7970 Performance Hits - 2D/3D

[EXCellR8]

((SYS SPECS BELOW)) *using Win7 64-bit*

I am having a slight problem with my new graphics card, it's an ASUS HD7970-3GD5. I noticed it happening just recently but I've only had the card installed about a week or two. Anyway, it seems to be a driver issue I would guess but after I do some gaming (say an hour) my framerate drops or becomes erratic--like the card is struggling no matter how low I put the graphics quality. It's very strange even with titles my previous card (a 1GB 5870) had no problem running, this new card eventually starts acting up.

I am using Eyefinity to run games @ 6048x1080 usually and I'm currently using the latest 64-bit drivers (12.8) at the time of this writing. The card is not overclocked and I make sure to adjust the fan accordingly before playing any games; I haven't seen the temps go past 55-60c so I highly doubt it's a heat issue. I've tried tinkering with v-sync, DVI frequency, and AA settings but nothing seems to affect/correct this. Card is getting plenty of power I have a 1000w power supply; both 6 and 8 pin connectors are properly attached. My idea jar is sort of empty at the moment.

I was wondering if anyone could shed some light on this, or give me anything to try. I've combed through my background process but I haven't made any changes since I upgraded the card. I'm about to change my AV software and reinstall the drivers but my gut feeling is that's not going to make a difference either. I guess I'll just wait for a new driver to roll out. The problem does seem to go away once the system is restarted but it will eventually come back like there is some kind of GPU activity I can't seem to find a source for.

Anyways, thanks in advance for any input... specs are below.

--

Intel i7 920 @ 4Ghz
ASUS P6T motherboard
ASUS HD7970-3GD5 (catalyst 12.8 x64)
6GB Mushkin Redline DDR3 1600
ULTRA X3 1000w PSU
Windows 7 64-bit Ultimate

 

[Jaydip]

Man at your resolution you need to turn down settings to have a smoother game play experience or go xfire.

 

[RussianSensation]

Try this fix:
http://forums.anandtech.com/showpost.php?p=33937729&postcount=55

Download MSI Afterburner and HWInfo64 to monitor your GPU temperatures and GPU clock speeds in 3D mode.

Use "Sensors" Yellow button to check GPU clock speeds, temperatures, etc. in HWInfo64

To show GPU Clock speeds in MSI Afterburner, go to Settings ---> Monitoring tab, then select the appropriate categories under "Active hardware monitoring graphs"

Check that your GPU clocks are 925 as expected and GPU loading is high, your GPU temperature are fine. Try older driver versions such as Catalyst 12.7.

Generally though that resolution is way too high for a single 7970 to handle. Are you sure when your games Auto-Detect that they aren't selecting much higher image quality settings than they did on your 5870? Did you run any benchmarks such as Metro 2033, Crysis 1/2, Unigine Heaven 3.0 to check against your 5870?

 

[chimaxi83]

One card is definitely not enough for that resolution.

 

[3DVagabond]

It sounds like you need to disable hardware acceleration for Flash, as RussianSensation said. Do that and I'll be surprised if it doesn't clear up.

As far as the res on a single 7970 goes, many games will be just fine. Not all games are like Crysis and BF3. On the more demanding ones med or high (not ultra) settings should be OK. Turn off stuff like Ambient Occlusion, Depth of Field, and Global Illumination. Where possible set reflections to "earth and sky" only. In 3D rendering they are all HUGE resource hogs. An O/C'd 7970 will trade blows with the 6990 and a lot of people were running Eyefinity satisfactorily just a few months back with them.

 

[chimaxi83]

What specific software are you seeing this issue in? Anything 3D?

 

[Silverforce11]

Smells like viruses, prolly is.

 

[Meghan54]

....after I do some gaming (say an hour) my framerate drops or becomes erratic--like the card is struggling no matter how low I put the graphics quality. It's very strange even with titles my previous card (a 1GB 5870) had no problem running, this new card eventually starts acting up.

Card is getting plenty of power I have a 1000w power supply; both 6 and 8 pin connectors are properly attached.


ULTRA X3 1000w PSU

So, you're saying the card starts off working fine but eventually begins to have problems with the frame rates slowing down? And the card initially starts off working well, correct? That the gist of the problem?


First, I'd follow what Russian Sensations said, but I'd also investigate your old power supply. But I'd also like to suggest another area to investigate, and pardon the length of my post.....

Despite the fact the X3 Ultras were decent enough power supplies when they were built (Andyson builds and was based on the Andyson AD800 platform, an 800W platform), it's now over 5 years old and probably has suffered from some time/heat degrading, esp. considering the "quality" of its internal components, such as Teapo capacitors.

But first thing, I'd ensure all the modular cable connections are tight and making good contact. While they may appear tight and act that way when you plug them in, given the unit's age, I'd bet one or more of the modular connectors have loosened a tad, which could give some electrical delivery problems as the unit heats up. You can test these out with a DMM (digital multimeter.)

And, the Ultra power supplies, even the "X" series, were never known as "great" power supplies. For instance, yours, while rated 1000W, could only output 840W, or 70A, on the +12V rail, something modern 850W power supplies can do.

Add to that the voltage regulation in those wasn't stellar, and neither was the ripple/noise suppression, and you have a mediocre power supply today that probably isn't capable of putting out anywhere near its rated capacity....without issues.

I only suggest this avenue of exploration because the problem seems to happen as the system is stressed over time, as you mentioned in the op. What's probably happening is something is heating up and either losing contact (modular connector with poor contact, bad solder point, wearing out cap/MOSFET/choke/rectifier/etc.) or a worn component having trouble producing stable electrical output under stress (again, the MOSFET's, chokes, rectifiers, coils, etc.). Or the component problem could be on the card itself....cold solder point, bad capacitor, bad memory IC, etc.

But I'd look into testing your computer with a more modern and known good 600W or better power supply and see if that solves the problem.

 

[BrightCandle]

i owned two 7970's for the first 6 months of the year and I had the exact same problem. Using GPU-Z I tracked it down to a progress GPU memory leak on the cards, which seems to have been present ever since the first release. I raised a bug with AMD but in the 6 months I never got a response.

You can confirm with GPU-Z and recording the memory usage, you'll see it climb until it starts to hit near top and the frame rate will then drop off.

 

[EXCellR8]

Wow, lots of feedback, thanks; sorry I couldn't respond sooner...

Looks like I have a few things to try and I will definitely check for that memory leak, which has me a little worried. I don't understand why many people are saying the card isn't powerful enough to run this resolution. Why would I buy a $400 card just to not use the maximum resolutions for games? And why would AMD bother to make it Eyefinity capable if you can't use the bigger resolutions? Hell, my 5870 would run Dirt 3 in DX11 @ 6048x1080 no problem, with all the candy, no sweat. I realize this game in particular doesn't compare to say BF3 or Syndicate or something, but the performance was there--and is still there, but not for long.

I consider myself quite savvy when it comes to video cards but I've never had this problem with anything else. I've run all kinds of stuff, which usually either works or doesn't, but this 7970 is the first to show this type of performance degradation. I suppose my power supply could be on its way out considering the age but I know if I put my 5870 back in all will be well, because I've done that. Different card yes but power consumption not that different.

Regardless if I can't find the problem this card is gone, don't need any more broken things...

 

[blackened23]

Do you have any processes open in the background? Do you have a gpu clockspeed monitor such as MSI afterburner? I know problems that you describe can be attributed to 2d clocks, when you have some things like flash or browser youtubes in the background it will cause lowered clockspeeds. Obvious question, you have all browsers / processes closed while gaming, yes? Load up the MSI afterburner OSD overlay and try to keep track of your GPU clockspeeds while this happens, see if they lower.

 

[EXCellR8]

I'll still need to download afterburner as I'm basically only using CCC at the moment. As far as clocks go I haven't touched a thing. I put the card in, clean installed the drivers, set up eyefinity and then started gaming. Like I said I've combed my background process and made sure my AV wasn't interfering--but seeing as I haven't changed any software (aside from graphics drivers) since upgrading I can't think of those contributing.

Whenever I game I make sure my browsers are closed and the only thing that is running aside from antivirus is Steam or Origin or anything like that. Those are obviously needed for some games and I've already ruled them out for others. This happens whether those services are running or not. A restart fixes it but then after awhile it comes back, so I'm thinking BC was right to suggest a memory leak.

EDIT

The clock speeds (according to CCC and GPU-Z) are 500mhz core and 1375mhz mem, which are the stock speeds. Core speed will occasionally go up to 925mhz but I believe that's normal because of the 'powerplay' function? The memory usage has remained low but I'll keep checking it. Dragging windows around on the desktop the performance takes a hit and the core clock will go up to 925 and then right back to 500mhz when I stop. I think that's normal too. Sensors also show 12V @ 12.41v and VDDC going from 0.939v to 1.031v. Everything seems normal to me...

 

[thilanliyan]

500MHz is the power saving clocks. 925MHz is the stock clock. If it is staying at 500MHz while gaming, you have a problem. At desktop, 500MHz should be fine I think.

 

[Despoiler]

Sounds like a memory leak in whatever game you are playing? What fixes it? Closing the game and restarting or rebooting the computer?

 

[blackened23]

I'll still need to download afterburner as I'm basically only using CCC at the moment. As far as clocks go I haven't touched a thing. I put the card in, clean installed the drivers, set up eyefinity and then started gaming. Like I said I've combed my background process and made sure my AV wasn't interfering--but seeing as I haven't changed any software (aside from graphics drivers) since upgrading I can't think of those contributing.

Whenever I game I make sure my browsers are closed and the only thing that is running aside from antivirus is Steam or Origin or anything like that. Those are obviously needed for some games and I've already ruled them out for others. This happens whether those services are running or not. A restart fixes it but then after awhile it comes back, so I'm thinking BC was right to suggest a memory leak.

EDIT

The clock speeds (according to CCC and GPU-Z) are 500mhz core and 1375mhz mem, which are the stock speeds. Core speed will occasionally go up to 925mhz but I believe that's normal because of the 'powerplay' function? The memory usage has remained low but I'll keep checking it. Dragging windows around on the desktop the performance takes a hit and the core clock will go up to 925 and then right back to 500mhz when I stop. I think that's normal too. Sensors also show 12V @ 12.41v and VDDC going from 0.939v to 1.031v. Everything seems normal to me...

Check the clockspeeds while gaming? 500mhz in 2d windows is normal. 500mhz while gaming is not normal. You need to download MSI afterburner, enable the overlay, and monitor your GPU clockspeeds while gaming (it will appear as an overlay while you game, you do not have to alt tab)

500mhz isn't stock clocks while gaming. If you get 500mhz while gaming you're running at 2d clocks which means you have a process running in the background, disable all other applications such as browsers or flash/ youtube, etc. Before you try it, you can't check GPU clocks while alt tabbed - this is why CCC and GPU-Z are worthless. Get MSI afterburner. You need this to check GPU clockspeeds while gaming, and not alt tabbed.

 

[RussianSensation]

I would also force PowerTune to +20% just in case. Also keep running HwInfo64 in the background when gaming to check stability on both the 3.3V and 12V rails. If you deviate below 3.15V and 11.5V, then your PSU may be insufficient.

Use programs such as Ad-Aware or similar to do a full system scan for ad-aware on your computer. Using MSI Afterburner is key to monitor your GPU clocks in gaming.

 

[Elfear]

I would also force PowerTune to +20% just in case.

+1

Your card shouldn't be throttling at stock clocks but if you got an ASIC with high power consumption, it may be the culprit.

 

[EXCellR8]

Okay, I will do some gaming and see if the clock is where it should be. I can actually monitor this right from my keyboard's LCD screen when GPU-Z is running. I've had it up for awhile now and there hasn't been any odd memory usage or anything from sitting on the desktop. Could be a different story when I run a game for an hour or so. I can also use CCC to adjust the power to +20% which is the max. Do I do that or should I use the MSI tool?

Sounds like a memory leak in whatever game you are playing? What fixes it? Closing the game and restarting or rebooting the computer

to clarify this... it is a full reboot of the system that seems to correct the issue.

Thanks again for all the help! I'll post back when I get a chance to do some 3D tests.

 

[RussianSensation]

Adjusting Power tune in MSI Afterburner works without problems. Set it to +20% and hit Apple. You can save various profile settings for your GPU which comes very handy for overclocks. Also, it allows you to create a custom fan profile curve should you not be happy with the card's BIOS fan setting profile. Don't be hesitant to install MSI AB as it is probably the #1 overclocking software for GPUs right now!

 

[EXCellR8]

Okay so here's what I found, it's... strange.

I installed MSI AB, put the power threshold up to +20% and decided to run a game I know my old card could handle at max settings and 'extreme' resolution (6048x1080). I ran Need for Speed Hot Pursuit, the new one, for appox. 35min before the problem showed it's ugly face once again.

I was using the LCD function of my keyboard w/ Afterburner so I could monitor the GPU off-screen. Firstly, when running in a 3D environment the core clock stayed @ 925mhz, normal. I watched the memory usage slowly creep up from around 800mb or so to around 1050mb which was around the time my framerate took a serious hit. That's obviously not the cap for the onboard memory and my temperature stayed around 56c, which is fine. Here's where it gets weird...

GPU usage was nearly CUT IN HALF! It dropped from approx. 85% to well into the 40s! It stayed like this until I shut the game down and now it's normal again sitting on the desktop. I paused the chart on AB and had a look...

You can clearly see where it takes a huge dive down, and this happened while I was driving around in the game. You can also see where I exited the game after and if you notice the GPU temp slowly going down as if the card just got lazy or something. The game ran like an absolute dream before this point; clock speeds remained normal during this slow down so I have no idea what to make of it.

So, do you guys think I have a faulty card here or is there something I can do? Should I maybe try a BIOS update or just wait for a new driver? Never before have I seen this happen so I'm definitely puzzled at the moment.

 

[Plimogz]

I was going to suggest CPU thermal throttling (dust, displaced heatsink, ?, etc...) but a quick search led me right back to our own AT Forums: http://forums.anandtech.com/archive/index.php/t-327089.html. And I now imagine a better hypothesis may be power throttling.

Going through that thread yielded this:

The P6T seems to start Turbo throttling somewhere around 4 GHz and 1.40 volts. The formula is based on current, power consumption as well as temperature so if you're air cooled and your temps are high then you might start to have this problem at a lower overclock.

Given that you're using the same motherboard and running that same frequency, I'd say that the correlation would be worth investigating.

I'd arrange the MSI Afterburner monitor/graphing to display GPU frequency as well, in order to see if the FPS drop coincides with a GPU clock change or not. Because if the clock stays stable at 925 while the FPS drops it would rule that out. (setting up the OSD and keeping an eye on clockspeed works as well, but wouldn't produce nearly as interesting a screenshot )

Furthermore, I'd use something to monitor and chart CPU speed+temperature at the same time as you're doing the same for GPU. IIRC Speedfan is good for this.

Should the Speedfan graph show CPU speed dropping at the same moment that Afterburner charts the FPS and GPU utilization drop, while showing GPU speed stable at 3D clocks, then I think CPU throttling of some kind would be undeniable.

 

[EXCellR8]

I would think if the system was power throttling it wouldn't take so long to affect gaming performance. I've been using this system and CPU speed for about 2 years now and this has never happened. My cooling system is pretty beefy as well... it's a 1/2" ID Swiftech X220-EDGE liquid loop with dual 120mm radiator. I've done plenty of gaming during this time period and I've been using the big resolution for about 9 months I think. The 5870 worked fine for 2 years, and I know the power consumption is different, but it's not drastic.

I still think it could just be a driver bug, but I will do a little more testing before coming to any conclusions. I think I'll try DiRT 3 for awhile, since it's DX11, and my previous card had much joy running it at eyefinity resolution. I've been choosing racing games because one, they are my favorite, and two because it's much more noticeable when the framerate takes a dive.

I'll post back my findings and I will make sure that I monitor the CPU as well.

 

[EXCellR8]

So I got around to testing the card in a DX11 environment with realtime gaming. Need for Speed only supported DX9 so I decided that Dirt 3 would provide a more relevant testing scenario. Here's a brief outline of the settings I used:

DiRT 3 ~1.5 hours play time

resolution - 6048x1080
texture quality - max
shadows quality - high
effects quality - high
post process - high
geometry/decals - high
anti-alias - 4x MSAA
sound - high w/ 5.1 surround

Using these settings the game ran very smoothly, without a hitch even, and not once did the performance degrade. I did notice that GPU usage was topped out @ 99% (instead of being the the 70-80 percentile) and the temps did nearly touch 70c, which I figured to be normal on full load. I believe about 1300mb or so onboard memory was used but I didn't see any anomalies with CPU frequency.

As you can see in this chart the execution here is much more normal, and you can see the high GPU usage in-game and then dropping to the regular level when I quit. You can also see where I adjusted the fan after etc. This is a game that my older 5870 would struggle with at this resolution if I enabled AA or maxed out the effects or post process quality--but otherwise it would run.

So, seeing as I didn't experience any issues with this game, do you think that maybe it's a DX9/shader issue or a more isolated problem? I honestly can't see any other factors but the graphics card in this, but I will continue to do some tests.

 

[RussianSensation]

Do these GPU % utilization issues you experience occur only in DX9 games? For example, if you play modern DX10/11 games, you experience no problems in those games? BF3, Dirt 3 DX11, Batman games, all fine?

 

[EXCellR8]

That's what it seems like, but still may be a bit too early to tell. I've played a couple of DX11 games as well as a DX10 title, and the problem didn't occur. I'm going to continue testing though as time allows and have a look at my DX installation. For now though it appears DX10 and 11 show normal patterns of GPU usage, and performance never degrades.