VRM cooling appears to be an important issue for the 4850

[dug777]

You may recall that I had suspected my Asus 4850's VRMs were causing my crashing issues at both stock and OC'd settings, as it uses the non-stock Asus Glaciator cooler, which provides no VRM cooling:

Card:

http://pics.bbzzdd.com/users/dug777/4850above.JPG

Thread with backstory:

http://forums.anandtech.com/me...240451&highlight_key=y

Anyway today I applied two little Zalman RAMsinks to the VRM groups (one per four little chips) and then strapped a whisper quiet Spire 80mm fan over the card using some fishing line through the many spare holes at that end of the PCB:

http://pics.bbzzdd.com/users/dug777/DSC_00900001.JPG

I just tied it firmly on at three corners of the fan.

While I had the case open I also turned my 120mm exhaust fan back to medium speed from high.

I then fired up furmark, and I was so confident of my mad skillz that I went straight in at 790/1028

Rock solid, despite the hottest GPU core temps I've ever seen on it; after 10 mins the core temp reading in furmark stability testing levelled out pegged at a toasty 101C...over 200F for you crazy yanks :Q

Looking back at my previous testing in the original thread you'll see that GPU temp clearly isn't the problem, and now with my VRMs properly and quietly cooled it's clearly showing this as it functions perfectly at far higher temperatures than I could previously attain without crashing.

Before I was crashing at stock 625/993 at 87'C GPU temp, now I'm solid and stable at 101'C GPU and 790/1028.

Three conclusions can be drawn:

1: The Asus Glaciator hsf isn't very effective at maintaining a low GPU temp.
2: The core can remain stable up to very high temperatures.
3: Proper VRM cooling is important for stability.

Comments/Thoughts/Questions?

 

[Avalon]

You're making me want to do that to my 4850 and OC it

 

[Zstream]

I purchased a couple S1's and used the stock VRM cooler. Works great.

 

[vj8usa]

What voltage are you using to hit that OC?

 

[nRollo]

Originally posted by: dug777


You may recall that I had suspected my Asus 4850's VRMs were causing my crashing issues at both stock and OC'd settings, as it uses the non-stock Asus Glaciator cooler, which provides no VRM cooling:

Card:

http://pics.bbzzdd.com/users/dug777/4850above.JPG

Thread with backstory:

http://forums.anandtech.com/me...240451&highlight_key=y

Anyway today I applied two little Zalman RAMsinks to the VRM groups (one per four little chips) and then strapped a whisper quiet Spire 80mm fan over the card using some fishing line through the many spare holes at that end of the PCB:

http://pics.bbzzdd.com/users/dug777/DSC_00900001.JPG

I just tied it firmly on at three corners of the fan.

While I had the case open I also turned my 120mm exhaust fan back to medium speed from high.

I then fired up furmark, and I was so confident of my mad skillz that I went straight in at 790/1028

Rock solid, despite the hottest GPU core temps I've ever seen on it; after 10 mins the core temp reading in furmark stability testing levelled out pegged at a toasty 101C...over 200F for you crazy yanks :Q

Looking back at my previous testing in the original thread you'll see that GPU temp clearly isn't the problem, and now with my VRMs properly and quietly cooled it's clearly showing this as it functions perfectly at far higher temperatures than I could previously attain without crashing.

Before I was crashing at stock 625/993 at 87'C GPU temp, now I'm solid and stable at 101'C GPU and 790/1028.

Three conclusions can be drawn:

1: The Asus Glaciator hsf isn't very effective at maintaining a low GPU temp.
2: The core can remain stable up to very high temperatures.
3: Proper VRM cooling is vital for stability.

Comments/Thoughts/Questions?

Ssshhhhhhh!
Mum's the word!

I'll likely be selling a VRM cooling-less 4850 soon, and if you clue people in to the need for this, my ill gotten Flea-Bay gains will be scuttled!

 

[dug777]

Originally posted by: vj8usa
What voltage are you using to hit that OC?

Rocking with stock voltages

It's just a freak.

@rollo:

I wouldn't worry about it, since partners sell them without VRM cooling.

Would be interested to see how most of those naked-VRM cards handle flat-out furmark, since mine was crashing at stock with mildly warm ambients in a case with decent if not exessive cooling...

 

[dug777]

Originally posted by: Avalon
You're making me want to do that to my 4850 and OC it

You do get a nice and noticeable boost

 

[aldamon]

VRM sinks are not vital to my PowerColor 4850's stability and I've had mine overclocked and pencil modded since this summer. I take the word of ASUS's engineering, or any other producer of a card with no VRM sinks, over any observations on a single card. You should have RMAd it.

Also, the temps produced by Furmark are nuts. No game on the planet comes close to doing that to a card. My PSU even squeals like a pig while running that POS program. Stop running "stability" programs and play some games. If Crysis doesn't melt the card, nothing else will.

 

[nerp]

I'm another person who doesn't have anything on my vram on my 4850 and I haven't had a single issue because of it. I run at stock. The zalman fan on it seems to blow enough air over the chips to keep them cool enough.

This is the card I have:

http://www.newegg.com/Product/...x?Item=N82E16814125238

 

[dug777]

Originally posted by: aldamon
VRM sinks are not vital to my PowerColor 4850's stability and I've had mine overclocked and pencil modded since this summer. I take the word of ASUS's engineering, or any other producer of a card with no VRM sinks, over any observations on a single card. You should have RMAd it.

Also, the temps produced by Furmark are nuts. No game on the planet comes close to doing that to a card. My PSU even squeals like a pig while running that POS program. Stop running "stability" programs and play some games. If Crysis doesn't melt the card, nothing else will.

I felt it would be foolish to return my card when it clocks like it does, I can equal your core without a voltage mod after all

I've also solved my problem very simply, but I suppose I can agree with you, clearly VRM temperatures ARE vital to my 4850's stability, but I can't neccessarily extrapolate that to any other 4850s...

Easy way to tell of course, leave the furmark stability test running at say 1280x1024 2xAA in a room that's about 25'C ambient and tell us how you go. I suggest earmuffs for the squealing, glad my machine doesn't do that!

If you had read the backstory I provided before getting all excited about me stopping running stability programs and instead playing games (), you'd note that I was getting crashes in FarCry as the weather warmed up, which delivers a power-draw that is very close to what furmark pulls (showing a consistently greater power-draw than Crysis and COD4).

I would guess Crysis and COD4 don't do it because they are hitting another limitation before they max out the core, maybe the mem-bandwidth?

Anyway, I only ran furmark to see if it was indeed a GPU-load related crash and to watch my realtime temperatures before it crashed. I'm glad I did, otherwise I would have thought it was a GPU-temperature related issue.