Pages:
Author

Topic: VRM temperatures (Read 6410 times)

sr. member
Activity: 322
Merit: 250
October 09, 2011, 09:10:16 AM
#22
Exactly. And even if you could measure the chip below the heatsink, you'd be measuring the surface temperature of the package, not the internal temp. I wouldnt expect a big difference (like when measuring the heatsink), but nothing will beat the precision of the internal sensors. IR is a somewhat useful alternative if you have a card which doesnt allow sensor measurement of certain chips, like VRMs, and will give you an idea,  but its a silly substitute if you do have such sensors.

Anyway, the mystery of my 5850 has been solved. It died. It got ridiculously sensitive to voltage adjustment, just increasing or decreasing the vcore by 0.01v made like 5C difference, as if the heatsink wasnt mounted properly (it was).  Soon after I started getting artifacting, blue screens and now it no longer boots. I suspect electromigration from accidentally pushing 1.65v through it because of a GPU-Z/everest bug killed it and no amount of aftermarket coolers was going to save it.  I tried baking it in a over too btw, but no dice. Its dead.

Unlucky man, its a sad day for miners when one of these great cards die. Maybe it was them heatpipes not doing there job. There few about these days also Sad New 7800's soon which i think will go down in price when the 7900 series is released next year. I dont see the 7900 being a better miner than the 7800 series due to the new architechre. Lets hope the 7800 becomes the 8800 and so on if my suspicions are correct about the new design Smiley

As for them IR, the better you buy, the better tyou get. All i can say. And if your unsure, Add 5-10 degrees, dependant on your 'cautiousness'. I got handheld thing used by mechcanics and its shows hotter than GPU-Z. I know what i believe. IR is 36 degrees bob on everytime in my ear.
hero member
Activity: 518
Merit: 500
October 09, 2011, 03:40:38 AM
#21
Exactly. And even if you could measure the chip below the heatsink, you'd be measuring the surface temperature of the package, not the internal temp. I wouldnt expect a big difference (like when measuring the heatsink), but nothing will beat the precision of the internal sensors. IR is a somewhat useful alternative if you have a card which doesnt allow sensor measurement of certain chips, like VRMs, and will give you an idea,  but its a silly substitute if you do have such sensors.

Anyway, the mystery of my 5850 has been solved. It died. It got ridiculously sensitive to voltage adjustment, just increasing or decreasing the vcore by 0.01v made like 5C difference, as if the heatsink wasnt mounted properly (it was).  Soon after I started getting artifacting, blue screens and now it no longer boots. I suspect electromigration from accidentally pushing 1.65v through it because of a GPU-Z/everest bug killed it and no amount of aftermarket coolers was going to save it.  I tried baking it in a over too btw, but no dice. Its dead.
donator
Activity: 1218
Merit: 1079
Gerald Davis
October 08, 2011, 06:25:57 PM
#20
You are aware that IR thermometers works by detecting infared light emitted by hot objects.

The heatsink is opaque to IR light thus the IR thermometer can't "see" the GPU.

Now the temp of heatsink may be similar to temp of GPU but you are reading the temp of heatsink no the GPU.
sr. member
Activity: 322
Merit: 250
October 08, 2011, 06:09:10 PM
#19

It's infared, it reads UNDER the heatsink to the hottest part, i.e, the max temp its giving out. Clever and simple devices Smiley


Oh really? I guess it looks right through the case too then...   Roll Eyes

no, just things close enough to be in range. I dont know if that was sarcastic so i took the time to google it for you....

http://en.wikipedia.org/wiki/Infrared_thermometer
hero member
Activity: 518
Merit: 500
October 07, 2011, 04:13:24 PM
#18

It's infared, it reads UNDER the heatsink to the hottest part, i.e, the max temp its giving out. Clever and simple devices Smiley


Oh really? I guess it looks right through the case too then...   Roll Eyes
sr. member
Activity: 322
Merit: 250
October 07, 2011, 12:51:58 PM
#17
And something along the lines of these is what you use for ACCURATLY MEASURING TEMPERATURE....
.

Great, so with one of these you can accurately measure the temperature of the heatsink. How do you measure whats below the heatsink?

Quote
As for my 5850's There Sapphire 5850 extremes (http://www.overclockers.co.uk/showproduct.php?prodid=GX-259-SP&groupid=701&catid=56&subcat=1711 (£160 they went to in 1 month from £100, i got them at £100)) and they all do 1000/300 @ 1.193V, i wont go above 1.2V cause of VRM's on this card. At that they are hitting 100+. SO i lowered clock to 965(975)/300 @ 1.187V which keeps them well below 90 degrees and cores run all below 80 degrees. All get over 400 m/hash at that speed and that was my target with them. YOu get squeeze 420430 quiet easily out of them Smiley

I must be doing something wrong. IN fact, I suspect I killed my accelero by bending it (even if just a tiny bit). The rear 2 heatpipes dont get hot, the front 2 get smoking hot as they should. Not sure it was like that before, but I think not.  Temps are still okayish, but just upping my voltage from a 1.088 to 1.100v causes my temps to rise 5C. Hate to think what 1.19v would do. Ill try another cooler later.

VRM temos are no longer a problem though. The R5 totally rocks. Nothing I do gets the VRMs above 60C, and I havent even bothered to put a fan on it.

It's infared, it reads UNDER the heatsink to the hottest part, i.e, the max temp its giving out. Clever and simple devices Smiley

I looked into them heatsinks too and they are sick, but use too much space and too costy, i'd go water cooled instead and get better cooling all over the card instead. But the cost again goes up and up. And the water blocks will not be compatible with newer cards. But the gain will be alot more significant than just H/S's
hero member
Activity: 518
Merit: 500
October 06, 2011, 06:10:56 AM
#16
And something along the lines of these is what you use for ACCURATLY MEASURING TEMPERATURE....
.

Great, so with one of these you can accurately measure the temperature of the heatsink. How do you measure whats below the heatsink?

Quote
As for my 5850's There Sapphire 5850 extremes (http://www.overclockers.co.uk/showproduct.php?prodid=GX-259-SP&groupid=701&catid=56&subcat=1711 (£160 they went to in 1 month from £100, i got them at £100)) and they all do 1000/300 @ 1.193V, i wont go above 1.2V cause of VRM's on this card. At that they are hitting 100+. SO i lowered clock to 965(975)/300 @ 1.187V which keeps them well below 90 degrees and cores run all below 80 degrees. All get over 400 m/hash at that speed and that was my target with them. YOu get squeeze 420430 quiet easily out of them Smiley

I must be doing something wrong. IN fact, I suspect I killed my accelero by bending it (even if just a tiny bit). The rear 2 heatpipes dont get hot, the front 2 get smoking hot as they should. Not sure it was like that before, but I think not.  Temps are still okayish, but just upping my voltage from a 1.088 to 1.100v causes my temps to rise 5C. Hate to think what 1.19v would do. Ill try another cooler later.

VRM temos are no longer a problem though. The R5 totally rocks. Nothing I do gets the VRMs above 60C, and I havent even bothered to put a fan on it.
sr. member
Activity: 322
Merit: 250
October 06, 2011, 04:22:11 AM
#15
Software isnt reliable, Buy a infared laser detector thing. MUCH more accurate and reliable.

I dont doubt the measured temps. My finger agrees Smiley. Besides, how else will you measure temps underneath a heatsink?

Quote
I use loads of 5850's all next to each other all clocked 950/300+ with no problems. THey all do 1000 (except 1) but i thought of wear on them and lowered them. 400 a card is fine with me Smiley

At what voltages and temps? Im currently at 875 Mhz with 1.1v and despite the rather elaborate cooling, Im getting near 70C on the GPU. To go beyond 900 MHz I need a voltage bump and just bumping my voltage to 1.15 causes my temps to rise to 75+C on the GPU as recorded by catalyst (which means 85+C on other areas of the GPU) and I dont like that. Im sure it will survive for a while, maybe even years, but it seems like a stupid bet for a few extra MH (particularly since the electricity cost increases disproportionately with vcore and higher temps).

Your finger? you relise that 60 degrees celcious is getting too much for your finder to accuratly measure? haha Jesus. I hope your joking. And something along the lines of these is what you use for ACCURATLY MEASURING TEMPERATURE....

http://www.ebay.co.uk/sch/i.html?_nkw=infared+temperature&_sacat=0&_odkw=infared+laser+temperature&_osacat=0&_trksid=p3286.c0.m270.l1313

Point and click. I got one, and it reads 36 degrees everytime in my ear. Perfect Smiley More you pay, the more 'distance' you can be away from the actual thing your measuring. Mechanics and all sorts use them for managing engines and all sorts.

As for my 5850's There Sapphire 5850 extremes (http://www.overclockers.co.uk/showproduct.php?prodid=GX-259-SP&groupid=701&catid=56&subcat=1711 (£160 they went to in 1 month from £100, i got them at £100)) and they all do 1000/300 @ 1.193V, i wont go above 1.2V cause of VRM's on this card. At that they are hitting 100+. SO i lowered clock to 965(975)/300 @ 1.187V which keeps them well below 90 degrees and cores run all below 80 degrees. All get over 400 m/hash at that speed and that was my target with them. YOu get squeeze 420430 quiet easily out of them Smiley

hero member
Activity: 518
Merit: 500
October 05, 2011, 03:05:13 PM
#14
Software isnt reliable, Buy a infared laser detector thing. MUCH more accurate and reliable.

I dont doubt the measured temps. My finger agrees Smiley. Besides, how else will you measure temps underneath a heatsink?

Quote
I use loads of 5850's all next to each other all clocked 950/300+ with no problems. THey all do 1000 (except 1) but i thought of wear on them and lowered them. 400 a card is fine with me Smiley

At what voltages and temps? Im currently at 875 Mhz with 1.1v and despite the rather elaborate cooling, Im getting near 70C on the GPU. To go beyond 900 MHz I need a voltage bump and just bumping my voltage to 1.15 causes my temps to rise to 75+C on the GPU as recorded by catalyst (which means 85+C on other areas of the GPU) and I dont like that. Im sure it will survive for a while, maybe even years, but it seems like a stupid bet for a few extra MH (particularly since the electricity cost increases disproportionately with vcore and higher temps).
sr. member
Activity: 322
Merit: 250
October 05, 2011, 01:56:04 PM
#13
Software isnt reliable, Buy a infared laser detector thing. MUCH more accurate and reliable. I use loads of 5850's all next to each other all clocked 950/300+ with no problems. THey all do 1000 (except 1) but i thought of wear on them and lowered them. 400 a card is fine with me Smiley
legendary
Activity: 2026
Merit: 1005
October 05, 2011, 05:10:58 AM
#12
DeathAndTaxes
may be your termal sensor is broken?
I`m not getting +100 C at any of my cards...
hero member
Activity: 518
Merit: 500
October 05, 2011, 01:07:52 AM
#11
its called "AMD GPU Clock tool". Word of warning, it doesnt like other apps, dont use alongside afterburner everest, gpu-z etc, unless you like 1.65v vcore. It also reset fans to default whenever you change clocks
donator
Activity: 1218
Merit: 1079
Gerald Davis
October 04, 2011, 05:58:19 PM
#10
What utility is that you are using?
hero member
Activity: 518
Merit: 500
October 04, 2011, 05:16:12 PM
#9
Some pics:






And some temps mining at 800 Mhz. Temps are even lower now I closed my case and the side intake fan is helping:



Is that awesome or what Smiley
hero member
Activity: 518
Merit: 500
October 04, 2011, 05:11:21 AM
#8
My thermal take VRM R5 arrived yesterday. Installing it in combination with my accelero twin turbo was a bit of a nightmare, as the VRM block doesnt fit underneath the accelero. Had to grind the R5 down, and bend the accelero a little bit to make room.

However, it does work incredibly. I havent even fitted a fan to the R5, and I likely wont as the VRM temperatures dropped 30-40C ! My VRMs are now cooler than my GPU. Awesome Cheesy

My 5850 is now testing at  890 MHz, 1.1v, VRMs are 59C.  Me happy Cheesy
legendary
Activity: 4354
Merit: 3614
what is this "brake pedal" you speak of?
September 25, 2011, 06:15:39 AM
#7
Seems like people here regard their VRM temperatures as trade secrets Cheesy ?

the cards I run (see sig) think its a secret, not one reports it Sad

I make sure minimum fan is 40% (like they ever actually get that low, most sit at 60-85%) and they have a good supply of fresh air. the 6870, my hottest running card, is slightly undervolted: 1.165 vs stock 1.200. other than that..

that  VRM HS is pretty serious. please post some pics and temp results when its in. 

EDIT: that link is for capacitors not VRMs. although as a rule the caps do sit right next to the VRMs and soak up some of their heat.
hero member
Activity: 518
Merit: 500
September 25, 2011, 03:17:03 AM
#6
Seems like people here regard their VRM temperatures as trade secrets Cheesy ?

I just stumbled upon this:
http://event.asus.com/mb/5000hrs_VRM/

Yes, its for a motherboard, but I suspect the VRMs are the same technology.
"VRM 5000hrs lifespan @105°C, 500,000hrs @65°C"

Seems like 100+C wont kill it outright, but temps have an enormous effect on lifespan.
So, Ive just ordered a Thermaltake VRM R5:
http://www.xbitlabs.com/articles/coolers/display/thermalright-spitfire_4.html

The results seem pretty spectacular. We'll see how it goes.
hero member
Activity: 518
Merit: 500
September 23, 2011, 12:37:46 PM
#5
I just switched to some watercooling and 3x 5970 have VRM running @ 63C

There are waterblocks that cover ram and vrms too? Nice!
But Im too chicken to use watercooling, so Ive been searching all over for "plates" that cover VRMs and/or ram heatsinks that have fins to use the airflow provided by my accelero cooler, I just cant find anything. Well, I saw precisely one, but it was for an older nvidia card IIRC. I find that odd. Surely not everyone using aftermarket coolers wants to rely on glued alu sinks that always fall off?

Anyway, I think Ill pick up a thermaltake VRM V4 or V5 to take care off the VRMs and cut my stock baseplate in half so it still covers the ram modules.
donator
Activity: 1218
Merit: 1079
Gerald Davis
September 23, 2011, 12:00:39 PM
#4
My aircooled 5970 (Core @ 920) runs 80C.  A pair of 6950s (unlocked and core @ 900) run 85-90C.  Both are with 25C ambient temps.  I don't have access to the other machines to check.

I just switched to some watercooling and 3x 5970 have VRM running @ 63C (max over 48 hours).  Grin  That is just at Core 880 taking it slow as that is a lot of thermal energy to dissipate.

A lot may depend on the cooling design and be board specific.  The 5970 reference design for example has an outlet "duct" from the fan that diverts some of the cool air directly across the VRM.  Since the fan is located at the VRM "end" of the card the coolest air actually hits the VRM (compared to a design with fan in the middle).
hero member
Activity: 518
Merit: 500
September 23, 2011, 11:43:15 AM
#3
Well, they do have protection that throttles their temp to around 130C; I noticed that when a GPU-Z/Afterburner bug caused my voltage to spike toa  whopping 1.65v. Which I only noticed after 30 minutes of roasting my card (ouch).  I suppose AMD set that limit at a point they are designed to handle, and they did handle 130C and 1.65v for 30 mins so I honestly dont know if 100C at regular voltage is a problem or not.

FWIW, the GPU's throttle at 100C, yet most people seem happy running them just 15C below that.

Anyway, Im mostly curious what VRM temperatures you guys that run your cards at 800-900-1000 MHz actually have. If they are not way higher than mine, then Im doing something wrong Smiley
Pages:
Jump to: