the 4870 LEDS

[RunawayPrisoner]

If you got the second red led, which means power supply is cut, then it's your voltage regulators that just overheated past 105C and died. (When voltage regulators die, power supply is cut, and the card would then complain about power supply... naturally)

Sensors on the HD4800 series are only correct if heat is distributed equally among the components. When heat is not so equally distributed, then they will be whack beyond imagination. So anyway... definitely have a LOT, and I mean... a whole... friggin... LOT of cooling on the voltage regulators. Actually, DO think about putting ALL of your cooling supplies on them. The core and mem are not that hot at all. It was the heat from the voltage regulator that heated everything up close to the 100C range when you were using the stock cooler.

P.S.: Same thing happens on HD 4850, too, and especially when excessive voltage mods are applied. I'll do a full report with proofs (pictures, benches, temperatures) and some other things later. But for now, my advice for you is to keep that area cooled above all else.

Just to demonstrate how whack the temp sensors are, here's idle temps from a HD4850 with AS1 and a 120mm fan coupled with a 80mm fan in a 22C room:

[Shadout]

Noob question.
Anyone can show a picture with the voltage regulators on that need the heavy cooling. Getting 4870 on monday and putting on a S1 shortly after, also bought some extra ramsinks to be sure to cover everything possible.

[msarp]

Exactly what I was thinking.

Room temp is 77F and case is closed .I have 5 S-flex SFF21F fans and I put Chill Factor 2 on the chip.

[dunn2953]

Noob question.
Anyone can show a picture with the voltage regulators on that need the heavy cooling. Getting 4870 on monday and putting on a S1 shortly after, also bought some extra ramsinks to be sure to cover everything possible.

you do not want to use a S1 with the HD4870, it does not cool the voltage regulators and passive cooling will not be powerful enough to cool it

[Shadout]

Oh, I surely wont cool the card passively.
Shouldnt S1 be more than enough, just the voltage regulators which is an issue?

[xMrBunglex]

the S1 is plenty to cool the GPU, but make sure you have heatsinks and airflow over those VRM's.

[MadMonkey]

Check your VRM temperatures! They are circled in the GPU-Z pic:

Those temps are different diodes at the GPU and DO NOT show the VRM temp. If you really wanna see the VRM temp download Everest from Lavasys (latest build is 4.50.1448 BETA).

Well FurMark really stresses the shader engine hard resulting in quiet higher temperature and GPU load. If you are looking for 100% stability, FurMark is ideal tool for testing.
Otherwise use RTHDRIBL tool and crank up some FSAA to increase GPU load in a linear way (1x - 16x FSAA).

Exactly, Furmark will stress the card like no other program. Try Furmark and Everest and don't be surprised if your GPU VRM temp is more then 100° C. This and this only causes your card to crash. I am sure everyone crashing while using Furmark have VRM temps way beyond 100°C

[babalouj]

It is infact furmark that is causing the LED problem with my card. My GPU VRM temp is never over 50C and furmark crashes almost instantly so its not a heat issue. This is with and without overclocking.

[chawks2]

It is infact furmark that is causing the LED problem with my card. My GPU VRM temp is never over 50C and furmark crashes almost instantly so its not a heat issue. This is with and without overclocking.

What version of everest/tool are you using to measure the VRM temp? Is it possible you're getting bad readings?

I have the same issue with Furmark also, however, in my experience it has been partially related to PCIe bus speed. If I run Furmark 1.4 @ 102 (stock mobo settings) I hang between 50-70% @ 1280x1024 (All test done at 4X AA). If I bump it up to 105, i can complete Furmark @ 1280x1024, but hangs at about 50-70% @ 1920x1200(on occassion loses video completely). Based on above it sounds like the VRM is getting out of control, losing power with the psu, causing the video to no longer display.

Must check this VRM value when I get home, thx for the info guys!

[babalouj]

What version of everest/tool are you using to measure the VRM temp? Is it possible you're getting bad readings?

I have the same issue with Furmark also, however, in my experience it has been partially related to PCIe bus speed. If I run Furmark 1.4 @ 102 (stock mobo settings) I hang between 50-70% @ 1280x1024 (All test done at 4X AA). If I bump it up to 105, i can complete Furmark @ 1280x1024, but hangs at about 50-70% @ 1920x1200(on occassion loses video completely). Based on above it sounds like the VRM is getting out of control, losing power with the psu, causing the video to no longer display.

Must check this VRM value when I get home, thx for the info guys!

I am using the latest GPU-Z for PWM temps and checking the card for hotspots with my finger. Nothing is anywhere near 70C because it would burn me if it was. I have sinks on everything that is hot and 120mm fans pointed at the card. Furmark crashes almost instantly for me.

[chawks2]

Furmark crashes almost instantly for me.

Just for poops and giggles, see if changing the PCIe bus speed results in any improvement.

[MadMonkey]

I am using the latest GPU-Z for PWM temps and checking the card for hotspots with my finger. Nothing is anywhere near 70C because it would burn me if it was. I have sinks on everything that is hot and 120mm fans pointed at the card. Furmark crashes almost instantly for me.

GPU-Z does NOT read VRM temps. Please try Everest. Right now I use build 4.50.1448 BETA

[babalouj]

@ chawks2 & MadMonkey - Will do, I will report back with info as soon as I get home from work.

[deathman20]

GPU-Z does NOT read VRM temps. Please try Everest. Right now I use build 4.50.1448 BETA

Really? Whats the other 3 temp sensors on the board then?

[MadMonkey]

Really? Whats the other 3 temp sensors on the board then?

I am pretty sure the HD48xx series GPU has now 3 different spots(diodes) of the GPU itself measured (in GPU-Z). Like in Everest you have CPU/CPU Core #1/CPU Core #2

Btw, this is the Everest build I use http://www.lavalys.com/beta/everestu...d0xlfnyc3c.zip

[chawks2]

Without question the GPU VRM seems to keep increasing in temps when benching. I ran rthdrbl for about 10 minutes and my idle @ 60c went to 77 running @ 640x480.

EDIT

Where do I put the snowblower on this card to get this temp down?



HOWEVER, I changed the clocks for the heck of it to see the temp differences. Now I cant get the GPU VRM sensor back?!?!

[babalouj]

Yeah, that Vitek chip definitely needs cooling. My VRM temps are 95C+ and the Vitek chip burns my finger under load. Im going to slap some ramsinks on there.

**EDIT** I just sinked the pulse and vitek inductors and VRM temps in everest are still high, how high are VRM temps with the stock cooler? I think I am going to stick my stock cooling plate on there with my waterblock to see what that does for me.

[Kondik]

Stock is pretty High. around 80 - 90, with help of the Accelero S1 Mosfet cooling heatsing [ a bit modified by me so it can cover all 5 Mosfets ] + 2 additional classic RAM heatinks on Pulses i managed to lower it on 55 degrees on a ventilated card and 70 on a non ventilated card in load, in idle it's around [ 1str card ] 45 / 60 [ 2nd ]

[chawks2]

Thanks to everyone, my problem was my underweight PSU.

I know this is a thread tangent, but anyone have any recommendations, would prefer modular.

EDIT

Launched Furmark, ran 4X and failed w/LED D601 lighting up.(Crit PWR failure).
Unplugged 3 of my HDDs, ran flawless 6X and no LED. PWR prob for me, if you were having issues with temps, I would imagine the designate LED would light.

[babalouj]

Stock is pretty High. around 80 - 90, with help of the Accelero S1 Mosfet cooling heatsing [ a bit modified by me so it can cover all 5 Mosfets ] + 2 additional classic RAM heatinks on Pulses i managed to lower it on 55 degrees on a ventilated card and 70 on a non ventilated card in load, in idle it's around [ 1str card ] 45 / 60 [ 2nd ]

Nice, I am doing that now. I will post back with results.

**EDIT** destroyed card by removing two VRMs (half of each) while trying to take the epoxied ramsinks off them. I will be purchasing a new card tomorrow.

[Kondik]

Nice, I am doing that now. I will post back with results.

**EDIT** destroyed card by removing two VRMs (half of each) while trying to take the epoxied ramsinks off them. I will be purchasing a new card tomorrow.

Call that bad luck, For MOSFET's you can use the Attached heatsink that's in the pack with the accelero. As for the pulses i put there small primecooler heatsinks but the card is cooled by two case Noctua P12 1300 rpm, so even the lava hot pulses ale kinda .. warm.

[deathman20]

I am pretty sure the HD48xx series GPU has now 3 different spots(diodes) of the GPU itself measured (in GPU-Z). Like in Everest you have CPU/CPU Core #1/CPU Core #2

Btw, this is the Everest build I use http://www.lavalys.com/beta/everestu...d0xlfnyc3c.zip

Aaah wasn't sure on that thought it was just mislabbeled, makes sense though. Anyways dang No VRM sensor on the 4850's.

[chawks2]

destroyed card by removing two VRMs

Ouch. Hopefully you get your new card quickly.

[dnottis]

Mine does the same thing. Got 2x120mm fans on the backside of the card, 1 x 120mm on the top side. Everything is sinked. Where the F is the VRM sensor listed in the newer builds of Everest?? RTHDRIBL will push the VRMs to about 90, Furmark will push them to 126c. Using Everest build 1454. This was taken right before I went to a black screen.

[dennismagic]

Mine does the same thing. Got 2x120mm fans on the backside of the card, 1 x 120mm on the top side. Everything is sinked. Where the F is the VRM sensor listed in the newer builds of Everest?? RTHDRIBL will push the VRMs to about 90, Furmark will push them to 126c. Using Everest build 1454. This was taken right before I went to a black screen.

VRM 126c! what is the limit?