

[bigben2k]

http://overclockers.com/articles638/

http://www.microlink.co.uk/microlink/tctable.html

That's all I'm going to say, other than if you want to get into some serious testing, you know where I am. Also, the DMM you picked, with a 1.2% error margin, is +/- ~7 degrees, and that's just the DMM...

[pHaestus]

I like the Dallas One Wire temperature probes for inexpensive reasonably accurate temperature monitoring. 0.5C accuracy out of box, 0.125C resolution, and not affected by wire length or very sensitive to electrical interferences. Combine a DOW sensor epoxied under the CPU core, a few more for waterblock in and out water temps, a few for air measurement, and then dump everything to log files either with MBM or with Crystalfontz' software (if you use a CF633 for the DOW support). Add in a swissflow flowmeter and measure flow rate as well and you should be able to set it up a fairly simple to use digital testing bench for under $200. Use a spare PC (everyone has those right?) for monitoring the temperatures and logging and K7Burn for loading CPU on test box.

[Etacovda]

bb2k - eye opening; not anything I didn't expect however! I was wondering if the 1.2% was across the range or the reading itself, but I wasn't sure. Thanks for clarifying that - after using the meter yesterday, I realise its not what I was after - it didnt even 'feel right' when I viewing it in the shop. Of course, I'm talking about using it for relative measurements here, but I guess any range of factors could effect that kind of couple; with a 7 deg deviance id imagine it would be very hard to call it a 'constant'; not something that I would want to put any hope in.

pHaestus -
http://www.crystalfontz.com/products/633/index.html with DOW sensors? I'll have a test bed, plus two spare boxes, so computers shouldnt be an issue at all. I dont mind sinking some money into something like that - if and when I stop testing, theres still a good use for it :)
I can see this becoming an addiction very quickly.

As for a second PC, im going to be building a testbed rig, from 1700+, an Asus A7N8X-X (decent mounting holes + diode reading is always good; i think you were going to get one of these at some stage from reading old threads?) - and I'll have to look at getting a better PSU than the one I have now, as I dont trust it in the slightest.
I just sold a KT333 based gigabyte motherboard, and a family member has generously donated a 1700+ (I'm building him a block- how could he refuse :))
So thats well on the way. However, to my dismay, after heading out to my car yesterday, I found the windsheild nicely cracked; I think some water must have gotten under the seal and frozen, so that may put this endevour on hold.

WRT the duron test bed; I really dont want to use that, as the raised lettering worries me - makes it pretty difficult to put on thermal paste properly (for this, I was going to use AS alumina (ceramique) - any thoughts? I think it settles faster than AS3 or 5) and since this cascade is borrowed, theres no way I'd put it on the duron - its surprising how much those raised letters ruin a base!

So, the next step will be to get the ASUS board, as the 1700+ should arrive today.
The person I borrowed the cascade off seems in no great hurry, so hopefully I will be able to keep it a bit longer. I may have to resort to diode temperature readings at this rate - not something that I want to do.

Thanks for reading my novel, and thank you for the input - a huge help, as always :)

[Etacovda]

Ignore me, fixed my problem.

[greenman100]

I like the Dallas One Wire temperature probes for inexpensive reasonably accurate temperature monitoring. 0.5C accuracy out of box, 0.125C resolution, and not affected by wire length or very sensitive to electrical interferences. Combine a DOW sensor epoxied under the CPU core, a few more for waterblock in and out water temps, a few for air measurement, and then dump everything to log files either with MBM or with Crystalfontz' software (if you use a CF633 for the DOW support). Add in a swissflow flowmeter and measure flow rate as well and you should be able to set it up a fairly simple to use digital testing bench for under $200. Use a spare PC (everyone has those right?) for monitoring the temperatures and logging and K7Burn for loading CPU on test box.


found info on connecting to serial port

http://www.vss.co.nz/new/gadgets/onewire.htm

cheapest source for DS1820 is $5 each, from http://www.futurlec.com/Dallas/DS1820pr.shtml ????

am I missing something easy?

apparently MMZ_Timelord's project is not getting off the grounmd, would like to investigate other options

[DrMemory]

Using techniques described in the Dallas Semiconductor (now Maxim) data sheets, these temperature sensors are capable of 0.25 deg. C accuracy. I currently have one glued (with Arctic Silver thermal epoxy) to the bottom of my water cooled Athlon 1000, one glued to a small piece of copper tubing used to sense the water temperature at the radiator inlet, and one that is in a 5 1/4" drive bay front panel used to sense the ambient temp. A standard 8051 type processor is used to read up to eight of the sensors once a second, and update a serial LCD display also located in the drive bay front panel. Overall accuracy is excellent for temperature monitoring, and OK for serious testing. One thing to remember is that like thermistors, these sensors are not hermatically sealed and should not be in direct contact with water. A coating of water proof (thermal?) epoxy could be used to fix this.

[greenman100]

drmemory

what model sensor did you use?

[DrMemory]

DS18S20N in a T0-92 (3 pin transistor type) package for the water and ambient temperatures, and DS18S20Z in an 8 pin SOIC (VERY small/flat IC) package for the CPU sensor. These numbers for these devices have been changed to DS1821. As far as I know the specs are the same.

[greenman100]

so I take it you grooved you CPU socket?

any thoughts on software to read teh sensost, besides what was linked? anything MBM5 compatible?

Where did you get your sensors? looking for a cheap place without requesting free samples.

[DrMemory]

P.S. Here is a picture of the LCD display. Notice that I've also glued a sensor onto the C drive.

[greenman100]

also, where'd you get the microcontroller/display?

[DrMemory]

I bought the sensors at Digi-Key (digikey.com) for a little over $5 ea.. Digi-Key is not the best (least expensive) place to buy electronics but it was the only place I could find them, which is usually the only reason I buy things from them.

As for the software. I wrote the software for the 8051 microcontroller (in 8051 assembly) for this project myself. It only reads the sensors, converts them to ASCII text messages and sends these messages and some LCD control messages to the serial LCD display to show the results. There is no reason these results couldn't be sent to the PC through the serial port instead (and it would also give you something to use the serial ports for ;) ) and displayed in a pop up window say using Visual Basic. I have not written any software to do this. MBM5 looks for known sensor IC's on the SMBus (a system maintanance bus that uses a slight variation of the I2C bus) . SMbus headers have been on nearly all motherboards for several years now. I don't know if the newest generation of motherboards intend to include it or not. Since my setup is not common and does not use the SMbus, MBM5 would probably not see it. Even if the hardware was changed to use the SMbus, MBM5 would probably not recognize it. And since MBM is no longer being maintained, and the source code is not being released, there is no way to add it.

I have also designed and written the software for a variation that can read eight temperature sensors and sense eight fan tach outputs and output ALL of them once every second. It could also control (on/off) up to eight fans. The fan tach outputs require more math than the 8051 can handle in one second to convert them to RPMs, so this would have to be done by the PC. I have considered using serial, USB, and SMbus to get this information to the PC where it would be displayed on the monitor instead of an LCD display much like MBM5. The original plan was to market it, but this has gotten nowhere.

[greenman100]

wow, you have been a _gigantic_ help, I thank you for your time.

am trying to decide if I want the $50 kit MMZ_Timelord is marketing, or a few of these $5 sensors (more like 5)

Would probably just read through a VB window.

[DrMemory]

The microcontroller (actually an Atmel AT89C52) can be bought at Jameco (www.jameco.com) and Digi-key (www.digikey.com).

The LCD display is another story. The hardest thing to find was an LCD display that displays 4 lines of 20 charactures and fits in a single 5 1/4" drive bay. Most of them are mounted to PCBs that are too tall. There is no obvious reason why they are ALL this tall. I finally found a graphics serial LCD display module (also does charactures) from Scott Edwards Electronics, Inc. that was way too expensive (I think it was $99 in single quantities) but actually fit in the drive bay. I wanted to find out the manufacturer of the LCD display itself (Scott Edwards Electronics builds the serial to LCD display interface board that is mounted on another circuit board behind the LCD display) and use this module as a prototype so I bought one. One thing I found out is that it is hard to get enough usefull display messages on 4 by 20 characture display when trying to display 8 temperatures and 8 fan RPMs without a LOT of screens. Another thing I foud out is that it is hard to see an LCD display unless the PC is sitting on top of a desk. This is not where I put my PCs. These were two of the main reasons why the "next generation" was going to send the information to the PC to display. The final nail in the LCD display coffin was the math needed to convert all eight RPMs could not be done by the microcontroller in one second.

[greenman100]

bought two maxim 6656? ICs, $9 ea at newark

transistor sensors, 8cents each at newark

will interface with SMBus

thanks for the insight, though

[DrMemory]

Using IC's that directly connect to the SMbus is a good solution that requires a minimum of programming. The Maxim IC's may even be recognized by MBM5. I believe there is a list of hardware that MBM5 supports somewere on its web site (still available?). Good luckl!