Water cooled 970's...

[senior520]

Hey guys, This might be a little long winded but wanted to try and include as much as possible , Ive developed a bit of a puzzling issue following a few changes ive made to my rig, and don't think it'll be one of the usual problems that can be solved with a bit of forum digging.......

With the launch of the new 10xx series cards id developed a serious itch to do something with my rig, but right now cant justify the cost of upgrading to a newer card as im running a pair of Asus 970 Strix's.
They do pretty much whatever i need at either solid 60fps or just below ( Ark being the exception ), main issue i was having was with noise, even with the quiet coolers on the strix cards prolonged gaming with good overclocks ended up with some fan noise. So i decided to go Water cooled for something to do and to drop noise levels.

Ordered a pair of kraken G10 AIO brackets and two Corsair h55 AIO coolers, and a pair of Noctua fans to replace the nasty white ones on the g10 brackets.....installed and started testing.,straight away temps are AMAZING, dropped from 75-80c on the top card after a long bench down to 56c MAX.

Now my previous best OC was around 1350mhz in Sli as id set a thermal limit in afterburner at 80 deg, so cards stabilized at around that after a prolonged bench.With the new coolers ive managed to get that up to 1501mhz, basically limited by card stability, start getting artifacts if i go much higher.

Now this is where i think my problem starts, Power limit has been increased to 120% in afterburner, and whilst benching the cards are showing around 118% maximum power usage and just under 1.2 volts, again temps still holding steady just below 60c after being thoroughly warmed , however after 4 or 5 loops of 3d mark stress testing without any real warning....Black screen, system restarts....... on boot i will have lost one of the two cards, doesn't even show up in Bios on PCI/DIMM POST. If i allow it to boot windows Nvidia control panel only Shows a single card also. However wait 10 minutes and reboot.....All is well, both cards showing in PCI / DIMM POST, and sli back working on win loading.

So this leads to the question, anyone seen or heard of anything like this? As its the first time ive been able to push these cards in Sli, always been limited by card temps in the past. Under powered? Overpowered and causing card to shut down? Im unsure...
Tried a few differnt combinations of lowered and raised power limit, with no real effect. Dropping the core clock to 1400mhz does eliminate the problem but surely i should be able to run a 1500mhz clock with temps so low and no visual artifacts?

Thanks in advance for any info or help.

 

[D1craig]

do you have heatsinks on the other vital chips of the card ie. the VRAM etc?

have you tested the cards seperatly for there max overclocks first? to me that sounds like the driver is crashing because the overclock fails.

 

[stooeh]

Not all 970s can do a stable 1500MHz. Only one of mine will.
I have seen blackscreen restarts when pushing too far in SLI whereas in single gpu mode it just tends to crash driver and recover. Hard restarts could also be related to power delivery I expect as well though, though I haven't seen a card 'go missing' after a restart.

 

[senior520]

No added heatsinks, although the DCII cooler didn't touch the VRM's either, only contact point was to the GPU with passive cooling from the fans and backplate. New setup has a dedicated fan for the VRM's.

I'll be honest id overlooked the thought of that as driver crashing had always presented to me with a system hang followed by recovery to desktop and a pop up system notification.

I'll give it a go see what i get clock wise individually.

 

[D1craig]

my evga sc 970 has a kind of metal plate that is (i think) in contact with some of the heatsink and the metal plate covers/touches some of the chips on the PCB.

one thing i noticed with my 970 was even with a custom cooler (prolimatech mk26 and heatsinks on the other chips/parts) the overclock was a lot better than stock cooler. then when i went full custom loop i could pushed the card even harder and the waterblock touches every chip on the board which must be the thing holding my card back.

 

[senior520]

Just ran a stress at 1450mhz in sli while waiting on a DL, took a little longer as the cards hadn't been pre-heated but still got the black screen and reset, this time the top card dropped out so had to swap HDMI to the bottom card to get display up, again no sign of the offending GPU in bios.....this time decided to try a different approach and fully disconnected AC power. Reconnected after 20 seconds or so, boot to bios and voila, card 1 back up and running.....odd
I'll come back to this one tomorrow with a fresh head and find out what each card can do on its own

 

[senior520]

Ok so following some extended testing ive done on these both individually and in sli, the cards seem to be suffering from heat problems of all things.... if i drop the power limit substantially i get no issues at all over prolonged bench or stress runs. As soon as i crank the limit back up black screen reboot.....and to be honest something smells warm, so i popped the side case off and had a feel around on the cards back plates, both are warm but not what id call hot on the front ( power connector end ) of the card and around the GPU, but down at the back, near the VRM controller and MOSFETs its bloody hot, the heatsink on the VRM controller is so hot i can barely hold my fingers on it......

EDIT: Just confirmed with HWinfo.... VRM temps are hitting 100c when i closed the benchmark, whilst Gpu temp was still down at 52c

Any suggestions for a good quality cooling solution for this area of the card? Got the side fan turned up a little to circulate more air over the heatsink but i think its going to need improving....heres a shot of the heatsink as it is now....