continuation from http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=2457&nowrap=true#49891

robertmiles
Send message
Joined: 30 Sep 09
Posts: 35
Credit: 193,794
RAC: 2,286
Message 49891 - Posted: 5 Jul 2011 | 1:51:27 UTC
Last modified: 5 Jul 2011 | 1:52:46 UTC
Is that model designed to have a fan that pushes air past the GPU, then out of the case, instead of circulating it within the case? When I switched from a model circulating the hot air within the case to a model using about the same power but blowing the hot air out of the case, the GPU temperature dropped about 30 degrees C.

Furl Hawkens
New member
Send message
Joined: 27 Jun 11
Posts: 5
Credit: 216,120
RAC: 14,845
Message 49885 - Posted: 4 Jul 2011 | 21:50:10 UTC - in response to Message 49874.

As for why I'm not using my 4890, only the 3850, its cuz the 4890 overheats. It always has unfortunately,

MW@H is very low dependant on memory-clock; that means, you can 'underclock' your card without significant loss of credit, but remarkable loss of heat. I use 'Afterburner' for that.

There were several discussions in the past. Many crunchers tested that, it works.

already tried. If I underclock my gpu, mem, or both, to points that are thermaly acceptable, I begin to get constant corruption of the data thats being processed.

I cant get it low enough, and have it stay stable enough, for DP calculations apparently...

Whats the CPU temperature (presumably without the card in and running, and no cpu overclock), and does the CPU have a stock or third party cooler?

How low did you have to take the GPU and GPU memory to get the heat under control, and what was that temperature ?

How old is the PC Case and what size is it? (make and model of the case would be good)

How many fans are in the case, front and back (plus others if there are any others) ?

Regards
Zy

What XFX has done here, this is an impertinence.
They sell an overclocked card, which you can make yourself, and take no measures in cooling. And then give no guarantee for the use. They even gamers who play 10 hours straight.

But back to your problem. The geometry of the PC-case also plays a major role. I have two HD 4850 and a Power Cooler Sapphire. The Sapphie is about 20 C hotter when they work alone. (55 or 75) When I installed both, then I have to rearrange the Sapphie so that it does not get hot air from the other card, but vice versa. Then the Power Cooler is also about 70 C warm.
Do you have on your board, a second PCI-e port? Then try to run the card once there.
You could also open nonused slots yet.

Did some digging on your card. The 4890 had a revised GPU, the RV790 - the latter was a redesigned RV770, the changes being to accomodate the higher clock speeds compared to the RV770, and cope with the additional stress/heat. The cooling is designed to throw the heat out the back of the case, not dump the lot inside, although of course some will go inside.

The card at full working order should run fully stressed @100% @76 degrees C, running 850/975. Heat was a real problem for 4870s, but not in 4890s, the new design resolved the 4870 heat hassles. Therefore if your card has widely differering temperatures at full stress and full fan, its either the case cant cope with the airflow needed, or the card is on its way out.

To narrow it down more, it will help enormously if you could post a screen shot of the sensor tab in GPU-Z with the card running at stock speed (initially to get some idea of how its running), and a second screen shot with some kind of load on it (again at stock speeds). If you dont have GPU-Z you can get it here : GPU-Z Download

Regards
Zy

As mentioned above the case cooling is important. Cases such as the Antec 300 have excellent cooling and are inexpensive (less than $50 on sale). I add 2 120mm fans to the stock config, one on the side and one the front, both blowing in. This will handle 2 high power GPUs. Another problem may be that the heatsink is not properly contacting the GPU or that there is not proper thermal compound.

I'm running 11 XFX GPUs (among others) but have always stuck to the 40nm chips for reasons of cooling and efficiency (HD4770 - HD5870). So far no problems with any of them (knock on wood). As far as your issue: if it's insufficient case cooling it's really not a card issue. If it's a problem with the HS or thermal compound then XFX should fix it. Have you tried opening up the case and pointing a small fan at it? If this solves the problem then you need better case cooling...

OK, let me try to give you guys some answers.

First off, I'm very PC-centric and know the issues are not related to other cooling systems.

The long and short of it is that XFX skimped out on cooling. They used inferior materials in the HS/F configuration. I actually questioned them, after pulling the AMD/ATi stock stated cooling system, as to why theirs was made with lesser materials (ie painted aluminum where AMD says there should be brushed copper...)

They told me they could not give an answer, but that they would not have made the card with improper cooling. yeah right lmao.

Stock speeds for my card are 875/975. With all settings on auto the driver will never raise the fan above 65% resulting in an overheat with stressful applications (like furmark, OCCT, looping 3D mark, or MW@H).

If I turn the fan up its insanely loud, and even @ 100% it will not keep the card cool enough (ie below 90c) when its running the above mentioned programs.

I can under-clock it, and the temps get under control, but I have to go down to 500/675 before it stays cool enough on auto fan control, and at that low of a setting I begin to get computational errors, which defeats the whole purpose...

As for other cooling, heres the full system


Apevia Full tower case, 5x120MM case fans (1 in front, 1 in top, 1 in side, 2 out back.

CPU is a Core 2 Quad Q6700 @3.6 with a coolermaster Hypere 212+, and 2x120mm (90CFM) fans in push/pull.

CPU runs ~30c idle, ~60c load

Northbridge runs 40c idle 50c load

GPU runs ~68c idle 85-97c load, sometimes passing 100c and crashing.

GPU VRM's (which are what I think is actually overheating) run at 80-90c idle, 100-130c load. The cooling is an out the back exhaust design, so not allot of hot air being recirculated.


I have already tried changing the TIM on the card (with XFX's ok), it helped about 3 degrees, but not much.

When I upgrade the card in a year or two I plan on making XFX warranty it out. I also hope that by then I might get something a bit nicer. 4890's will be old by then, and they have to replace it with either a 4890 or something "comparable in performance", maybe a newer card?...

OMG that sounds so horrible what you are telling us about XFX - So how much dit u have paid for that card??

...
can under-clock it, and the temps get under control, but I have to go down to 500/675 before it stays cool enough on auto fan control, and at that low of a setting I begin to get computational errors, which defeats the whole purpose...

:
This could maybe has another reason, witch maybe was solved read here:
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=2468&nowrap=true#49949

When I upgrade the card in a year or two I plan on making XFX warranty it out. I also hope that by then I might get something a bit nicer. 4890's will be old by then, and they have to replace it with either a 4890 or something "comparable in performance", maybe a newer card?...

Well when they give u a new card with "comparable in performance", today this would be an hd 6770 but this card has no Dopple Precision witch you don't need for gaming, but 4 MW :-(

Can you control the cards voltage, e.g. using MSI Afterburner? If you can lower clocks and voltage simultaneously you should be able to get the same power consumptio & heat output you're now having at 500 MHz at crudely estimated 700 MHz. On that card you should be able to run your memory at 400 - 500 MHz without problems. Most tools don't allow you to go that far down, though, which is really annoying. ATI Tray Tools can be configured to do it.

My HD4870 (approximately same power consumption at 750 MHz as your chip at 850 MHz) runs happily at 825 MHz (used to be 830) with an Artic Cooling Accelero S1 Rev. 2 with 2 120 mm fans attached via zip ties. They run at ~1000 rpm, nicely quiet. Chip temperature running MW is 60 - 70°C depending on ambient. The cooler itself is comparibly cheap, but quite large - mkae sure you've got the space if you want to mount it (or something similar). Also the mounting process is OK, but not for the faint of the heart (seeing your naked GPU die exposed..).

MrS

---

Scanning for our furry friends since Jan 2002

Nice case etc, good setup, so providing you've done the usual housekeeping re dust etc, which by the looks of CPU temperatures you do, there is no problems on that side of life.

GPU runs ~68c idle 85-97c load, sometimes passing 100c and crashing.

Provided thats a varied loading and not the artifical Furmark full blooded assault - and the fan its going at 100% - thats bad news, and indicates its history IMHO.

GPU VRM's (which are what I think is actually overheating) run at 80-90c idle, 100-130c load.

{My bolding}
If that level on the VRM has been "standard" when in use - its toast. Thats way over any kind of safe VRM temperature (the latter tends - varies a little with the card - to be 120degrees for VRMs, and normal idle is around 45-55 degrees. Sustained use over that only has one result, VRM burnout.

You can verify the latter by accurately measuring what is being drawn by the card. GPU-Z, whilst not exactly accurate to scientific standards, will give a good indication in the Sensor tab - look at the amp draw @12v - if as I suspect its high, then a VRM - Voltage Regulator - has burnt out. I would take it down to a local IT shop to get it properly verified (VRM burn out). If confirmed its then a case of what you want to do.

If its been Overclocked fiercely, XFX will not warrenty it because AMD will not cover burnt out cards after overclocking. I did manage to get a 5970 RMA'd with XFX, but it had not been overvolted, and VRMs were intact, so in the end I got a replacement. If the VRMs are burnt out, its vertually certain they will not replace under warrenty .... but lose nothing by trying :)

Regards
Zy