Sign in with
Sign up | Sign in
Your question
Solved

SLI GPU Unstable Clock and Crash

Last response: in Graphics & Displays
Share
February 7, 2014 10:50:32 AM

I have two EVGA GTX 650 Ti Boost graphics cards hooked up in SLI. (I'll post the rest of my computer stats below). I originally started with one, and occasionally would have an error that the "display driver has stopped responding and recovered." Upon posting here I was able to stop that error from happening by going into the NVIDIA options and changing my power settings to "max performance" instead of "adaptive."
Then I got a second card and set it up in SLI. Now I get bursts, maybe 3 or 4 in a row, of these "display driver stopped responding" errors. All my options are the same that fixed it last time. Downloading the EVGA precision X software, I can see both GPU's. I notice that I get these errors when my clock speed is around 348 MHz, and when it clocks up to 1071 MHz it is fine. My first question is why is the clock changing from .3 Ghz to 1.1 GHz, is this just normal operation?

Secondly, I notice that both of my GPU's have different clock speeds, and voltages. My initial GPU has 1175 MHZ clock and 1174 mV voltage, while the second (new) GPU has 1071 MHz and 1075 mV Voltage. Is this operation okay or is there a way I can get the two cards to sync together in both voltage and clock?

Any help or comments would be appreciated. thanks

Specs:
Gigabyte Ga-990FXA-UD3 motherboard
750 W PSU
AMD FX-8120 3.1 GHz CPU
8 GB Kingston Hyper X Ram
500 GB Harddrive, 7200 RPM
Windows 7
a b U Graphics card
February 7, 2014 10:55:54 AM

use afterburner, I never had much luck with that tool gave me issues, msi you can sync the clocks to match.

300mhz is probably the 2d clock speed and when going to 3d it ramps up to full speed
m
0
l
a b U Graphics card
February 7, 2014 12:26:24 PM

it could be that the cards are different clock rates or power fluctuations or a combination of the two. I use Precision X on my system and my cards are at Super Clock frequencies. they are identical though.
m
0
l
Related resources
March 14, 2014 7:53:24 PM

ylweyez said:
I have two EVGA GTX 650 Ti Boost graphics cards hooked up in SLI. (I'll post the rest of my computer stats below). I originally started with one, and occasionally would have an error that the "display driver has stopped responding and recovered." Upon posting here I was able to stop that error from happening by going into the NVIDIA options and changing my power settings to "max performance" instead of "adaptive."
Then I got a second card and set it up in SLI. Now I get bursts, maybe 3 or 4 in a row, of these "display driver stopped responding" errors. All my options are the same that fixed it last time. Downloading the EVGA precision X software, I can see both GPU's. I notice that I get these errors when my clock speed is around 348 MHz, and when it clocks up to 1071 MHz it is fine. My first question is why is the clock changing from .3 Ghz to 1.1 GHz, is this just normal operation?

Secondly, I notice that both of my GPU's have different clock speeds, and voltages. My initial GPU has 1175 MHZ clock and 1174 mV voltage, while the second (new) GPU has 1071 MHz and 1075 mV Voltage. Is this operation okay or is there a way I can get the two cards to sync together in both voltage and clock?

Any help or comments would be appreciated. thanks

Specs:
Gigabyte Ga-990FXA-UD3 motherboard
750 W PSU
AMD FX-8120 3.1 GHz CPU
8 GB Kingston Hyper X Ram
500 GB Harddrive, 7200 RPM
Windows 7


I am actually going through the very same frigging thing....I do not know why but one will be running at 1200 mhz and the other will be running in the 300's and yet they are in sync.... I have the voltage set to the same numbers and everything else, but they keep being at huge variable differences, and I wanted to ask if you ever found out what was the problem or how to maintain similar speeds wiht your gpu's
m
0
l

Best solution

a b U Graphics card
March 14, 2014 8:08:44 PM

* What brand and model power supply do you have?
* Also, what temperatures do you get right before the crash for your video card?


------------------------
How to calculate power supply requirements.


Quick summary.
Goal: Find out the most power your computer would consume under the worst conditions and then pick a power supply that can provide that much power.
To do that, you would:
1. Find out the maximum power each component uses under heavy load.
2. For simplicity, you can just focus on two components: the video card(s) and CPU ... as these use up almost all the power.
3. Use DC power numbers (not AC, if possible).
4. Add all the number together and then add some extra padding to those numbers to account for random extra parts and so you don't run at 100% capacity.
5. Check if your power supply can handle that on the +12v rails.

Practical example
Here's a more practical example of what I currently do to figure out power supply requirements.

1. video card power
1a. Go to http://www.techpowerup.com/ and lookup your model; then find the "Peak" power draw for you model. For example, this page shows that the MSI R9 290X uses 263W at peak load:
http://www.techpowerup.com/reviews/MSI/R9_290X_Gaming/2...

2. CPU power
2a. Go to http://www.xbitlabs.com/articles/cpu/ and search for your CPU model or one that has the same TDP as your model. Then find out the max power draw under full load for all cores.
2b. Note: xbitlabs gives you AC power for the CPU + motherboard & RAM. DC power is preferable (it's more accurate), but this will work well enough. (Sometimes I convert the AC power to DC if I need more accurate numbers.)

3. Add the numbers together and this is your minimum DC power that you need on the +12v rail in order to run at full load. Why the +12v rail? Because the CPU and video card mostly use the +2v rail and you need to make sure you have enough on that rail.

4. Add some padding to those numbers so you don't run your power supply at full load. Use this as your realistic target number for the +12v rail.

5. BONUS: estimating overclocking power requirements. If you want to estimate power requirements to overclock a video card, there is a neat trick that *might* work:
http://www.playtool.com/pages/psuconnectors/connectors....
The PCI-Express slot uses 75W (DC) (6.25amps on +12v rail)
The PCI-E 6-pin plug uses 75W (DC) (6.25amps on +12v rail)
The PCI-E 8-pin plug uses 150W (DC) (12.5amps on the +12v)
So, just add the number of plugs your video cards has + the PCI-Express slot power to get the theoretical maximum your video card could use during an overlock. While I don't know if this is an accurate way to check, it seems like it *might* work in theory.


And that's all there is to it. However, if you would like more details... and an explaination as to why I did things this way, read on.

Basic Concepts you Need to Understand
1. Power supplies split their power up into different rails (or voltages). Different parts need different voltage types, such as +5v, +12v, etc.... While a power supply might have a high wattage rating, it may only provide a fraction of that amount to certain rails. Thus, you must ALWAYS check how much power is provided to each of the rails.
2. For simplicity, the ONLY rail you need to worry about is: +12v Both the CPU and Video card mostly use the +12v rail, so you can generally ignore the others. So, basically, always check the +12v rail on a power supply.
3. Power supplies convert AC power to DC power, but lose some of that power in the process. (All your computer parts use DC.) For example, when a power supply converts 100W AC power to DC, it may only end up with 80W (DC) power that you can actually use. This would be an 80% efficient power supply.
4. You can convert DC power (watts) to amps by dividing by 12. So 100W DC power is 8.3amps on the +12v rail.

Thing to be aware of if you want to do things correctly
While I have provided you with enough information to calculate power supply amounts fairly accurately, there are lots of little issues you should be aware of that could cause you to get things wrong.

* Never trust a review site that does not isolate components. You need to get power usage numbers that completey isolate a particular component (like your video card) from the rest of the system. Techpowerup.com does this correctly for video cards. They tell you the power draw ONLY for the video card and nothing else. Many sites, however, do this incorrectly, but telling you power draw for the entire system. The reason this is worthless, is because what if you have different parts in your system (which everyone does), then those numbers are meaningless to you. Unfortunately, nobody isolates only the CPU in their reviews; however, xbitlabs comes close: they disable the video card and just give you numbers for the CPU+motherboard & RAM.

* Some sites might subtract the power at full load from the idle power and claim that as the power draw for the component. This is also useless, since parts already consume power under idle, so you will never know the real max power draw this way.

* DC power numbers are better (and more accurate than AC) power numbers. If a review site gives you power draw "from the wall" or in AC, then this is quite inaccurate. The reason is each power supply has a different efficency rating... and not only that, but at different loads, the power supply even has a different efficiency. So, you will never know what the true DC power load really is. Also, you cannot use AC power directly, because your power supply will have a different efficiency than their power supply, so the numbers won't match your system. Again, techpowerup.com does this correctly for video cards. Another major flaw in using AC power numbers is they don't tell you what rails the power is using... which is very important sometimes.

* A good power draw test must stess out the part being test. In order to get useful numbers you need the maximum power draw from that part under the heaviest load. As a side note, this also reveals another flaw in testing both the CPU and video card together... since many reviews stress out only one part and not the other... which tells you nothing as that does not maximize the load on the system.

Examples of good video card testing (which you can use to get power draw numbers):
http://www.techpowerup.com/ -- look for "Peak" power usage; this number is ONLY for the video card and it is in DC power for the +12v rail, so it is very exact (and useful).

Examples of good CPU testing (which you can use to get power draw numbers):
http://www.xbitlabs.com/ [CPU:good; GPU: bad]-- look for the "100% Load" test... the one that uses ALL cores, not just 1 core. The good thing about xbitlabs is that they don't connect a monitor, so the video card is less of a factor in the power numbers. Better yet, some newer reviews even provide DC power numbers and will isolate the CPU and motherboard separately (so they are even more accurate)! There are a few flaws to note: (1) they combine CPU + motherboard + RAM into the test; however, this can actually be kind of useful as you don't have to add as much padding later (2) they test using AC power ... so you will either need to convert to DC by looking up how efficient their power supply is... or just use the AC numbers and think of it as "extra padding to be safe" since AC numbers are higher than DC. Note: Do NOT use them for video card power; they test video cards wrong.


Examples of sites that get some tests right and some tests VERY wrong. Use them for the good tests; ignore them for the bad ones:
Examples of sites that *might* be ok:
anandtech.com [CPU only; not GPU] - their video card numbers are no good. However, their CPU tests seem to isolate from the video card properly, however, I think they are also AC numbers that include the motherboard and RAM and base parts in the amount.
tomshardware [GPU only; not CPU] - their CPU tests are no good, but the video card test appear to show DC power draw only for the video card. However, there is no data to confirm if this is true... and you have to be careful which graph you read as some of them are the DC power and some of the charts show AC power for the entire system (which you don't want).


Examples of VERY bad power supply test/reviews (NEVER USE these sites for power draw numbers -- these sites are beyond bad!):
Here's what they do wrong:
for video cards they fail to isolate the video card from the CPU; instead they give you total system power in AC, which gives you wildly inaccurate and useless numbers
for CPU, they do the same thing: they fail to disable the video card and they also give you AC numbers, so again, you get wildly inaccurate (and useless) numbers
A list of sites to NEVER use for power requirement numbers:
anandtech.com (video card only)
bit-tech.net -- (both)
guru3d.com -- (both)
hardocp.com -- (both)
hardwareheaven.com -- (both)
legitreviews.com -- (both)
tomshardware.com -- (cpu only)
vortez.net -- (video card)
xbitlabs.com -- (video card only)

Disclaimer: Some of these sites are actually very good review sites, when not testing power draw. In fact, some of them, I've used for years now ... so I'm equally surprised that so many good sites get power draw testing wrong. However, some of them are also very bad at testing other things too (and I wouldn't trust them for any review types). I will not comment on which is which.

Share
!