Efficiency issue?

[Etacovda]

I'm not entirely sure how to word this in order to not get flamed by jaydee

My block I constructed, as in my previous thread, is completed and running on my testing system. After some tests, with a chaintech 7NJL1 (?) and a T-bred A 2000XP, I think I've hit a design flaw in my block. Basically, a huge temperature rise from 0.1 increase in voltage. This is all LM90 based, on the motherboard, and not the best testing method i realise (i am picking up a multimeter with a K type thermocouple shortly, which should be a bit closer to realistic), but I was wondering what a quick analysis from you guys would come to.

At 1.92ghz, 1.85vcore, i was seeing a 37 degrees under load.
At 1.92ghz, 1.95vcore, i was seeing a 42 degree under load temp. This shot up pretty damn quickly too.
Load was achieved using 2 instances of prime95 TT.

Im thinking its base thickness, as its only about 3 mm. The block is pretty low restriction, with my full set up connected (rad, res, block, pump) I'm seeing 1.5gpm, roughly. I think I'm right in thinking that the thickness is reducing the efficiency of the block, but was wanting a more experienced opinion.

[bigben2k]

I'm not sure what your block looks like, but 3 mm is rather thin for a simple-low flow block design. A 5 C temp is not unrealistic, given that.

I see... this is a #Rotor style design. Maybe #Rotor has some insight, for baseplate thickness?

What pump did you end up using?

[jaydee]

Originally Posted by Etacovda

I'm not entirely sure how to word this in order to not get flamed by jaydee

Hehe, At least people are reading.
Doubt it is a block design issue. .1V shouldn't add 5C at the same MHZ setting on any block no matter how poor it is. Probably a Chaintech design issue.

[Groth]

A water temperature (or ambient) would be handy. If I remember correctly, lm90 + Tbred has an error of +6/-4, nearly all of which is systematic error (the offset will be constant for a given lm90/Tbred pair). You can roughly determine that error by plotting delta-T (CPU minus water) vs multiplier (keep Vcore, FSB, etc. constant).

Prime95 isn't a good heat loading program, and running two instances just increases OS overhead and risks cache thrashing. Use K7burn.

A good possibility for the excessive heat increase is that your CPU is at the edge of stability (a distinct possibility at 1.92 with a Tbred A). At the lower voltage the CPU is doing less work than expected due to cache ECC errors and is idling during the subsequent calls to memory to fetch the correct values. At couple more voltages above, between, and below your two data points could tell the tale.

[pHaestus]

I just read a review today that suggested the Chaintech diode/voltage monitoring hardware wasn't very good.

[Etacovda]

Hehe, At least people are reading.
Doubt it is a block design issue. .1V shouldn't add 5C at the same MHZ setting on any block no matter how poor it is. Probably a Chaintech design issue.

Yeah, I thought it might be that (I knew that would get a giggle too ) - I also read that PH had issues with his test bed when his PSU was doing some wierd things. This psu (very cheap 300w version) is doing some weird things, I think that 1.95 was too much stress for it - in fact, im certain of it, because the ATX wires were pretty damn warm... (wtf?). That, and I don't trust chaintech myself, and would never buy their mobo's or touch them with a 10 foot pole - wasnt my board or cpu.

A water temperature (or ambient) would be handy. If I remember correctly, lm90 + Tbred has an error of +6/-4, nearly all of which is systematic error (the offset will be constant for a given lm90/Tbred pair). You can roughly determine that error by plotting delta-T (CPU minus water) vs multiplier (keep Vcore, FSB, etc. constant).

Prime95 isn't a good heat loading program, and running two instances just increases OS overhead and risks cache thrashing. Use K7burn.

A good possibility for the excessive heat increase is that your CPU is at the edge of stability (a distinct possibility at 1.92 with a Tbred A). At the lower voltage the CPU is doing less work than expected due to cache ECC errors and is idling during the subsequent calls to memory to fetch the correct values. At couple more voltages above, between, and below your two data points could tell the tale.

Well, I dont really have access to any decent thermal equipment, I'll have a look around tonight (not my house, dont want to poke around too much ). One of my cases comes with a front sensor with 3 sensor inputs, so I might check that next to a mercury thermometer with some warm water and see if its fairly accurate or not. Its very responsive, not that that means much. I'll be investing in a multimeter with K type thermocouple... if you go to www.dse.com.au and type "k type" in the search field, it'll come up. They dont allow direct linking, unfortunately (well, not that i can get to work).

I normally use BurnK7, but this was done in a bit of a rush. Hell, I don't even know if my seating was alright or not, but the thermal paste print didnt look off, It looked ok. If the mounting pressure was too low, would could this be a symptom of it?

I'm not sure what your block looks like, but 3 mm is rather thin for a simple-low flow block design. A 5 C temp is not unrealistic, given that.

I see... this is a #Rotor style design. Maybe #Rotor has some insight, for baseplate thickness?

What pump did you end up using?

Its a semi-blend of designs; I didnt feel right completely copying designs, I aimed for a high turbulence above the die with an easy route out of the block. The next block will be considerably different, I think - probably a bit more restrictive and turbulent.

The pump is similar to an eheim 1048 in performance. Not particularly impressive, but I measured 1.5gpm though the entire loop, which I dont think was too bad. After looking at Ph's performance curves for WB 1.5gpm seems a good enough area to be aiming for?

I just read a review today that suggested the Chaintech diode/voltage monitoring hardware wasn't very good.

Yeah, i gathered this might be the case. I'm getting a DFI lanparty Ultra B NF2 board this week, but I think they're thermistor only, which is a bit dissapointing. Not entirely sure though, haven't done a whole lot of research yet.

Thanks for the help guys, hopefully with some better equipment I'll be able to see more clearly where everything is at.

[bigben2k]

1.5 gpm would be ideal.

If your PSU wires get hot, that'll be because the power connector on the mobo is making poor contact. You should focus on that (use conductive grease, tighten up /re-seat the connectors,etc...). The heat then travels up the wires.

In very rare cases, the wire itself can be overloaded, but I don't believe that to be the case.

[Etacovda]

Wow, thats pretty interesting, I never really though of bad contact as a clue.... With the next rig I'll be setting up on (my actual rig, I've been waiting for this for a while) I'll be eliminating all of those problems, I hope. I was talking a VERY dodgy 300w power supply, so I wouldn't be surprised if it was a combination of bad rails, voltage fluctuations and a bad connector - it only had 10A on the 12v rail, and its made by augira

At present my 2500+ does 2.35ghz under air pretty easily on 1.75vcore - ive had it boot into windows at 2.5ghz on a cold day at 1.85vcore. I'm hoping for a 2.5ghz at around 1.9vcore stable with water, that would be excellent. This will be on a DFI board with an enermax psu, both are fairly new so I'd imagine the psu and mobo ATX connectors shouldn't have any issues. With this DFI board, im not to willing to start butchering anything just yet - I have a ka266 motherboard that I'm going to do the ZIF mod on, and I'll look at getting a second hand 1700+ or something similar, for voltage/temps etc testing. I'd like my next block to perform quite a bit better (a degree or two would be nice, I think that won't be hard as the this block isnt really even close to what it should be) and im not willing to trust motherboard temp sensors for my results. I think the K type thermocoupled multimeter will be good enough for comparasions sake.

Thanks for the help

[|kbn|]

see if the mosfets are hot, and how much variation in cpu voltage there is when loaded and unloaded.. I found over 1.8v on my nf7-s that voltage would get very unstable without mosfet sinks, causing the cpu to not get nearly enough power..

Also when you increase voltage remember it also increases current, so 0.1v means it gets quite a bit more power, and could probably add 5-10w depending on the cpu..