Sapphire 7970 drops to 10% gpu load in game. Game goes into slowmo

[cmb21]

Hi,

OK card was purchased 2nd hand. I installed it and ran 3Dmark11 a few times with no problems. It has been playing BF3 brilliantly until i had a weird problem last night. I have put about 15 hours of BF3 on the card with out a single issue until now.

Anyway i was about an hour into playing then the game went into slow motion for me, i could move and shoot but it was all in slow motion, i had gpu z open on my 2nd monitor and it showed the GPU load had dropped from 99% to 10%

I restarted my machine, went back into BF3 and it was fine again, played for another 45 minutes maybe before chaning maps and while i was in a jet it happened again. GPU load dropped to 10% and the game went into super slow mo for me.

Temps maxed out at 48 Celsius while i was playing (card is watercooled with an EK full block).

Just wondering if anyone has any ideas what has happened and if there is anything i can do to fix it. Could it be psu or other component related or is the 7970 dying?

Other gear is
i7 950
MSI X58 Pro
12GB GSkill Trident F3-16000CL9T
I think my psu is a super flower 750W modular
Sapphire 7970 with EK waterblock.

 

[El Tigre]

If you can, I would run a stress stest with Furmark to test the GPU. I would also run a memtest to test the integrity of your memory.

You can also try reinstalling the GPU driver.

 

[cmb21]

If you can, I would run a stress stest with Furmark to test the GPU. I would also run a memtest to test the integrity of your memory.

You can also try reinstalling the GPU driver.

Thanks i will try those now, i just installed furmark, does it matter which test i run? Burn in or benchmark?

EDIT: I just ran the Benchmark 1080 preset and it got a score of 2569 with no problems, the test is quite quick though so the GPU only got up to 37c. Should i run the 15 min burn in or not?

 

[cmb21]

So memtest has been running for an hour and 45 minutes now and this is what it has come up with so far. (how long does memtest usually run for?)

Figured it was easier to take a photo of the screen than try type it out lol.

Hoping maybe its a stick of ram causing the problem and not the video card, i dont have much experience with memtest or ram errors, so not sure exactly what this result means.

 

[El Tigre]

Let the memtest finish, it should be finished shortly, then I can help you once all the results are in.

Also, run Furmark under Burn In for about an hr and have either MSI Afterburner and/or GPU-Z to monitor the temperature an such of the card while it's going through the stress test.

 

[Graphite946]

Do mem test with just one stick of ram in. Easiest way to narrow down which is the problem.

You said you have a water block. What are you doing to keep the VRAM cool?

 

[cmb21]

Let the memtest finish, it should be finished shortly, then I can help you once all the results are in.

Also, run Furmark under Burn In for about an hr and have either MSI Afterburner and/or GPU-Z to monitor the temperature an such of the card while it's going through the stress test.

Thanks. I will let it finish then rum Furmark.

Do mem test with just one stick of ram in. Easiest way to narrow down which is the problem.

You said you have a water block. What are you doing to keep the VRAM cool?

I was under the impression that the EK Full Cover waterblock for the 7970 cooled the VRAM as well.

 

[cmb21]

Just left mem test alone as i had other things to do, its now been running for almost 6 hours and still going.

I simply booted from the memtest disc and it went straight into running the test, i did not configure it or anything. So how many passes is it meant to before it finishes? It shows it is on Pass 3.... is that really necessary? The error i posted before is still the only one to have come up.

 

[Kari]

memtest86 will continue looping as long as you let it run... I usually do at least 5 passes if it doesn't get any errors.
Anyhow it has already caught an error (or 12). You could rerun it with just one stick installed to see which one is faulty. It might be fixable by just slightly increasing ram voltage or relaxing the timings a bit. (basically just like when overclocking to get it stable) Or one of the sticks can also be completely broken and needs to be replaced...

also what are the temps on the CPU when the slow downs occur?

 

[unksol]

Just left mem test alone as i had other things to do, its now been running for almost 6 hours and still going.

I simply booted from the memtest disc and it went straight into running the test, i did not configure it or anything. So how many passes is it meant to before it finishes? It shows it is on Pass 3.... is that really necessary? The error i posted before is still the only one to have come up.

Edit: what Kari ^ said. Best me lol

Memtest doesn't stop still you stop it. It will run for ever on it's own. Some people let it run 24 hours but 3 passes is plenty and you already know you have a problem that needs addressed. try it one stick at a time to find the problem.

 

[cmb21]

Well when i stopped it at 3 passes (6 hours) it had 27 in that errors tally, but still had only the one failing address listed like in the picture.

CPU temp was about 65c when the slow down happened.

 

[Kari]

ok, so the temps are all fine then. bit weird really, since the problem description fits perfectly for an overheating situation...
and memory errors usually show up as system freezes or graphic artefacts... might be completely different problems...

 

[jjb8675309]

link to the psu? no offense but the brand sounds really sketch and performance loss like that could indicate a dying or frying psu...

 

[cmb21]

link to the psu? no offense but the brand sounds really sketch and performance loss like that could indicate a dying or frying psu...

None taken, at the time i needed a PSU quickly as i fried my previous one and the local stores didnt have any of the decent ones in stock. Its done a decent job for the last 3 years.

Although it was only running a 5830. So it seems strange that after 2 weeks of a new GPU i have issues. I dont think its the GPU i ran Furmark bench test for 45min without an issue and the GPU got up to 50c, which is less than what it is in BF3. Hoping like hell its not the GPU.

http://www.super-flower.com.tw/products_detail.php?class=2&sn=6&ID=22&lang=en


Thought i better add this:

The 2nd time i had the issue, i just waited it out and the GPU load picked up again and the game went back to normal.

EDIT:

Ok so Furmark burn in (settings are full screen 1920x1080) has now been running for 1hr11min GPU temp us hovering around 50/51 Celsius, FPS - min 40 - max 46 - avg 43. If there had been any load drop during the test then i think you could safely assume it would be reflected in the min fps.

Can you run something like Prime95 at the same time as furmark burn in? As the cpu load is only 15% atm and when the gpu load dropped it was around 45-50%. I was thinking maybe by stressing the CPU it would pull more power from the PSU and that would hopefully replicate the problems (assuming its CPU related).

Going to leave furmark running a little longer then i will start testing individual sticks of ram using memtest.

 

[cmb21]

Furmark burn in went for 1hour 35 min without a problem.

Been trying to get the ram tested in between having to do other things. I tested with just 1 stick (i have 2 6GB kits in the machine so 6 x 2gb sticks all up). I only managed to test the 1 so far but at 2 passes it had come up with 10 errors and 1 entry in the list for a failing address.

 

[alexander0884]

You said you got the card secondhand? Could have some form of damage on it. Where did you buy it from? Ebay? Does the seller sell hardware regularly?

Change the stick of ram and redo the test to eliminate RAM altogether. Here is what you should have done in order:

#1: ensure you are running BIOS at default. NO overclocking on ANYTHING.
#2: Install a stock cooler (If you have it) on your card
#3: redo the mem test with the same stick of ram.
#4: if errors persist, redo test with same stick of ram but higher voltages in your BIOS
#5: If same errors persist, redo the test with different stick of RAM
#6: if same errors persist, this is no longer worth your time, RMA and get a new part else contact ur seller and demand a refund for faulty item and let him RMA it for you.

*** Side note, even though you bought the item, dont mention it is second hand during RMA, the warantee should not be affected. Just say you dont know what happened, suddenly starts lagging when you play games. You could contact your seller and ask him for any backups/receipts and any additional warrantees/gurantees the card might have to support your claim.

PS: Paste ur PSU details here. Model number, make etc. Recommendation is that you run that card with a single rail 12v which would give you 54 amps @ 650Watt PSU. Post ur tech details of your PSU go go go.

 

[cmb21]

I dont quite understand why i would be doing an RMA on the GPU if the ram errors persist. The reason i was running memtest on the RAM is another member suggested i do so, a little confused as to why ram errors would warrant the RMA for the GPU (as i said i dont have much experience with memtest and what the results mean). Can the gpu cause ram errors in memtest? i didnt think it had anything to do with the gpu.

In all honesty after stress testing the GPU for in total over 3 hours i am finding it hard to believe its the card. I posted after the problem occured twice within a short amount of time, so it is not happening all the time. I haven't been able to replicate the problem since it initially happened which is why i am leaning towards the PSU being a problem. Thats why i asked if you can run a CPU stress test such as prime95 at the same time as a GPU stress test to try and put more load onto the PSU and see if that can reproduce the problem. Since CPU load was 50% when it happened and its only 14% during the cpu stress test so it wouldnt be drawing a huge amount of power.

Ram failing is generally blue screen as far as i understand.

I put a link in another post to the PSU

Super Flower (yeah i know it sounds sketch) 750W 80Plus Bronze
SF-750R14HE

The seller supplied a copy of the receipt when i purchased the item. And yes i have the stock cooler.

Nothing is overclocked everything is running completely stock settings.

 

[alexander0884]

I suspected it might be your PSU, the guidelines I wrote out are simple ones to eliminate VRAm vs your normal RAM incase of some form of damage on 2nd hand products. Its not unusual.

I looked at the specs, I can gurantee you 100%^ its your PSU.

DC OUTPUT
+3.3V
+5V
+12V1
+12V2
+12V3
+12V4
+12V5
+12V6
-12V
+5VSB

　

　

Max output Current
24A
30A
20A
20A
20A
33A
33A
20A
0.5A
3.0A

If you look at the figures above, you have multiple 12v rails, and none of them exeed over 33amps. Your system simply throttles the card as not enough power.

 

[cmb21]

I suspected it might be your PSU, the guidelines I wrote out are simple ones to eliminate VRAm vs your normal RAM incase of some form of damage on 2nd hand products. Its not unusual.

I looked at the specs, I can gurantee you 100%^ its your PSU.

DC OUTPUT
+3.3V
+5V
+12V1
+12V2
+12V3
+12V4
+12V5
+12V6
-12V
+5VSB

　

　

Max output Current
24A
30A
20A
20A
20A
33A
33A
20A
0.5A
3.0A

If you look at the figures above, you have multiple 12v rails, and none of them exeed over 33amps. Your system simply throttles the card as not enough power.

Yeah thats fair enough in regards to the ram.

So the 7970 needs a fair chunk more than that? Does it make a difference that it uses both the 6 pin and 8 pin connector? (edit: im guessing the 6 and 8 pin power connectors would still only be 1 12v rail so wont provide enough current for the card? or have i missed the mark completely?)

Also seems strange that it has only happened twice, but i guess you would have to assume the card is trying to draw to much power so the PSU must be getting over worked and starting to show the results of that.

 

[alexander0884]

Yeah thats fair enough in regards to the ram.

So the 7970 needs a fair chunk more than that? Does it make a difference that it uses both the 6 pin and 8 pin connector? (edit: im guessing the 6 and 8 pin power connectors would still only be 1 12v rail so wont provide enough current for the card? or have i missed the mark completely?)

Also seems strange that it has only happened twice, but i guess you would have to assume the card is trying to draw to much power so the PSU must be getting over worked and starting to show the results of that.

I can not tell, a quick google search indicates that the card doesnt ened more than 30 amps, however from previous experience I know that I should have at least 54amps capacity on a 12v rail. However with these newer cards they are supposed to be mroe power efficient, I am not entirely sure how it fits in that retrospect.

A quality PSU would have a single or a double 12v rail, that should in theory support any high end power requirements. For a 650watt PSU it should be a single rail, for an 850 you can do double but splitting it further I dunno. I know modular PSUs (such as yours) are not great at power output, only at power distribution. Had this eben 1-2 years ago,I would defo say its PSU but with these enwer cards I can not be sure.

Best make a thread on the PSU forum and link back to this just to get it verified before you do anything.

 

[Kari]

actually in some atx standard it was specified that the max current for a 12volt rail to be 22A or something like that and if more was needed it was recommended to use multiple rails.
anyways the psu is wired so that the different rails feed different output sockets on the psu, so can you swap those pcie connectors around a bit... though as the furmark went without problems there isn't any pcie power problems as such.

and yes, you can run prime95 simultaneously. that will really make the system sweat though so keep an eye on the temps and whatnot... (and the cpu/mobo should be wired to different 12v rails, so it shouldn't cause any problems with the pcie power quality as long as power usage stays below 720W as specced)

reference 7970 can use up to 270Wats under furmark, 190W in games (but it draws that power from 3 different sources, pcie slot on the mobo is specced to deliver up to 75W, 6pin 75w and 8pin 150w... and it seems these are connected to different rails on the psu)
http://www.techpowerup.com/reviews/Sapphire/HD_7950_Flex/26.html

((and from personal experience: faulty system memory can cause graphic artefacts for example in games...
edit and also: booting windows with only the faulty stick installed can mess up your windows installation ))

 

[cmb21]

Thanks for all the help so far guys, really appreciate it.

I have some more testing to do, but i have played a few hours of BF3 and its yet to repeat itself.

However out of interest i did a quick test with furmark burn in and prime95 running at the same time.

Its a pain in the arse to try and copy the data from the notepad document that gpuz logged to so i will just do a couple lines as an example.

Load VDDC Current [A] VDDC Current In [A]
83 122 13.2
95 122.3 13.1
95 99.8 11.8
89 118.3 11.9
97 114.5 12.0
86 119.5 13.1
84 38.3 3.8
94 38.5 3.8

Now as soon as i stopped prime 95 the data went back to being very consistant

Load VDDC Current [A] VDDC Current In [A]
100 114 11.4
98 113.5 12.3
99 109.8 12.2
99 111.3 11.9

And it carries on like that.

That just doesnt seem right to me.

Edit: I did space that out nicely but it removes the spaces when i submit......

 

[RussK1]

1 x 6pin AMP requirement = 75 x 2 = 150w / 12v = 12.5amps

2 x 6pin AMP requirements = 75 x 3 = 225w / 12v = 18.75amps

6 + 8pin AMP requirements = (75 x 2) + 150 = 300w / 12v = 25amps

2 x 8pin AMP requirements = (150 x 2) + 75 = 375w / 12v = 31.25amps

 

[Graphite946]

A lot of very, very good info in this thread.

 

[cmb21]

No thoughts on the weird inconsistant power readings when i run prime95 in conjuction with furmark?

Or should i be asking that in another section?