ATI 5970 overclocking problems?

[thilanliyan]

http://www.legionhardware.com/document.php?id=868&p=1

Not surprising really...a card that is pushing 300w at stock...of course it's gonna run hot and throttle if you really try to push it.

 

[Hauk]

Furmark..

 

[lavaheadache]

I'm betting that third partly cooling companies will have something out that will address the vrm cooling flaws associated with the stock cooler, if that's what you can even call it , (a flaw)

 

[dank69]

There is more info on AT homepage.

 

[bunnyfubbles]

cooling the vrms is what got me in trouble with my 5850

cooling the GPU is easy peasy, my Accelero S1 + 120mm fan idles at 26C and loads in the low 40s while game, apps like OCCT and Furmark can bring that load up to a 'whopping' ~49C

Its the VRMs that are a bitch. When installing the S1 I had to remove the base plate of the stock heatsink/fan, which acts as a heatsink for the RAM and VRMs. Simple stick-on heatsinks weren't adequately cooling the VRMs, quickly sending temps soaring to 120+ when using those "power virus" apps. After a lot of grief due to fault of my own, I was finally able to get the base plate back on (there was a moment where I thought I might have killed my 5850) and VRM temps are back under control. Granted, I've learned to avoid using Furmark and OCCT at ridiculous levels (higher resolutions and/or AA will really stress the VRMs) and I will just monitor my temps while playing games and running game benches.

 

[nitromullet]

Interesting that AT had no problems in actual games, while Legion Hardware mentions throttling in STALKER CS.

 

[KIAman]

In a perfect world, the VRMs would be attached via a separate daughter card which has its own independent cooling.

Or in this case, it is already taking up 2 pci slots so put both GPU on a single pcb and the VRMs on a separate pcb along the top.

I think I mentioned this somewhere already but mark my words. We will have GPU with it's own AC wall connector, lol.

 

[Painman]

Throttling during FurMark doesn't surprise me at at all, considering the job that a 5970 stock cooler has. Throttling while playing Stalker would indicate to me that they might have had a wonky review sample.

 

[Idontcare]

Looks like the nicely proved that AMD's throttling technology works as advertised.

 

[Daedalus685]

Did anyone actually ever think that a stock cooler was goign to be able to keep up with the levels of overclocking many expect from a 5970? Has any reference stock cooler been able to keep up with overclocking like that? ever?

More reason to wait for a nice waterblock/very creative air before anyone buys one of these with the expectation of running it like a 5870x2 I suppose.

Were these nubmers all done at 5870 voltage? I'd be interested in seeing more, as the AT blog aludes to, about what voltage is actualyl required to get 5870 clocks.

 

[uclaLabrat]

Did anyone actually ever think that a stock cooler was goign to be able to keep up with the levels of overclocking many expect from a 5970? Has any reference stock cooler been able to keep up with overclocking like that? ever?

More reason to wait for a nice waterblock/very creative air before anyone buys one of these with the expectation of running it like a 5870x2 I suppose.

Were these nubmers all done at 5870 voltage? I'd be interested in seeing more, as the AT blog aludes to, about what voltage is actualyl required to get 5870 clocks.

According to the AT review, that's precisely what was advertised. The card has a stock TDP of 300W per the ATX spec, but the cooler has 400W thermal draw. They beefed up the cooler so it COULD be OC'd, to reach its full potential

 

[Daedalus685]

According to the AT review, that's precisely what was advertised. The card has a stock TDP of 300W per the ATX spec, but the cooler has 400W thermal draw. They beefed up the cooler so it COULD be OC'd, to reach its full potential

Mentioned in marketing slides and advertised are different things.

The cooler could very well be capable of handling 400W of heat. This makes no stipulation on where these 400W have to come from. Obviosuly the thermal conductivity from the VRM was not sufficient, but this in no way means that the cooler could not handle 400W.

The AT review mentioned that the card seemed designed for higher clocks, but evidently this is not the case.. at least as far as the design of the cooler goes.

 

[Painman]

There's an AT blog post about the issue... linky

I had forgotten that AT ran into the issue as well. But it does appear to be related to VRM cooling. Maybe it's an uneven contact issue, maybe too much localized heat dump into one part of the 'sink? A bit of both?

In the picture in that blog post, you can actually see that the VRMs at the top of that vertical row of 6 were getting more contact pressure than the ones lower down. The intentations in the waxy TIM tells the story. It's pretty well squashed at one end, and hardly at all on the other end.

 

[Daedalus685]

There's an AT blog post about the issue... linky

I had forgotten that AT ran into the issue as well. But it does appear to be related to VRM cooling. Maybe it's an uneven contact issue, maybe too much localized heat dump into one part of the 'sink? A bit of both?

In the picture in that blog post, you can actually see that the VRMs at the top of that vertical row of 6 were getting more contact pressure than the ones lower down. The intentations in the waxy TIM tells the story. It's pretty well squashed at one end, and hardly at all on the other end.

Aye, the contact seems a bit off, but the major issue is that there is actually no heat sink above them, just the fan and a metal bar. The things are not even under the vapor chamber.

This cooler is clearly not desinged, for whatever reason, to keep teh VRMs cool. Either they didnt think they would get that hot, or other limitations made them cut back (size comes to mind). In any event, the cooler is not sufficient for high overclocks, more due to its position than anything else.

Im scared to see models with the fan hanging off the back... 15" anyone?

 

[nitromullet]

Mentioned in marketing slides and advertised are different things.

The cooler could very well be capable of handling 400W of heat. This makes no stipulation on where these 400W have to come from. Obviosuly the thermal conductivity from the VRM was not sufficient, but this in no way means that the cooler could not handle 400W.

The AT review mentioned that the card seemed designed for higher clocks, but evidently this is not the case.. at least as far as the design of the cooler goes.

You're right... It sounds like ATI's legal team made sure they did the requisite CYA.

from http://anandtech.com/video/showdoc.aspx?i=3679&p=4

Make no mistake: this card was designed to be a single-card 5870CF solution; AMD just can’t sell it like that. In our discussions with them they nearly (as much as Legal would let them) promised that every card will be able to hit 850MHz core (after all, these chips are binned to be better than a 5870), and memory speeds were nearly as optimistic, although we were given the impression that AMD is a little more concerned about GDDR5 memory bus issues at 5870 speeds.

...that doesn't mean 5970 owners should be happy about this. Luckily, right now this is very few people. Supply issues for ATI should give potential customers the time (whether they want it or not) to evaluate the purchase of a 5970 vs dual 5870s.

 

[Painman]

Aye, the contact seems a bit off, but the major issue is that there is actually no heat sink above them, just the fan and a metal bar. The things are not even under the vapor chamber.

This cooler is clearly not desinged, for whatever reason, to keep teh VRMs cool. Either they didnt think they would get that hot, or other limitations made them cut back (size comes to mind). In any event, the cooler is not sufficient for high overclocks, more due to its position than anything else.

Im scared to see models with the fan hanging off the back... 15" anyone?

If the VRM cooling design is similar to the 5850/5870 scheme (I'd imagine it is) then yeah, there's a copper bar bonded to the heatsink on the other side of the base plate.

I just noticed that the pic in the blog post is a link to a much larger pic. In the large pic, you can see that the VDDC slave at U89 has a totally clean top... it's not making heatsink contact at all. The one at the very bottom of the row has good contact though. There's some warpage happening here for sure.

I don't think I'd recommend the 5970 right now unless AMD gets some better QA, or the purchaser plans on using 3rd party cooling.

 

[Daedalus685]

Certainly it is not something to be happy about. But yet another reason to be a pecimistic SOB when a company tells you that the stock cooling is epic.

I don't know.. If someone were to post in the CPU forum that their brand new i7 920 was balls because they could not get a stable OC with nice temps on the fan it came with I think I can see the responses.

Don't get me wrong, it seemed certainly clear to start that AMD wanted this thing to run at 5870 clocks, at least the board it self, but it seems as clear that when they chose to run it a bit slower they also probably shortened the heat sink to save on size and cash. I thught the first pictures of the thing about an inch longer, though I could be mistaken.

The stock cooler being insufficient to OC is not at all the same to me as the card itself falling short. It is certainly a bit strange to put such a good cooler on it yet forget to cool the VRMs.. and a bit misleading to advertise the OC potential yet cool it inadequately. However, this is not at all that shocking.

 

[nitromullet]

If the VRM cooling design is similar to the 5850/5870 scheme (I'd imagine it is) then yeah, there's a copper bar bonded to the heatsink on the other side of the base plate.

I just noticed that the pic in the blog post is a link to a much larger pic. In the large pic, you can see that the VDDC slave at U89 has a totally clean top... it's not making heatsink contact at all. The one at the very bottom of the row has good contact though. There's some warpage happening here for sure.

I don't think I'd recommend the 5970 right now unless AMD gets some better QA, or the purchaser plans on using 3rd party cooling.

They discuss this in the comments...

Bad contact? by Mr Perfect, 4 hours ago
It looks like one of the VRMs on the right didn't even touch the heatsink, it has no thermal goop on it and there is n oindentation on the TIM either. Maybe better contact would help? Is Anand's card similar or wose for contact? Could be a quality control thing.

Reply | Report Post

RE: Bad contact? by greywood, 3 hours ago
Second that - - from the photo, it looks like at least two VRM's at the bottom right side and all three at the top center are making little if any contact with the TIM. Might be interesting to clean off the generic goop, re-apply some AS5 (or such) then really "cinch-down" the HSF and try re-testing?

Reply | Report Post

RE: Bad contact? by Rajinder Gill, 3 hours ago
I think Ryan scraped that off to read off the FET part numbers. I asked him the FET model numbers because I wanted to find out if ATI had used 45amp slaves.

Reply | Report Post

 

[Daedalus685]

If the VRM cooling design is similar to the 5850/5870 scheme (I'd imagine it is) then yeah, there's a copper bar bonded to the heatsink on the other side of the base plate.

I just noticed that the pic in the blog post is a link to a much larger pic. In the large pic, you can see that the VDDC slave at U89 has a totally clean top... it's not making heatsink contact at all. The one at the very bottom of the row has good contact though. There's some warpage happening here for sure.

I don't think I'd recommend the 5970 right now unless AMD gets some better QA, or the purchaser plans on using 3rd party cooling.

Keep in mind that we don't know how many times the thing was opened and clsoed before they took the picture.

Also, the thing works jsut fine at the settings it ships with, so what are ya gonna do.

Here is a pic of the 5870 cooler:
http://pic.xfastest.com/sxs112/ASUS/5870WEB/pictures/ASUS5870-31.jpg

The bar on top of the VRMs is not actually attached to the heat sink.

It is the same on the 5970, the VRM pad is not actually attached to the vapor chamber at all.

The 5970 has one more VRMs under the bar than the 5870, it is no wonder they get hotter, and since there is no thermal conductivity to the heat sink (the part rated at 400W) we run into cooling problems VERY quickly.

 

[Hauk]

Keep in mind that we don't know how many times the thing was opened and clsoed before they took the picture.

Also, the thing works jsut fine at the settings it ships with, so what are ya gonna do.

Here is a pic of the 5870 cooler:
http://pic.xfastest.com/sxs112/ASUS/5870WEB/pictures/ASUS5870-31.jpg

The bar on top of the VRMs is not actually attached to the heat sink.

It is the same on the 5970, the VRM pad is not actually attached to the vapor chamber at all.

The 5970 has one more VRMs under the bar than the 5870, it is no wonder they get hotter, and since there is no thermal conductivity to the heat sink (the part rated at 400W) we run into cooling problems VERY quickly.

A solid observation. Thermal patches and a copper bar taking the VRM heat nowhere!

Revision time..