5970 Waterchilling REV 2 Final

[Scaniris]

How is it possible to have differences as normally (as far as I imagine...) all OEM's are using the reference design at the product launch. Is it because OEM are responsible for some components integration? Or is it because ATI works with different providers to build the cards?

[Dude_Person]

weird, wonder why 0x MSAA is making vrms goto 120c and only 75c with 4x MSAA @ 1.3v.

I suspect it's memory bandwidth bottleneck. Furshading and multisampling is very bandwidth intensive.

[X2beach]

I can tell you the Xbrackets on the back are not all the same. And VRM hightS are little different..

[Romir]

Romir.. Be carefull running furmark and 1.3 Vcore on you 5970.... The VRM Get Hot very hot... Doesnt matter how well u cool them.. I had my Water Temps down to 15C the block was COLD.. adn the VRM still get very close to 120C Just so much heat coming out of such a small spot..

I Currently have (3) 5970 in my hands Sapphire, XFX, HIS and they are all little different.. componets different heights ect ect

I'm definitely looking for the lowest voltage I can run 850/1200 at and settling there. I got what I wanted, a quieter water cooled 5970 @ 5870 speeds for less than the cost of two of those.

Interesting note on the component differences, thanks. I didn't expect that from such early batches.

weird, wonder why 0x MSAA is making vrms goto 120c and only 75c with 4x MSAA @ 1.3v. Couple questions:

What happens to temps @ 1.3v 4x MSAA if it runs longer than 100 seconds?

What temps do you get at stock voltage 0x MSAA?

And finally, are those 120-140c temperatures real (verified with your Temp IR gun on back of card or block)? Just making sure it isn't a sensor glitching.

1. Here are the temps after 35m. (4x MSAA Furmark @ 1.3v 1000/1225). The GPU load never comes close to being maxed out with MSAA.



2. GPU-Z stopped showing vrm temps on the second card again so here's 30m of fullscreen furmark with the stock gpu1 results:



3. At 1.2v 900 core I measured 90c on the acetel above gpu2s vrms. Gpu1's vrms were 98c avg at the time so figure at least 10c higher for gpu2s. The little uncovered silver bumps/components to the right of the second gpus VRM (right edge of the block) also get quite hot. The second group down were hitting 80c with the others mostly in the 60s.

So I'm not the only one with a space heater then

Romir, could you try the following settings:
Core: 1.25v/900mhz
Mem - no overvolting - 1200mhz

..and see what VRM temps you get?

Overvolting my memory doesn't seem to help me, or is worth so little it wouldn't be worth the hassle. My card can't handle go beyond 1230 or so without errors.

I tested 20m at 1.2v 900/1200, sorry I had the wrong voltage. I'm sticking with 1.14v or whatever my card can run at 850/1200 from here out.

[Dude_Person]

GPU-Z stopped showing vrm temps on the second card again

It often does this with my card when I open two GPU-Z instances. You could just flip one back and forth? You loose the chart then though...

[nusihb]

It was at 1.25v 900core/1200 mem I got close to 150C on the VRMs, so you definitely are doing better than me.
I'll try a DD block next week, it'll be interesting to see if my temps will improve.

[jaredpace]

Thanks for the that testing and the results Romir. I see now that 4xMSAA Furmark is only using ~60% of the gpu, that's why vrm temps are lower. So at 1.2v, 100% gpu usage, after 20 mins of furmark, vrms are @ 100c. AFAIK, they say keep them below 105c and worry around 120c, while the gpu has a much lower threshold - keep them below 90c, and worry around 105c

I still think they need to make waterblocks concentrated on cooling these vrm areas. The gpu cores aren't even reaching 50C completely maxed out, while the vrms can still go to the 120c range on water. And that is verified on like 3 or 4 different blocks by different people.

[nusihb]

I swapped the Koolance block to a DD block, and now my VRM temps are better... upper 70s when overvolting, where I had up towards 150C with the Koolance block.
When I asked the Koolance tech support if they're seeing anything similar in their testing, they told me 'we don't have a testbed here for checking temps'... hm.