GPU clocks going higher than manufacturer's default

[Skeptical]

Hello,


I recently acquired a bnib GTX 1050Ti Cerberus 4GB that was barely used. After some time and messing around with it I realized it's temperature was going quite high whenever the gpu reached 100%, and it's cap temp is set to 83C(which I thought it was weird because it's usually 90). Strangely enough, it easiliy reaches that temp, I just need to run some heavy application. So I used gpu-z to check It's default clocks which are shown below:

Gpu-Z


As you can see these numbers also doesn't match the manufacturer's website and software (GPU Tweak II)



So whenever I run Haven I get over 1600Mhz. What is going on here? Did the previous owned flashed a modded bios? The void seal is intact, I'm pretty sure the vga was never opened. Whenever I run msi afterburner or any other software the GPU clock goes way above the limit, reaching even 1700Mhz sometimes. I will post some pictures to show exactly what's going on. Is there a way I can fix it? I'm not much of a gpu expert! Thanks

 

[oxrufiioxo]

This is normal behavior for a modern gpu since pascal. They all boost now based on a variety of factors voltage, Temps, power.

 

[Skeptical]

As you can see, the graphics shows 1900+ in Haven, and the gpu clock goes up to 1709 even though the "boost" clock is set to 1455... As you can see it easily reaches 83C which is the temp limit, when searching why the temperature was that high I found out that this gpu is "boosting" itself more than 300mhz

This is normal behavior for a modern gpu since pascal. They all boost now based on a variety of factors voltage, Temps, power.

Thanks for the reply, so the vga was not flashed? Can you look at my second post and confirm this behavior is standard? As you can see I am using all the default settings in the software (Max set to 1455) but it goes above 1700mhz anyways. Maybe that's what's making my gpu so hot? If so is there a way to lower it to the standard value? I'm not comfortable with my gpu reaching over 80C that easily and with fans at 100% all the time

 

[oxrufiioxo]

Yeah, looks normal to me. The heatsink on that card looks pretty dinky even though it's got two fans. You can try to lower the power limit to like 90% or downclock the core. You could also repaste it and see if that's the culprit.

 

[Fouquin]

So whenever I run Haven I get over 1600Mhz. What is going on here? Did the previous owned flashed a modded bios? The void seal is intact, I'm pretty sure the vga was never opened. Whenever I run msi afterburner or any other software the GPU clock goes way above the limit, reaching even 1700Mhz sometimes.

The card is boosting over 1455 because it's operating within GPU Boost thresholds.

The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation

www.anandtech.com

GPU Boost – Nvidia’s Self Boosting Algorithm Explained

Graphics Card technologies have progressed by leaps and bounds over the last few generations with each generation bringing a substantial improvement not

appuals.com

An easy solution, is to reduce your power target by 10-20%. If you have temperature target control, set it lower than 83C if you're uncomfortable with that temperature. The 83C limit has been in place on the majority of NVIDIA's cards since 2012, but you can set it higher or lower on most cards. in reality there is nothing at all wrong with the card running at 83C, it's well far below where you would want to worry about degradation or damage (95-110C). Your card is operating as expected.

 

[looniam]

yeah lower the PL is quick but repasting the card, or at least increasing the fans in AB (get rid if trixx!) and under-volting (plenty of online guides) would be best.

really you got a "better chip" if its boosting higher on "default" voltage; why you really wont lose performance under-volting . .

 

[DeathtoGnomes]

Has Nvidia called wanting their binned chip back?

 

[qubit]

Yup, these things run very hot. Most cards can't be touched for several minutes after running a game and must be allowed to cool down first.

 

[agent_x007]

Alternatively, you can "lock" your GPU to a point on V/F curve which is fine for cooling to cope with (here's MSI's afterburner example, you lock a point with first selecting it, then pressing "Ctrl + L" shortcut, and lastly confirming your choise on main screen of Afterburner).

However, if you don't want to see it going higher than that point (and still have 2D mode), I recommended doing something like this :

^This lowers max. voltage GPU can use to selected value on x-axis (since higher points have the same frequency on y-axis).
Important note : You have to adjust this based on cooling and stability (higher frequency with lower voltage = less stability).
Start with default curve (in above case, 1400MHz-ish at 762mV [0,762V]), and keep increasing frequency until you find instability.

 

[eidairaman1]

Yeah, looks normal to me. The heatsink on that card looks pretty dinky even though it's got two fans. You can try to lower the power limit to like 90% or downclock the core. You could also repaste it and see if that's the culprit.

Yup, these things run very hot. Most cards can't be touched for several minutes after running a game and must be allowed to cool down first.

This cards pcb is basically what is called an Asus Dual/Mining. This name came about when nv was trying to force companies to sell nv cards under top names such as rog and amd cards under lower names such as Ares.

Asus Dual/Mining/Cerebus is bottom tier

Asus TUF is Mid Tier

Asus ROG is Top Tier.

Do not change thermal compound on it for at least a year till warranty expires.

 

[Skeptical]

The card is boosting over 1455 because it's operating within GPU Boost thresholds.

An easy solution, is to reduce your power target by 10-20%. If you have temperature target control, set it lower than 83C if you're uncomfortable with that temperature. The 83C limit has been in place on the majority of NVIDIA's cards since 2012, but you can set it higher or lower on most cards. in reality there is nothing at all wrong with the card running at 83C, it's well far below where you would want to worry about degradation or damage (95-110C). Your card is operating as expected.

That is a relief, tysm. As a former radeon owner I had no idea nv cards had this feature at all and I was almost panicking!

@agent_x007 Thanks for the guide. I ended trying this curve realizing the gpu peaks at 1700Mhz@1230mV. Then I made my curve like this:

900 Voltage was the minimum I could run at max "boost frequency" without instabilities and my temperature went down drastically. I hardly hit 75C now on very demanding games and that is very conforting. Thanks for the help everyone, I truly appreciate it.

This cards pcb is basically what is called an Asus Dual/Mining. This name came about when nv was trying to force companies to sell nv cards under top names such as rog and amd cards under lower names such as Ares.

Asus Dual/Mining/Cerebus is bottom tier

Asus TUF is Mid Tier

Asus ROG is Top Tier.

Do not change thermal compound on it for at least a year till warranty expires.

Thanks for the reply. Huh, so this card is a bottom tier 1050Ti? I always regarded ASUS as a top tier manufacturer. I was reading it's specifications and I thought It to be a very resistent piece of hardware, supposedly 15x more tested than the industry's standard. I guess I was wrong then, tough luck for me.

It is also weird that I have to go out of my way as an user and undervolt the card by myself in order to make it run cooler and not at 100 fan speed. And all this set by factory defaults.

It is all good now : 44C idle 75C max load, but I had to undervolt it, common users wouldn't go such lenghts, indeed even although it is a very powerful card (at least compared to my old radeon) , this cerberus seems to be is a bad designed product.

 

[maxx2575]

my 1050Ti boost to 1923 when at load.

 

[Skeptical]

my 1050Ti boost to 1923 when at load.

If mine if Cerberus, yours is Behemoth. It was supposed to run at 1500mhz default...

 

[R-T-B]

Thanks for the reply. Huh, so this card is a bottom tier 1050Ti? I always regarded ASUS as a top tier manufacturer. I was reading it's specifications and I thought It to be a very resistent piece of hardware, supposedly 15x more tested than the industry's standard. I guess I was wrong then, tough luck for me.

I'm sure it's built fine it's more that its performance is lackluster vs other higher performing chips from the same generation (1080 etc). But you probably knew that when you bought it.

 

[looniam]

Thanks for the reply. Huh, so this card is a bottom tier 1050Ti? I always regarded ASUS as a top tier manufacturer. I was reading it's specifications and I thought It to be a very resistent piece of hardware, supposedly 15x more tested than the industry's standard. I guess I was wrong then, tough luck for me.

i don't think eidairaman1 was talking manufacturer so much as the SKUs.

i'll use evga as an example being a fanboi and all

their black cards are vanilla/reference design pcb (power connectors, phases, ect.) with nvidia's spec base/boost clock speeds.
their super clocked are the exact same thing but the chip are binned for an (usually ~100mhz) above nvidia's spec'd boost clock.
the FTW3 (ultra) are custom pcbs, more vrms, power connectors, (yadda yadda) and binned chips ~150Mhz boost increase.
kingpins are uber custom pcbs . . . and binned chips ~200Mhz+ boost speed increase.

its sorta like i mentioned before w/golden chip; what you have would have been binned in a ROG Strix card if they had made one for the 1050ti.

and i did pull those evga boost increase speeds out of the air and are an example only.

oh maybe i read you wrong but don't get hung up on a specific number for boost clocks. they ought to have an "*at least" next to it. any kepler, maxwell or now ampere card i had was ~100Mhz faster out of the box.
except one dog of a 780ti classy

 

[eidairaman1]

i don't think eidairaman1 was talking manufacturer so much as the SKUs.

i'll use evga as an example being a fanboi and all

their black cards are vanilla/reference design pcb (power connectors, phases, ect.) with nvidia's spec base/boost clock speeds.
their super clocked are the exact same thing but the chip are binned for an (usually ~100mhz) above nvidia's spec'd boost clock.
the FTW3 (ultra) are custom pcbs, more vrms, power connectors, (yadda yadda) and binned chips ~150Mhz boost increase.
kingpins are uber custom pcbs . . . and binned chips ~200Mhz+ boost speed increase.

its sorta like i mentioned before w/golden chip; what you have would have been binned in a ROG Strix card if they had made one for the 1050ti.

and i did pull those evga boost increase speeds out of the air and are an example only.

oh maybe i read you wrong but don't get hung up on a specific number for boost clocks. they ought to have an "*at least" next to it. any kepler, maxwell or now ampere card i had was ~100Mhz faster out of the box.
except one dog of a 780ti classy

Yeah i was only talking about product lines with in the Brand.

 

[agent_x007]

It is also weird that I have to go out of my way as an user and undervolt the card by myself in order to make it run cooler and not at 100 fan speed. And all this set by factory defaults.

It is all good now : 44C idle 75C max load, but I had to undervolt it, common users wouldn't go such lenghts, indeed even although it is a very powerful card (at least compared to my old radeon) , this cerberus seems to be is a bad designed product.

I'm glad this worked for you.

Manufacturers simply max out voltage where they can so that they can use more leaky dies on mass production products. It's simply cost effective option vs. throwing them away as trash.
They could do what you just did in factory, but that would mean second or third round of stability checking, and that's not something they want to waste factory time on.
On the other hand... performance matters. They can't just pick lower voltage point (as "max boost") and call it a day, because that reduces performance way too much (sometimes even below stock/boost in NV spec).

PS. Voltage curve I provided is from my Titan Xp card.
Which by default uses ~1.03V on vGPU on stock cooler as max. boost value. My card has NV ref. cooling, so keeping frequency stable at that kind of voltage is simply impossible for it.
In my case, I lower my vGPU so much to get acceptable performance without jet engine (and/or 83C temp/power throttling on top of that), kicking-in. It works grerat for me.

 

[eidairaman1]

I'm glad this worked for you.

Manufacturers simply max out voltage where they can so that they can use more leaky dies on mass production products. It's simply cost effective option vs. throwing them away as trash.
They could do what you just did in factory, but that would mean second or third round of stability checking, and that's not something they want to waste factory time on.
On the other hand... performance matters. They can't just pick lower voltage point (as "max boost") and call it a day, because that reduces performance way too much (sometimes even below stock/boost in NV spec).

PS. Voltage curve I provided is from my Titan Xp card.
Which by default uses ~1.03V on vGPU on stock cooler as max. boost value. My card has NV ref. cooling, so keeping frequency stable at that kind of voltage is simply impossible for it.
In my case, I lower my vGPU so much to get acceptable performance without jet engine (and/or 83C temp/power throttling on top of that), kicking-in. It works grerat for me.

1 size fits all approach with volts

 

[R0H1T]

This is normal behavior for a modern gpu since pascal.

Maxwell actually, my boost clocks are 25(?) MHz higher than advertised.

 

[freeagent]

On my Maxwell card the default clock is supposed to be 1230 I think, but it always runs at 1410. I can’t remember what the default clock is on my Ti but it will boost to 1980 on its own, and if I push it will do 2115-2130.

 

[agent_x007]

Maxwell actually, my boost clocks are 25(?) MHz higher than advertised.

Actually, it's since Kepler (GTX 680 does this if power/TDP is fine), later Keplers get more heavy on temp instead of power.
GTX 780 Ti : 876MHz default, 930-ish boost. It goes to 1020MHz as max. boost in first few seconds (drops fast based on load vs. TDP/temp limits from there).

 

[r9]

Hello,


I recently acquired a bnib GTX 1050Ti Cerberus 4GB that was barely used. After some time and messing around with it I realized it's temperature was going quite high whenever the gpu reached 100%, and it's cap temp is set to 83C(which I thought it was weird because it's usually 90). Strangely enough, it easiliy reaches that temp, I just need to run some heavy application. So I used gpu-z to check It's default clocks which are shown below:

Gpu-Z
View attachment 246547

As you can see these numbers also doesn't match the manufacturer's website and software (GPU Tweak II)

View attachment 246548

So whenever I run Haven I get over 1600Mhz. What is going on here? Did the previous owned flashed a modded bios? The void seal is intact, I'm pretty sure the vga was never opened. Whenever I run msi afterburner or any other software the GPU clock goes way above the limit, reaching even 1700Mhz sometimes. I will post some pictures to show exactly what's going on. Is there a way I can fix it? I'm not much of a gpu expert! Thanks

Don't be skeptical.
It's normal behavior.

 

[oxrufiioxo]

Maxwell actually, my boost clocks are 25(?) MHz higher than advertised.

My 2x 680, 2x 970, and 980 ti all ran pretty close to the rated boost clock my 2x 1080s, 2x 1080ti, and Titan Xp all ran over 200mhz over the rated boost clocks. So yeah technically they all boost but at least from what I've observed it wasn't till pascal that cards boosted 200mhz+ over what they were rated at.

 

[maxx2575]

If mine if Cerberus, yours is Behemoth. It was supposed to run at 1500mhz default...

well it is also a zotac card

ZOTAC GTX 1050 Ti OC Edition Specs

NVIDIA GP107, 1506 MHz, 768 Cores, 48 TMUs, 32 ROPs, 4096 MB GDDR5, 1752 MHz, 128 bit

www.techpowerup.com

 

[vmarv]

This is normal behavior for a modern gpu since pascal. They all boost now based on a variety of factors voltage, Temps, power.

The card is boosting over 1455 because it's operating within GPU Boost thresholds.

The NVIDIA GeForce GTX 1080 & GTX 1070 Founders Editions Review: Kicking Off the FinFET Generation

www.anandtech.com

GPU Boost – Nvidia’s Self Boosting Algorithm Explained

Graphics Card technologies have progressed by leaps and bounds over the last few generations with each generation bringing a substantial improvement not

appuals.com

Thanks God I've found this thread. I used AMD since 2016 and recently I bought a used GTX 1070 TI: looking at the boost speeds in benchmarks I thought that the owner had modified the bios. So I was looking for a bios editor for Pascal to see the values, but I discovered that NVIDIA encrypted the bios of this gen cards. So, I said wtf is going on here, how the hell did he modified the bios?
And instead everything was as supposed to be.