Powercolor HD7970 Core Voltage Issues

[Jetlaw]

I've disabled ULPS via the registry and I'm using MSI Afterburner to set new voltages and overclock my 7970. I'm also using dual monitors.

Using HWiNFO64 my voltage readings are consistently about 10mV below what I have in Afterburner, but that's not my problem. My problem is the damn card under load.

With no overclock at all my card runs stable at 1050 core/1400 memory and I have increased the voltage from stock 1175mV to~1180mV.

Now when I hit up anything that pushes the GPU to 100% (like Battlefield 3 for instance) the core voltage readout (GPU VRM Voltage Out (VOUT/VID) in HWiNFO64) drops. At points more core voltage will be down to 1120mV.

Now this is only another minor gripe of many. My card somehow retains stability despite the core voltage dropping by ~50mV under load.

The major gripe comes here - And this is what I want to fix. While livestreaming using XSplit (and only with XSplit, it's a documented issue between XSplit and 7xxx series cards) I'll get regular driver crashes within 15-60 minutes of starting streaming regardless of if I've just launched the game or it's been running for 3 hours and then I decide to start XSplit. Pre-crash while my monitors still display images before blacking out HWiNFO will report my lowest core voltage usually as in the 1120mV range. The second my drivers recover, the lowest core voltage is usually 900mV.

I'm taking a wild stab in the dark that my core voltage is dropping to 900mV which is causing the driver crash. Of course with the driver crash comes the restarting of whatever game I'm playing and the restarting of my live stream and this isn't very nice.

What reason does my card's core voltage decrease under load and is there any correlation between the 900mV minimum reading just after each driver crash and the driver crash happening at all?

 

[m1dg3t]

System specs? PSU? Tried 1 card? Actually measured with a DMM to confirm software reading?

Go! Go! Go! Go! Go! Lol

 

[razaron]

Does afterburner have a logging feature? If not, get GPUTweak and in the monitor options tick enable log. This should help you find out what the vGPU is during crashes. Also, vDroop scales with temperatures so your card must be getting toasty. Mine was >1.155V at 80C.

 

[Jetlaw]

Power supply is 850w. Using a Socket 2011 i7 3930k OC'd @ 4.4ghz, 24GB of Kingston HyperX @ 1600 CL-9 in 6x8GB setup. Mobo is an ASROCK x79 Extreme7.

Haven't used a multimeter to measure the actual voltages.

I didn't know about or what Vdroop was, so I did some research and it seems in line with the standard voltage reductions I'm seeing. Will use GPUTweak and log what happens prior to the driver crash where my voltages drop to 900mV.

 

[Jetlaw]

GPU Tweak doesn't seem to be capturing the VDroop at all, but here's the readout from the latest test:

http://pastebin.com/iQfHZC3w

Search for "*** DRIVER CRASH ***" for the point when the driver crash happens right at the bottom. Until then it looks stable. Any fan discrepancies can be accredited to me creating a custom fan profile around 2 minutes in when I realise that I didn't set up my usual super aggressive fan profile on GPU-Tweak (I like the fans to ramp up to 100% a little bit past 60 and use smooth scales in MSI Afterburner rather than instant 50% > 100%).

I'll see if HWiNFO64 has logging. At least that should hopefully show if the VDroop is causing problems.

 

[erocker]

With no overclock at all my card runs stable at 1050 core/1400 memory and I have increased the voltage from stock 1175mV to~1180mV.

Is it a Ghz edition card? If so, are you using the 12.7 beta drivers that are supposed to be used with that card?

 

[Jetlaw]

It's a standard 925 core/1375 mem card. I've tested it from a mild underclock, to stock, to my current overclock.

I'm using the most current 12.7 beta, this problem has also happened with 12.4 > 12.6.

It is "supposedly" an interaction between XSplit and AMD's ZeroCore tech on the 7xxx series card according to AMD tech support talking to XSplit devs, but this strikes me as a standard tech support answer because I've got ULPS disabled via the registry which from what I've read should also disable ZeroCore.

HWiNFO64 supports up to 100ms polling for sensors and thus I've gotten a much sharper picture of what happens when the drivers crash:

http://pastebin.com/qu0DN9ZJ

Best switched to raw view then copied into notepad with word wrap off.

MSI Afterburner voltage settings have the card at 1181mV, 1050 core clock and 1400 mem clock.

Prior to the driver crash my card is running at ~100% use (As per BF3's tendancy) at a respectable 57c GPU thermal diode/63c VRM. When the crash hits, voltage is dropped from ~1170mV to ~950mV, core clock is dropped to 500, mem clock is dropped to 150. I appear to have accidentally snipped it out for some reason but GPU memory use also plummets to 40MB for a brief instant in time.

Would this be consistent with Zerocore getting in the way? It seems perculiar that this interaction only happens with XSplit and due to the reply they got from the AMD devs they're loathe to do anything about it as AMD has seemingly admitted it's a problem on their end. If it was Zerocore, surely disabling it would alleviate the problem?

Quote:

The Radeon HD 7XXX series cards have the AMD Zerocore Power technology which might be causing an issue with Xsplit.

Since the cards are still fairly young, we are continously working on fixes and patches in our drivers to address these issues.

Thank you for the link to the thread as i've forwaded that as well to our developers.

[Additional Note]
My flatmate is also running a near identical 3930k/7970 setup and he's using an identical card other than it being Sapphire rather than Powercolor (Onl;y difference is my Powercolor has a quiet optimised fan and have added an additional DVI port over the reference single port making any attempts to flash BIOS completely blasted out of the water as it appears my card is unique in having the second DVI port.) - He also suffers the same issues.

 

[m1dg3t]

So you're complaining about crashing on a 200Mhz OC?

 

[Jetlaw]

Do I honestly have to underclock by 100mhz core and memory from default, overvolt to 1250mV, run the same tests with the same crashes and then post identical logs with identical crashes under identical circumstances?

Okay. That what a rehetorical question, but I'll do it anyway then I can complain about crashing with heavily overvolted, heavily underclocked settings and hopefully someone will have some idea as to if the problem is with the XSplit software, the 7k series Zerocore or something else entirely.

 

[m1dg3t]

You said the card is 925/1375 stock, it runs fine @ 1050/1400 & crashes @ anything over that, am I missing something?

 

[Jetlaw]

Driver crashes happen at any clock speed and any voltage purely for 7k series cards with the streaming software XSplit, but nobody is entirely sure as to the why. The cards are 100% stable otherwise and inclusion of XSplit causes some kind of conflict with something only in 7k series cards which by extension leads to a driver crash. Does XSplit's interaction with the GPU trigger something like Zerocore which will just drop your card's voltage and clocks and basically force a driver crash as we're seeing, or does the dropped voltage occur simply because the drivers crashed? I can't say, I don't have the experience.

All we know is that at some point, only while XSplit is actively recording, the drivers will crash and the card will drop to ~900-950mv/150 memory/500 core (For all 7k series cards and all currently known drivers) with one proposed workaround being to just ram your voltage up to 1300mV (Which doesn't work for 100% of people and infact led to sooner driver crashes using XSplit for me in particular).

AMD have stated this could be due to their Zerocore systems added in the 7k series, but disabling ULPS should also disable Zerocore according to what I've read (And I've disabled Zerocore) and the issues are still present.

I'm not entirely sure if the driver crash is causing the ~0.5s drop to 900mV/non-existant clocks, if someone else is causing it (which is in turn triggering the driver crash) or if these symptoms seem to be a part of something else. The card isn't overheating, the VRMs aren't overheating either. Everything seems to be stable. Heck, the card isn't even touching 60c thanks to super aggressive fan profiles I use.

It just seems that if you run XSplit at some point with only 7k series cards your drivers will crash. It doesn't matter if you're streaming a game or your desktop, streaming over the internet or recording locally.

It's a weird and frustrating experience and I'm wondering if anyone else has encountered anything similar, is able to diagnose the problem more precisely than "Could be our Zerocore" or if they have any idea about what I could possibly do to stop it if anything.

People here seem to know a lot more about graphics cards, architecture, driver crashes and more than elsewhere which is why I'm trying to ask enthusiasts if they've experienced anything like it (as opposed to asking people who don't know a damn thing about graphics cards at all. No use asking someone who has as much knowledge as I do if that.)

 

[m1dg3t]

Gotcha, was misunderstanding you before Sounds like it is a script/code issue, perhaps gather as many complaints as possible and bring them forward? Whoch seems like what you are trying to do