EK-5970 vs stock cooler tests

[Lord_Farken]

I take the water block out and re do the pad and grease
it help a lot ,beside only in xtreme burning mode reach that temps .never in crysis max out ,but is good to know the limit before s**thappen




now I can reach a new level thank to better cpu and gpu cooling
the part or the card that get really hot is the back plate ,but only during the furmark test

[Lord_Farken]

Yeah your going to have to put some thicker pads on them man.

Here my 5970 on air
GPU 1 and 2 5970
GPU 3 285GTX

Max temp on core with extreme burning mode was 80.5 C
Max temp on Vram was 99C extreme burning mode can do upto 120 before it effects anything.


Heres the screen shots.



link to higher quality screen

that all can do for the block side ,the one that is hot is the back side
what I gona do put a another cooler behind?
no way I keep it like that, for games they stay at 36c max that fine in my list

[jaredpace]

Farken, nice work on that Crysis benchmark. 89 FPS WOAH! Is that 1 5970, or two? heheh.

Also, now that you have reseated the waterblock and doubled up on the padding & paste, whats the new VRM maximum temps after gaming for a few minutes in Crysis? I don't see it in your screenshot.

[Lord_Farken]

Farken, nice work on that Crysis benchmark. 89 FPS WOAH! Is that 1 5970, or two? heheh.

Also, now that you have reseated the waterblock and doubled up on the padding & paste, whats the new VRM maximum temps after gaming for a few minutes in Crysis? I don't see it in your screenshot.

the max temps are in 2nd pic on above post just zoom with the browser
is only one but my set up beat some user scores that got tri-fire
they are really pissing me of by saying that is bull and cheated when I use the same tool
now at last I can play crysis and warhead all the way without slowdowns
at everything max ,it like playing a new version ,everything behave so different
thanks for the comment

[jaredpace]

Well if you notice, in that gpu-z sensor tab window card reads idle clocks of 157/300, 0% gpu usage. VRMS of 32C in crysis & 105C in furmark doesn't jive. I would expect them to be around 75-85C

[Lord_Farken]

Well if you notice, in that gpu-z sensor tab window card reads idle clocks of 157/300, 0% gpu usage. VRMS of 32C in crysis & 105C in furmark doesn't jive. I would expect them to be around 75-85C

yes i see that I don't know what up with that 0% ,while the test in running the osd show full load on both gpu's and same mhz
the temp I just going to leave it alone ,the performances is solid ,so if not broken don't fix it

[controlyar]

The 5970 reference design uses the *best* available components for both board and heatsink construction - I mean that quite literally.

It is critical that if the cooling solution is replaced, VRM cooling is seriously taken into consideration when selecting a replacement cooling solution. The Volterra regulator slaves (small shiny silicon components) must be kept cool for proper operation (125C rated, 150C shutdown) and to prevent the board from throttling clocks (typically at 125C).

The only recommended thermal interface materials (TIM) are:

Volterra Slaves (small shiny components) - T-Pli 225/230 - 230 is slightly thick and is easier to handle , best TIM available @ 6W/m-k
GDDR5 Memory/Inductors - black components (some marked CPL/Cooper/Pulse) - thin conformal pad, high thermal conductivity - few available

If the inductors are not thermal coupled to the waterblock, the waterblock will encase the part and allow it cook in this environment.

If you are "upgrading" to a liquid cooling block, you may want to carefully consider transferring the thermal material to the waterblock. It can be removed and kept intact with a fine blade.

To validate the VRM cooling is working , you can look at the graph/logs of GPU-Z for the VDDC Phase #1/2/3 temperatures running your favourite application - the readings are very accurate and directly reflect the regulator silicon junction temperatures.

I have seen single piece waterblocks struggle to provide adequate pressure to all components on such large dual GPU boards - there are many components to cool and it is very challenging to perform this function using a single piece of CNC'd copper. The tolerances are extremely tight, so you may have to overtighten screws to help pull the copper plate against the board - but then you risk damaging board components.


Otherwise, contact the manufacturer (DD/EK/Coolance etc.) and request that they consider using the same thermal interface materials as used on the reference design in the future.

Feel free to post this info on other forums. Much of it applies to other boards as well.

As an aside, I would not recommend running Furmark - it will only thermally stress your hardware (potentially shortening its useful life) and provide little indication on stable overclock speeds, which is the main reason users generally run it. The board may also throttle speeds, which again will not provide any insight into the maximum achievable speeds.

[Lord_Farken]

The 5970 reference design uses the *best* available components for both board and heatsink construction - I mean that quite literally.

It is critical that if the cooling solution is replaced, VRM cooling is seriously taken into consideration when selecting a replacement cooling solution. The Volterra regulator slaves (small shiny silicon components) must be kept cool for proper operation (125C rated, 150C shutdown) and to prevent the board from throttling clocks (typically at 125C).

The only recommended thermal interface materials (TIM) are:

Volterra Slaves (small shiny components) - T-Pli 225/230 - 230 is slightly thick and is easier to handle , best TIM available @ 6W/m-k
GDDR5 Memory/Inductors - black components (some marked CPL/Cooper/Pulse) - thin conformal pad, high thermal conductivity - few available

If the inductors are not thermal coupled to the waterblock, the waterblock will encase the part and allow it cook in this environment.

If you are "upgrading" to a liquid cooling block, you may want to carefully consider transferring the thermal material to the waterblock. It can be removed and kept intact with a fine blade.

To validate the VRM cooling is working , you can look at the graph/logs of GPU-Z for the VDDC Phase #1/2/3 temperatures running your favourite application - the readings are very accurate and directly reflect the regulator silicon junction temperatures.

I have seen single piece waterblocks struggle to provide adequate pressure to all components on such large dual GPU boards - there are many components to cool and it is very challenging to perform this function using a single piece of CNC'd copper. The tolerances are extremely tight, so you may have to overtighten screws to help pull the copper plate against the board - but then you risk damaging board components.


Otherwise, contact the manufacturer (DD/EK/Coolance etc.) and request that they consider using the same thermal interface materials as used on the reference design in the future.

Feel free to post this info on other forums. Much of it applies to other boards as well.

As an aside, I would not recommend running Furmark - it will only thermally stress your hardware (potentially shortening its useful life) and provide little indication on stable overclock speeds, which is the main reason users generally run it. The board may also throttle speeds, which again will not provide any insight into the maximum achievable speeds.

well men you really do your home work
yes I am sure now that the card throttle down under stress
only 3 times on air I get 27K score
every time I try again at the same settings that got me there
on water ,when the test start it go to 110 fps, then 2sec later boom it throttled back
if not a big issue really because game performances seem unaffected
thanks I'm honor that you register on 2006 and waited 4 year for a post worthy of your knowledge LOL

[Lord_Farken]

some guy ask to check and post gpu 2 clocks and
what the hell this means ?

[controlyar]

No homework here - I designed the board

Long-time lurker.

[lowdog]

No homework here - I designed the board

Long-time lurker.

Realy


So what do you suggest then, if we use for example the EK block to watercool the card then we should use all the tim pads from the stock cooler with the waterblock rather than the grey rubbery pads provided with the waterblock


Also, why does this card (all 5XXX series for that matter) seem to exhibit the freeze/grey screen/stipes issues so often......driver or hardware fault???

[controlyar]

I am not aware of the specific TIM the LC vendors use on the blocks.

The thermal conductivity of the gap pads on the reference design heatsink are of the highest conductivity - however, the block and reference heatsink are likely not exact matches in terms of height, so the gap pad requirements may not be identical (ie. one requires a slightly thicker pad to make contact). The material we select is actually simulated to ensure the pressure exerted will not damage the part once the heatsink is tightened.

Ideally, the block vendor will use comparable TIM's. If not, you can transfer the TIM over and then use software to validate the cooling performance. Again, the cooling of the Volterra slaves is the most critical so those particular locations should be using the T-Pli TIM.

On the grey screen issue, it is being investigated with considerable effort I assure you.

[lowdog]

I am not aware of the specific TIM the LC vendors use on the blocks.

The thermal conductivity of the gap pads on the reference design heatsink are of the highest conductivity - however, the block and reference heatsink are likely not exact matches in terms of height, so the gap pad requirements may not be identical (ie. one requires a slightly thicker pad to make contact). The material we select is actually simulated to ensure the pressure exerted will not damage the part once the heatsink is tightened.

Ideally, the block vendor will use comparable TIM's. If not, you can transfer the TIM over and then use software to validate the cooling performance. Again, the cooling of the Volterra slaves is the most critical so those particular locations should be using the T-Pli TIM.

On the grey screen issue, it is being investigated with considerable effort I assure you.

Okay, thankyou very much.

[nusihb]

Wow, we get to 'talk' to the board designer directly - cool!

The 'pad thickness' problem makes sense, when I was using the Koolance block the VRM temps flew up towards 150C during O/C (and using furmark) no matter what I did...
I couldn't get the water block tightened up enough, and not much of a imprint in the (fairly thick) pads.

But when using the DD block, no problem - the highest VRM temp I seen so far is around 55-60C, and that's with core at 990mhz, voltage at 1.25, running furmark for 20 minutes or so.
Measuring the temps with a IR meter on the back of the board seems to verify these temps.
DD is using thinner pads, so I assume the tolerances are tighter...

Controlyar - do you happen to have any insight in the '2D lockup' problems some of the 5970 owners here are experiencing?
I have a buddy at work with a 5970 that get this problem at (cold) bootup - and he's using the OEM heatsink, no O/C.
Seems like disabling ULPS and rising 2D clock speeds helps for some...

[controlyar]

On the low temp issue, this is not something we have seen internally. The product is qualified down to 0C ambient and the board is allowed to soak to allow it to approach this temperature after which a large matrix of tests are run.

This may be related to another issue under debug, so I will try to follow up after.

[Lord_Farken]

On the low temp issue, this is not something we have seen internally. The product is qualified down to 0C ambient and the board is allowed to soak to allow it to approach this temperature after which a large matrix of tests are run.

This may be related to another issue under debug, so I will try to follow up after.

ohh don't worry about it , find out yesterday that a don,t have that issue that lowdog got a all ,my problem was a bug in the OSD of After Burner
if the osd is monitoring the test it affect the fps drastically ,why I don,t know maybe you can get that fix with the next drivers set
with the osd on I get 25k now that I know that I get this


that quite a difference why is that ,this make no sence
but a happy now is beast still

[affiliate13]

On the low temp issue, this is not something we have seen internally. The product is qualified down to 0C ambient and the board is allowed to soak to allow it to approach this temperature after which a large matrix of tests are run.

This may be related to another issue under debug, so I will try to follow up after.

Hi controlyar,
Thanks for coming on here to talk to us, that is really good of you.

If I might hijack Lord Farken's thread for a quick question,

I am able to run my 5970 @ 1Ghz using 1.235v (EK install results here) Where would you consider the 24/7 voltage limit for these cards to be?

My VRMs temps as reported by GPUz are always well under 50c (usually mid 30s) only using Furmark do these hit 100c.
GPUs barely touch 40c during prolonged gaming even with a higher room ambient, like 25-30c.

1000mhz is plenty for day to day use but I think it can go higher, only i don't want to degrade this card so early on.

Thanks.

[controlyar]

Reliability lifetime is kind of black magic but I won't get into that or the specific details for products. It is more meaningful to discuss it in relative terms. Engineers are generally on the conservative side as well.

Generally speaking, the lifetime degradation with increased voltage is much more pronounced than increased temperature. With LC installed, the ~50C temperature reduction will allow you boost the voltage slightly with a negligible net reliability impact.

With DPM enabled, you have to consider your effective activity factor - ie. how many hours per day will you be playing games and running at the higher voltage? . If low, then again you can consider a small voltage boost.

Practically, VDDC = 1.1875V setpoint should not degrade the product with extended gaming (5870 uses this setting). Anything beyond that and you are impacting lifetime, but consider the factors above.

The hardware is effectively "obsolete" within 1 year and is generally not worth using after 3 years, especially in the enthusiast space - the boards burn too much power for the performance they can deliver relative to the latest products.

So achieving high reliability (ie. many years) and enthusiast gaming experiences are almost in direct conflict.

[affiliate13]

Yes, your last points bring it all into focus.
I have a pile of 4850, 4870 and two 4870x2s sitting here now redundant unless i can think of a use for them.
Around a year is what i will expect to use it for, and it will only be using the VDDC of 1.1235v for 2-3hrs most evenings.
Any increase will be for short benching runs

Thanks for such a detailed reply.

[t_ski]

I just received my FC5970 from Performance PCs yesterday. Thought I am not new to watercooling by any means, I still followed the instructions carefully. After I got the loop filled and bled, I fired up the system and booted into Windows (x64 Ultimate). After about 5-10 minutes, I started noticing flickering on the screen, then got the gray vertical stripes (never had that before with this card). Since then I have had several lockups, even with uninstalling the drivers and installing the latest hotfix. Even locks up in the BIOS. This only happened after I changed the stock cooler out for the waterblock.

Anyone have any suggestions? I'm going to have to take the block off and see if something's not making contact right. Damn tight loop though

[Lord_Farken]

I just received my FC5970 from Performance PCs yesterday. Thought I am not new to watercooling by any means, I still followed the instructions carefully. After I got the loop filled and bled, I fired up the system and booted into Windows (x64 Ultimate). After about 5-10 minutes, I started noticing flickering on the screen, then got the gray vertical stripes (never had that before with this card). Since then I have had several lockups, even with uninstalling the drivers and installing the latest hotfix. Even locks up in the BIOS. This only happened after I changed the stock cooler out for the waterblock.

Anyone have any suggestions? I'm going to have to take the block off and see if something's not making contact right. Damn tight loop though

mine never give that problem ,however my vrm temps were over 130c at max oc so even when the pad supplied by ek should be enough
I buy an extra pad kit and paste remove it and repad everything double and even in components that no supposed to bet doing contact with a bit of paste and now nothing go over 30c
so check what is the max temps with gpu-z in all the sensors
the is more threads about this issues just search for lowdog posts
he got a similar problem
I hope you get fix ,is such a big improve over running on air
just today i get over 30,160 gpu points in vantage

[t_ski]

Thanks - I'll see if I can find his posts.

[Philwong]

alright I run furmark xtreme burning test for 10min set gpu-z to high reading
here the pic,the vddc are to hot I guess they are not making good contact with cooler ,which one in the pbc are dose ?
the gpu's temp drop 10c with the new cooling ,how they can be such a huge temp differences ,40c gpu vddc 126c damn

Before you remount your block, disable voltage monitoring in AfterBurner and test the temps again. My 5870's VDDC can skyrocket to 1.65V which causes 100+ degree VRMs when I've more than one app (Everest Ultimate + MSI AB + GPU-Z) monitoring the GPU's voltages.

Phil

[t_ski]

I think lowdog and I may have the same problem:

http://www.xtremesystems.org/forums/...2&postcount=14

I turned off my fans and turned my D5 vario pump down to 1 and the systems is stable again. Need to do some more reading on this...

[Lord_Farken]

Before you remount your block, disable voltage monitoring in AfterBurner and test the temps again. My 5870's VDDC can skyrocket to 1.65V which causes 100+ degree VRMs when I've more than one app (Everest Ultimate + MSI AB + GPU-Z) monitoring the GPU's voltages.

Phil

yes you are right, that was before notice the bug
yeah that why can use OSD in AB no more ,it got a bug the even affect the fps in benchmarks
it my card woul realy get to over 130 it would be toast