5850 - VDDC #2 hitting 110+ C

[Devastator]

I currently have two Sapphire 5850s in Crossfire. I just began to overclock them, and got a stable core of 1023MHz @ 1.287v. This overclock is FurMark stable.

Now the strange thing is, either I can get a stable core of 1023MHz while the memory is at stock, or a stable memory of 1230MHz while the core is at stock - but I can't get both at the same time at all. When the core is overclocked, I can't overclock the memory to crap (1023/1150 results in instability).

Recently, I discovered that the reason the memory isn't overclocking while the core is overclocking is because of the extreme temperatures I'm getting on the VDDC. VDDC #2 hits 110+ C while running FurMark while VDDC #1 and #3 are around 95-100 C. The overall temperatures are about 10 C lower on my second 5850 (the one farthest from the CPU).

Now my first question is a technical one, and what exactly is overheating from the photo below:



This photo shows the 5850 with the Twin Turbo Pro memory/voltage heatsinks on. My question is, are the three chips shown in the red rectangle the ones that are overheating? These chips are only cooled by air (there are no heatsinks or thermal pads on them even with the stock cooler). Would it help my VDDC #1, #2, #3 temperatures if I put heatsinks on them while installing my Arctic Cooling Twin Turbo Pro?

This is the photo of the 5850 with the heatsinks off if anybody needs to see under the heatsinks:



Thanks for your help!

[w4tch0]

It's the mosfets under the heatsinks right next to your "red rectangle" that are so hot.

I've seen some tests of twin turbo pro on a some german site and I know the mosfet temps incresed a lot compared to using the stock heasing. Fact is, those passives from twin turbo package are waaaay too small comapred to the massive cooling ability of the whole fullcover from the stock heatsink.

Get bigger passives on those chips.

And to directly answer your question: no you don't need to add passive cooling on the parts you marked. And it would not help at all.

//edit: here is the test http://www.hardwareoverclock.com/Arc...urbo_Pro-4.htm
It's in german but the graphs are clear. The power phases temps (VDDC) incresed a lot compared to stock.
Setting the Twin turbo manually to high RPM seems to help as well.

[Devastator]

I'm actually still using the stock cooler, those photos I linked in my post above were from the same German review you're talking about. I was planning to buy the Twin Turbo Pro in hopes that it would solve these high VDDC temperatures I'm getting, but I guess that isn't the case.

What should I do to decrease the VDDC temperatures (I'm using the stock cooler at the moment)? I don't think airflow is the issue - my system is currently open (not in a chassis), and a 230mm fan is blowing in that opening next to where you plug in those two six-pin connectors for both my video cards.

[w4tch0]

Oh in that case it's a different story.

I don't have experience in Overclocking 5850 as high, mine only goes up to 950core with maxed out Vcore... anyway either the mosfets simply can't put up with the power requiremnet and get so hot even with proper cooling or maybe the stock heatsink is from factory installed incorrectly (missing theremal pads maybe? dunno).

[Zeus]

Take the stock cooler of and see if something is amiss. I think the stockcooler has a hole there so that the turbine is blowing some air onto that area.

It's not unusual to have a trade off between core and mem overclock by the way. I think it's due to Vreg limitation on the card. (traces/resistance, noise, overheating mosfets etc.)

[Devastator]

I know for a fact it's due to the overheating of these VDDC things. When I lowered my core clock to 1016MHz @ 1.237v, the temperature of VDDC #2 dropped to 105 C under load (this was when the memory was at stock). I changed the memory to 1200MHz, and it's FurMark stable until the heat builds up and the VDDC #2 hits around 120 C (yes, the temperature increased when the memory clock increased) - then it crashes again.

My explanation might not be spot on, but it definitely has something to do with VDDC temperatures (I'm thinking it's mostly VDDC #2).

I'll take the stock cooler off tomorrow and see what's going on in there.

[jaredpace]

Make sure the VRM slaves have contact with that thermal pad strip, you may want to double the pad thickness there if there is no impression on it. Make sure the screws above & below are tight

[Devastator]

Make sure the VRM slaves have contact with that thermal pad strip, you may want to double the pad thickness there if there is no impression on it. Make sure the screws above & below are tight

How exactly would I do this?

[jaredpace]

5850 has 3 + 2 phase, so to reach those high clocks (1050/1200) you have to push the same power through less power management regulators than a 5870 (4+2). So they get hotter cause less units have to do more work. There are 3 in a row there, plus one missing on the bottom (on 5850), and two for VDCCI/VTT on the top. (There is also a 6th Volterra slave to the left of the top row of memory chips).

BTW, do you get 115C temps in games, or just in Furmark?

[jaredpace]

get something like this:

http://cgi.ebay.com/Thermal-Conducti...1#ht_742wt_939

[Devastator]

Ah, I see.

An answer to your previous question - yes, these high temperatures only occur during FurMark stressing. They are nowhere as high during gaming (checked the temperatures during gaming right now by running the Unigine Heaven Benchmark).

[jaredpace]

If they are probably around 70-80c while in Crysis or Heaven bench, then I wouldn't worry about 115C in Furmark burn. ATI calls Furmark a "power virus" anyway, and the only thing that will happen is those vrm slaves will reach 125-140C and start to throttle power and downclock your gpu.

[w4tch0]

Yeah 5800 has a built in protection of the VRM's that was implemented mostly because of software like furmark and OCCT.

It's not possible to achieve such amount of stress as Furmark puts on the GPU VRM's in any real world application anyway.

[RoadconeTuning]

i just let furmark run with my stock cooling and clocked at 922/1200 and VDDC 2 just hit 128C.... holy sh1t... core was 90C and this was within 30 seconds... fan had ramped up to 2500+ rpm via stock fan profiles... im backing my card back down untill my EK waterblock gets here




well i thought it was too much voltage, i still had it jacked up to 1.35 for when i was shooting for 1GHz and i dropped a good bit of amperage but after a minute or so the temps shot up to freak 135 on the VDDC! down to 1.28v and it was stable but temps are waaayyyyy too high for me


after 3 minutes at stock everything...


i guess the stock cooler is just no where near good enough to keep up with this cards voltage circuits.

[RoadconeTuning]

Btw, what voltage is the vddc amperage rated at? Because 90 amps should be causing a freaking arc... I'm pretty sure that the gpu-z utility is a decimal point off...

[[XC] gomeler]

Btw, what voltage is the vddc amperage rated at? Because 90 amps should be causing a freaking arc... I'm pretty sure that the gpu-z utility is a decimal point off...

93 amps at 1.3vdc or whatever he had set. That's nothing compared to some larger GPUs(GT200).

[RoadconeTuning]

now im trying to figure out how that is even posible with such small traces lol.. my dad is an electrical engineer and were gonna model it in ETAP tomorrow at work... it just doesnt seem like physics should work right.. we guessed you'd need a #6 wire to do that lol

[[XC] gomeler]

now im trying to figure out how that is even posible with such small traces lol.. my dad is an electrical engineer and were gonna model it in ETAP tomorrow at work... it just doesnt seem like physics should work right.. we guessed you'd need a #6 wire to do that lol

The traces are not small. The vGPU power plane spans several centimeters in width in order to handle all the current.

[jaredpace]

its 1.3v or whatever, not 12v like the rest of the system. so it's 93A / (12/1.3) = actual amperage. So like approx ~ 10 amps, or approx ~ 120 watts pull @ 12 volt system power

[RoadconeTuning]

i realized that after i made the post... i knew there was no way in hell it was 90amps at 12v... that'd be a helluva wire for 12v dc

[Philwong]

Can you guys log the GPU-Z's readings into a file and check what is the highest VDDC value? I hit 1.6V which is ridiculous when I didn't even tweak my 5870's vCore setting. Weird! Not sure whether MSI AfterBurner is going haywire here. I'm using this app only to customize the fan speed profile.

Phil

[RoadconeTuning]

supposedly if you monitor with two programs you end up with a massive freakout from the chip and they'll shoot up to 1.65v so you ahve to disable voltage monitoring in afterburner if you are using gpu-z as well... looks like we might have found a bit of the problem.

[joshd]

Old-ish thread, but I'll post here instead of making a new one.

My card idles at 55C (dual monitors) and loads under furmark at 90C with 110C VDDCs. Should I really care about this? It's more like 75C load and ~70C VDDCs under Heaven bench, which is fine. I've not really tried clocking the card yet due to the temps it reaches on the stock cooler under furmark.

[Kain665]

Let me get this straight - you're getting under 110c vddc temps with the stock TTP heatsinks and 1.287v? I get easily up to 120c with only 1.225v... and I'm using a modified stock base plate with the provided heatsinks stuck onto the VRM heatpipe...

[Necetra]

One item. Thermalright VRM-R4.

Reduced my VRM load temps to less than my GPU core load temps.
Nothing can touch it. Probably overkill, but the cooling performance is epic.