TR Forums

This isn't a new problem but since I just upgraded to windows 10 and redoing my GPU overclocks, it came up again and started to annoy me a bit, thought I might just ask and see if anyone has a solution.

So trying to overclock a titan X, but sometimes, for unknown reasons, the card will lock to a very low frequency cap, like 405Mhz and will never go beyond that, in GPU-Z the reason for performance cap is read as "utilization". The only way to fix this is to reboot, and then it's ok. Seems to only happen when I'm messing around with the clock settings and won't happen any other time. Yes I'm using a moded BIOS, if that's the problem then I'll just not bother because there's no point if I can't take full advantage of the hardware with a custom BIOS.

Are you 100% certain the GPU is actually running at those clocks? In other words, if you fire up a benchmark does it only give you half the performance you're expecting?

GPU-Z is an awesome tool but it's written by one guy and I wouldn't be surprised if the custom BIOS means GPU-Z is just reporting the wrong values.

What is it you're wanting to unlock with the custom BIOS - higher voltages?

This is normal function with modern GPUS. They throttle when they're not being utilized. Even with light gaming, they won't clock to the their highest/boost speeds unless you force them. When i play LoL my 980Ti likes to chill in the 880 Mhz range. DE:HR it goes up to 1200 Mhz. To make the GPU hotter and cost you more money even when it doesn't need to, go to Nvidia Control Panel and adjust the GPU power settings for Max performance. That should improve things. It still may throttle on Windows, but you'll hit max clocks more often. Windows 10 is the unknown variable in the equation; don't know how drivers are working with the new OS. I would've waited to upgrade til they get some of the kinks out, personally.

How can you be confident it's not the custom BIOS either directly causing the low clocks, causing incorrect low reporting of clocks, or making the card need to be at 405 MHz?

Also, did you mod it yourself or find one like this, where somene on the internet snagged a refernce BIOS from one messgae board and "fixed" it to raise the power target to 350 W at 100%, 425 W at 121% (the does not compute), "fixed" it to raise the max voltage to 1.281 V (!) and uploaded it to Dropbox then posted the link on another message board, despite having no Titan X card to test it on?

1. I know because I can test the performance in like you know? games?

2. The BIOS mod is mainly to provide more power, I made it 140% but I'm not entirely sure how it works so I can't say how much that actually is, it could be more than 140% of the original 250W.

3 I expected more from here than "oh it's normal you're just not utilizing it enough" Really? You really think I didn't know if there was actually any problem with it?

4 I moded it myself with I forgot, maxwell BIOS tweaker or something. It could be the BIOS for all I know, even though it was based on a stock titan X BIOS. It's just not a huge problem because it never happens during normal use. It only happens when I'm constantly changing clock speeds with an OC software, and like I said, it's fully fixed when I reboot.

chengong wrote:

3 I expected more from here than "oh it's normal you're just not utilizing it enough" Really? You really think I didn't know if there was actually any problem with it?

It really is supposed to do that, sorry. If I cap the FPS to 60 on my GTX's at just 1080p for example, they will stay utilized on something like GTA V 35-40% tops. Clock speed is 907Mhz most of the time and it's hitting 60 fps solidly without a hiccup.

You can turn it off in the Nvidia CP - It's labelled as "Power Management Mode" under the Manage 3D Settings section. Set it to "Prefer Maximum Performance" and you'll see it aim for max boost clocks under any kind of workload instead (mine hit 1525Mhz for example if I do that by just playing a YouTube video).

geekl33tgamer wrote:

chengong wrote:

3 I expected more from here than "oh it's normal you're just not utilizing it enough" Really? You really think I didn't know if there was actually any problem with it?

It really is supposed to do that, sorry. If I cap the FPS to 60 on my GTX's at just 1080p for example, they will stay utilized on something like GTA V 35-40% tops. Clock speed is 907Mhz most of the time and it's hitting 60 fps solidly without a hiccup.

You can turn it off in the Nvidia CP - It's labelled as "Power Management Mode" under the Manage 3D Settings section. Set it to "Prefer Maximum Performance" and you'll see it aim for max boost clocks under any kind of workload instead (mine hit 1525Mhz for example if I do that by just playing a YouTube video).

It isn't the problem here.

If the GPU reaches the throttle point, it will fall into a low locked clockspeed.

Only a reboot will correct that state.

chengong wrote:

It isn't the problem here.

I doubt you've actually tried it, so just saying.

In the absence of ANY thermal issues, it can only come down to the dynamic clocks as I stated or that custom BIOS just being outright a load of crap.

Maybe this is too obvious. My 460 used to lock to low gpu clocks (around 400 iirc) after driver reset, restarting fixed it. Not sure if maxwell does this too. Not allowed to post links but if you were to search for 405mhz core you might find a few threads about this.

I had some problems today with GPU downclocking to between 200-600 Mhz in LoL even when I was only getting 30 fps with 144 fps cap. It would go up to 800 Mhz when alt tabbing back to desktop. Thinking it might be the new 353.62 drivers messing with my 980Ti in Win 7. Since I didn't upgrade to Win 10, I'm rolling back to 353.49 and hoping that fixes it

geekl33tgamer wrote:

chengong wrote:

It isn't the problem here.

I doubt you've actually tried it, so just saying.

In the absence of ANY thermal issues, it can only come down to the dynamic clocks as I stated or that custom BIOS just being outright a load of crap.

I can't believe I have to confirm basic stuff like this here. Yes I have tried I know because it's running like 10fps in Witcher 3 when normally it runs 35fps.

If the clock table on the bios is broken then explain why it doesn't happen in normal use? Obviously restarting doesn't do anything to the BIOS. And I can use whatever OC setting I want, so long as I don't fumble around with it too much.

chengong wrote:

geekl33tgamer wrote:

chengong wrote:

It isn't the problem here.

I doubt you've actually tried it, so just saying.

In the absence of ANY thermal issues, it can only come down to the dynamic clocks as I stated or that custom BIOS just being outright a load of crap.

I can't believe I have to confirm basic stuff like this here. Yes I have tried I know because it's running like 10fps in Witcher 3 when normally it runs 35fps.

Well I'm sorry my peasant response was beneath you. You were asked a little further up if it impacted performance and you didn't answer the question. IMO, saying "I know because I can test the performance in like you know? games?" comes across a touch arrogant / sarcastic. I can't help you if I'm not in receipt of all the facts while my crystal ball is out of service.

Hopefully someone else helps you.

For someone wanting help, your tone comes off a bit rude. Legitmate questions were asked due to a lack of initial information.

Reverting to old drivers fixed the problem. Gorgeous 144 fps at ~1100 mhz. The newest Nvidia drivers apparently have tons of problems.

Westbrook348 wrote:

Reverting to old drivers fixed the problem. Gorgeous 144 fps at ~1100 mhz. The newest Nvidia drivers apparently have tons of problems.

That's not an option as he is using Windows 10? There's only 362.30 and 362.62 (I think) released for the RTM version.

Improper overclocking causes these symptoms.

My card usually locks into a non-performance power state after too many hardware faults due to overclocking. Like others have said, a reboot will solve it, but double-check the stability of your overclocking.

The driver will not necessarily throw you a TDR ("Display driver stopped responding") error, either. The overclock can fail silently after X errors.

geekl33tgamer wrote:

You were asked a little further up if it impacted performance and you didn't answer the question. IMO, saying "I know because I can test the performance in like you know? games?" comes across a touch arrogant / sarcastic. I can't help you if I'm not in receipt of all the facts while my crystal ball is out of service.

This, 100%. Most of us aren't idiots, but without all the facts we have to assume that you're not one too and with snarky know-it-all comments you're not winning any favours when asking for help.

1) Make sure the performance is actually suffering and it's not just a reporting issue.
2) Eliminate potential causes of the problem one by one until you find the issue.
3) If step 2 is unsuccessful, RMA your card.

For step 2, the obvious culprits are:

• Custom BIOS
• Driver version - with Windows 10 you may have to actually migrate the card over to a W8 machine to go back to known-working drivers
• Overclock voltages/frequencies
• Cooling issue triggering a safety throttle
• Driver settings (Powertune profiles etc)

Eliminate each of those one at a time in a controlled and methodical manner, and then if you can prove it's none of those things RMA the card because it's probably faulty.

Chrispy_ wrote:

1) Make sure the performance is actually suffering and it's not just a reporting issue.
2) Eliminate potential causes of the problem one by one until you find the issue.
3) If step 2 is unsuccessful, RMA your card.

For step 2, the obvious culprits are:

• Custom BIOS
• Driver version - with Windows 10 you may have to actually migrate the card over to a W8 machine to go back to known-working drivers
• Overclock voltages/frequencies
• Cooling issue triggering a safety throttle
• Driver settings (Powertune profiles etc)

Eliminate each of those one at a time in a controlled and methodical manner, and then if you can prove it's none of those things RMA the card because it's probably faulty.

Like I said, if he's going to do elimination, then first of all he should start with changing or disabling the overclocking.

Meadows wrote:

Like I said, if he's going to do elimination, then first of all he should start with changing or disabling the overclocking.

Step 2 is not an ordered list. Any one of them could be the problem.

My vote is that the problems is the newest Nvidia drivers. Reducing the OC won't solve the underlying problem if that's the case. If you can't roll the drivers back because of Windows 10 then I suggest finding a way to get back to the old version of Windows until MS/Nvidia get all the bugs fixed. If you can't, then just wait it out. Lots of people are having problems with the latest drivers and are in the same boat, so I'm assuming they'll get a hotfix driver out soon.

This is why I won't be a guinea pig by being one of the first to upgrade. Stuff like this always seems to happen.

"I mess with my clocks a lot and eventually the card gets in a bad state."

I'm sorry, but this has been around forever when overclocking GPUs ever since dynamic clocks were introduced. You're running a custom modded BIOS (granted, shouldn't change much) but it implies that you have some experience here. If you've changed clocked enough (and jumped to any sort of unstable state) the GPU will drop into this weird state.

Reboot, move on with life, and continue your OC.

I had this issue for a while. My fix was to enable Kboost in Precission X. That fixed the problem, but, your GPU is going to be going full out all the time with Kboost enabled. Not a problem for me since I use my GPU's for distributed computing projects when I'm not gaming.

Some people don't like the Precission X software. All I can say is I have not had any issues with it so far.

Chrispy_ wrote:

geekl33tgamer wrote:

You were asked a little further up if it impacted performance and you didn't answer the question. IMO, saying "I know because I can test the performance in like you know? games?" comes across a touch arrogant / sarcastic. I can't help you if I'm not in receipt of all the facts while my crystal ball is out of service.

This, 100%. Most of us aren't idiots, but without all the facts we have to assume that you're not one too and with snarky know-it-all comments you're not winning any favours when asking for help.

1) Make sure the performance is actually suffering and it's not just a reporting issue.
2) Eliminate potential causes of the problem one by one until you find the issue.
3) If step 2 is unsuccessful, RMA your card.

For step 2, the obvious culprits are:

• Custom BIOS
• Driver version - with Windows 10 you may have to actually migrate the card over to a W8 machine to go back to known-working drivers
• Overclock voltages/frequencies
• Cooling issue triggering a safety throttle
• Driver settings (Powertune profiles etc)

Eliminate each of those one at a time in a controlled and methodical manner, and then if you can prove it's none of those things RMA the card because it's probably faulty.

I'm sorry it came through like that, but please take a close look and see if you'll also get irritated by the same situation. In my OP post I clearly stated that GPU-Z reports a performance cap reason as "utilization", which obviously means that I am aware of this fact. Knowing that, and still questioning if I actually did stress the GPU implies only one of two possible things:
1. I don't know the meaning of the word "utilization", AND I never looked it up
2. I'm an idiot and didn't think to actually give it utilization even when it clearly says that there's not enough utilization.
Besides, the screenshot I provided shows 100% GPU utilization, AND it shows that there are no thermal issues.

As for the possibility of a detection error, it too, can be seen in the screenshot. Temperature is 33 degrees, unless you think I'm using some kind of a phase-change system, that's clearly not a possible temperature for a 250w GPU running at maximum frequency. TDP is also labeled as 34%, obviously, not running at full speed. Voltage at 0.89v again, obviously not a full load voltage. Unless of course, you believe that GPU-Z read all those numbers wrong, at the same time, and they happen to match up with each other (temperature with tdp and frequency). Or may I suggest, the more likely problem is that there really is something wrong and it's stuck at 405mhz. I tend to think that the latter is much more likely.

I think anyone would be irritated when people constantly ask things that are already made clear in the OP.

Yeah, it's clear the hardware's not being stressed.
It depends what was happening when you took that screenshot:

• If you were getting low framerates in a real game/app and in need of more performance that's a fault.
• If you were running a light app with Vsync on and getting framecapped, that's powertune doing its job, whether correctly or not.
• If you were running Furmark, both AMD and Nvidia detected that in the driver in the past and used a modified power-virus profile, It's possible that profile is triggering a bug in this specific behaviour newest W10 driver that might be a bit rough around the edges....

I'd be inclined to call it a software fault (driver) but you'll just have to be patient and test each thing one at a time. Do you have an old SSD with Windows 7 or 8 install kicking around anywhere that you could use to test a previous driver?

To fix the problem, you need to know what the problem actually is - the 400MHz clock appears to only be a symptom, not a cause. It's also hard to eliminate the overclock as causing the problem if you haven't had this overclock stable before. GPU's are really really complicated and the default "two sliders and some voltage adjustment" is really just the tip of the iceberg, what actually happens when you overclock involves many things that are partly managed by the bios, partly managed by the driver, and partly managed by the OS. Seperating those things out is almost beyond the scope of us mere mortals and way beyond what you should be relying on tools like GPU-z for.

chengong wrote:

I'm sorry it came through like that, but please take a close look and see if you'll also get irritated by the same situation. In my OP post I clearly stated that GPU-Z reports a performance cap reason as "utilization", which obviously means that I am aware of this fact. Knowing that, and still questioning if I actually did stress the GPU implies only one of two possible things:
1. I don't know the meaning of the word "utilization", AND I never looked it up
2. I'm an idiot and didn't think to actually give it utilization even when it clearly says that there's not enough utilization.
Besides, the screenshot I provided shows 100% GPU utilization, AND it shows that there are no thermal issues.

As for the possibility of a detection error, it too, can be seen in the screenshot. Temperature is 33 degrees, unless you think I'm using some kind of a phase-change system, that's clearly not a possible temperature for a 250w GPU running at maximum frequency. TDP is also labeled as 34%, obviously, not running at full speed. Voltage at 0.89v again, obviously not a full load voltage. Unless of course, you believe that GPU-Z read all those numbers wrong, at the same time, and they happen to match up with each other (temperature with tdp and frequency). Or may I suggest, the more likely problem is that there really is something wrong and it's stuck at 405mhz. I tend to think that the latter is much more likely.

I think anyone would be irritated when people constantly ask things that are already made clear in the OP.

Your situation is that you're screwing with clocks and run into a condition that you caused, that's not happening in normal operation, and expect someone to tell you anything other than not to change the clocks aggressively?

Find your sweet spot, reboot if it sticks, and move on. I'm surprised you're being so indignant when you're the one causing the problem.

I want to bring this thread back alive as I am having the same issue as chengong.

Here is my deal;
-I remember having the same throttling issue since about a year ago, however I did not notice it right away. I did not make a proper diagnosis until a few months ago after talking with Nvidia and Asus tech support
-I am running a Asus GTX 770 OC edition
-I am not OCing the GPU
-I have tried a better PSU
-No custom bios
-I am using windows 8
-Have run tests in Unigine valley and 3DMark
-Have changed the preference to "prefer preformance" in the Nvidia control panel

Here is what I have found;
-Version 335.23 allows me to run unthrottled. I have only tried it and the most recent driver.
-When running the newest driver, the GPU clock speed gets throttled to 405Mhz. It will run at "full load", although the temp stays at 32 degrees. The scores in 3DMark and Unigine Valley are significantly lower than they are when using older drivers.

I don't know if this is unique to this specific GPU, or if its because the GPU is about 2 years old now or anything else. I don't really mind using the old drivers but it would be cool if Nvidia/Asus would take a look at this. I am clearly not the only person having the problem, and it is apparently very hard to find a sure answer to fix it.

Any questions? I'll be happy to answer!

I had the same problem, on some occasions my GTX 570 would thermal throttle and go to super low clock speeds and only rebooting would fix it bit if yo go into the NCP and select prefer maximum performance it does fix it. When it throttled games that i would usually run at high would run 20 fps on low and i knew it wasn't the card because it may be a bit older but can run Fallout 4 on high with 60 fps and gta v on high with 30 fps but it really got me mad when it throttled and ran Fallout 4 an 15 fps and gta v at 20 fps. Like seriously i have a C2Q Q9450 and 8 GB of RAM and when it throttled instead of running cs:go at like 115 fps i was running at 35 fps. I mean its playable but for god sake my 2010 Macbook pro can run it better. But the only reason it thermal throttles is because sometimes the fan speed does not increase when im playing a game and reaches 75 degrees C. i turned on a custom fan speed chart on MSi Afterburner and turned on force fan speed update on each period but 1/10 times i launch a game it does not auto update the fan speed.

I have had this exact isue, it's emanates from Nvidia's driver. If you're using version 376.33, try reverting back to 375.86, that driver version doesn't have that problem with low clocks.
P.S. This occurs in my older as well as newer cards. If on a 580, it locks to low-3D clocks, 405/162/810. It also does a similar thing with my 780 Ti and 980 but I can't recall the exact clock speeds.