is my 3080 TUF OC a Golden Sample? (2160 Mhz while gaming)

[Deleted member 193596]

so i have a 3080 TUF OC since around a week now and i am overclocking it every day (one 15 Mhz Bump every day after several hours of 1440p gaming)


i currently sit at +180 Mhz on the core which results in Cold boosts up to 2200 MHz with the stock fan curve.


my average Gaming Clock speed is 2145 MHz and it bumps up to 2160 Mhz.

In very demanding titles that are going to the max 352W non stop the card runs at around 2145 to 2160 Mhz.

btw the boost clock is 165 Mhz above the EVGA FTW3 Ultra

here is the clock speed at full power consumption and utilization 2145 to 2160 Mhz average.

 

[Zach_01]

You “measure” power draw by readings on HWiNFO?
Just curious... Does this sensor report the whole card (GPU+VRAM) or just the GPU?
I think that nVidia is not providing any info about VRAM other than clock. You may have to add another 30-40W for VRAM power draw on top of that 350W for total card power draw, given the fact that GDDR6X is more power “hungry” than GDDR6 which is around 25W max for a 8x1GB (14Gb, 1800MHz) VRAM configuration.

Very impressive clocks though... I was under the impression that 30series AIB cards have limited power draw, worst that the FE ones.

 

[Deleted member 193596]

You “measure” power draw by readings on HWiNFO?
Just curious... Does this sensor report the whole card (GPU+VRAM) or just the GPU?
I think that nVidia is not providing any info about VRAM other than clock. You may have to add another 30-40W for VRAM power draw on top of that 350W for total card power draw, given the fact that GDDR6X is more power “hungry” than GDDR6 which is around 25W max for a 8x1GB (14Gb, 1800MHz) VRAM configuration.

Very impressive clocks though... I was under the impression that 30series AIB cards have limited power draw, worst that the FE ones.

the 350W are board power of course.

in GPU Z the GPU pulls around 200W and the board power is 352W.

the 110% Powerlimit are the 352W.

 

[Zach_01]

Only 200W for GPU and the rest is 150W... it’s hard to believe.
May I see what short of sensors GPU-Z is reporting for the card?
I’m interested in such info.

 

[Deleted member 193596]

Only 200W for GPU and the rest is 150W... it’s hard to believe.
May I see what short of sensors GPU-Z is reporting for the card?
I’m interested in such info.

that's completely normal for a 3080.

the memory alone pulls 70W

 

[Zach_01]

Screenshot of GPU-Z sensors?

 

[Deleted member 193596]

When idle? That’s normal. During 3D load how much is it...

Screenshot of GPU-Z sensors?

see post above..

 

[Zach_01]

Ok, I saw it now

 

[Deleted member 193596]

Ok, I saw it now

that's why the memory on the FE runs at over 100°C.. it pulls as much as a whole 1050 Ti

 

[delshay]

so i have a 3080 TUF OC since around a week now and i am overclocking it every day (one 15 Mhz Bump every day after several hours of 1440p gaming)


i currently sit at +180 Mhz on the core which results in Cold boosts up to 2200 MHz with the stock fan curve.


my average Gaming Clock speed is 2145 MHz and it bumps up to 2160 Mhz.

In very demanding titles that are going to the max 352W non stop the card runs at around 2145 to 2160 Mhz.

View attachment 171612

btw the boost clock is 165 Mhz above the EVGA FTW3 Ultra View attachment 171613

here is the clock speed at full power consumption and utilization 2145 to 2160 Mhz average.View attachment 171615

Let's see your score with Superposition 1080P Extreme. ..Upload screenshot

Download Here https://benchmark.unigine.com/superposition

 

[mrthanhnguyen]

Run time spy or port royal and see if it can hold the clock. If it holds, its very good.

 

[wolf]

Could be a golden sample, but I find my with 3080 TUF it really depends on what you have loaded the card with.

Could you spin up Quake RTX ? (first few levels are free), and a few heat soaked runs of timespy extreme? Those are the brutaliser tests. I have a feeling that after 10+ minutes of Quake RTX the clock will be a fair bit lower, but I could be wrong, would definitely show how much of a golden sample it is.

 

[xkm1948]

RTX games is the real test. Fire up Metro exodus and play for a solid hour and report your clock

 

[wolf]

RTX games is the real test. Fire up Metro exodus and play for a solid hour and report your clock

Absolutely and more so even with non-RTX games, the older and 'lighter' they are, the higher the card can and will clock. I don't play either of the games he's shown screenshots for, but to my eyes it's wow and modern warfare?

 

[Deleted member 193596]

Absolutely and more so even with non-RTX games, the older and 'lighter' they are, the higher the card can and will clock. I don't play either of the games he's shown screenshots for, but to my eyes it's wow and modern warfare?

i played battlefield V, Hitman 2, R6 Siege, NFS Heat and crysis remastered now.

in all games the clock speed hovers between 2125 and 2160 Mhz. except in Siege with everything maxed out..it runs at over 300FPS and the card drops down to 2085-2100 Mhz

 

[mrthanhnguyen]

My 3090 hit 2160mhz in bf5 but in time spy and port royal, it drops to stock clock. Cant hold it.

 

[delshay]

My 3090 hit 2160mhz in bf5 but in time spy and port royal, it drops to stock clock. Cant hold it.

You & I ask the OP to do an benchmark test & has done neither test sofar. OP needs to show results, then we can judge if it's a golden chip.

 

[Deleted member 193596]

You & I ask the OP to do an benchmark test & has done neither test sofar. OP needs to show results, then we can judge if it's a golden chip.

ever heard of time zones?
i posted this at midnight, went to bed and i have a job.

i am home since a few hours and only replied in my lunch break.

here is a random time spy run

I scored 18 142 in Time Spy

Intel Core i9-10850K Processor, NVIDIA GeForce RTX 3080 x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

 

[P4-630]

ever heard of time zones?
i posted this at midnight, went to bed and i have a job.

i am home since a few hours and only replied in my lunch break.

here is a random time spy run

I scored 18 142 in Time Spy

Intel Core i9-10850K Processor, NVIDIA GeForce RTX 3080 x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

When running this benchmark, run GPU-Z sensors in the background and post a screenshot of it afterwards a benchmark.
The GPU won't stay at maximum boost, you can try other benchmarks as well like superposition, heaven, valley or other.

 

[mrthanhnguyen]

Only 1982mhz in time spy. Not that impressive.

 

[wolf]

here is a random time spy run

I scored 18 142 in Time Spy

Intel Core i9-10850K Processor, NVIDIA GeForce RTX 3080 x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

As said above - Average clock frequency 1,982 MHz - which aint bad but I don't think it's golden sample territory. Can you heat soak it and run Timespy Extreme specifically with your best clocks?

 

[nguyen]

Ah why so harsh against OP, the fact that his 3080 can reach 2160Mhz means that the chip reach very high clocks.
Golden sample or good ASIC score means that the chip can reach high clocks, these chip tend to be leaky, meaning high power consumption for a given freq/voltage. These chip will do wonders when the PL is removed.
Vice versa a chip with low ASIC score tend to be less leaky for a given freq/voltage, so it might get higher Timespy score when locked to the same PL as high ASIC chip, but can't reach clocks above 2050mhz under no circumstance.
TL;DR OP has to shunt mod his card in order to get the most out of his 3080

 

[wolf]

Ah why so harsh against OP, the fact that his 3080 can reach 2160Mhz means that the chip reach very high clocks.

Definitely not trying to be harsh, just trying to get a couple of things out of it, for one an apples to applies comparison against my own RTX3080 TUF.

The chip definitely can touch and potentially hold some very high clocks, but the 3080 is an interesting beast in the sense that, well, it is SUCH a beast and it takes a lot to load it to the maximum extent possible. Power draw / power limiting / holding boost clocks is wildly variable depending on, among other things, the workload. To really push into seeing what the GPU is capable of in terms of sustained boost clocks, you need to saturate it with the heaviest workloads imaginable. 4K with all the trimmings is one such avenue, and why I also recommend Quake RTX as it's fully path traced and will bring any RTX capable card to it's knees sustained clocks/power draw wise. It's more nuanced than any GPU I've ever owned before in this way and that for example two different games could show 99% GPU utilisation, but one pull 270w and another pull 350w of TDP when settings are a locked max clock/voltage, which is a bit different to OP using the core clock offset slider.

But at the end of the day I can appreciate that if these are the games OP plays, and he is getting 2100+ in those games, I mean that ain't bad by any means. My methodology is more to test and set profiles around my worst case scenario, rather than best case scenario. It does make me wonder however about my example of 270w vs 350w, if I take the game that only wants 270w from the card, I can probably tweak a profile with much higher clocks just for that game.

 

[nguyen]

Definitely not trying to be harsh, just trying to get a couple of things out of it, for one an apples to applies comparison against my own RTX3080 TUF.

The chip definitely can touch and potentially hold some very high clocks, but the 3080 is an interesting beast in the sense that, well, it is SUCH a beast and it takes a lot to load it to the maximum extent possible. Power draw / power limiting / holding boost clocks is wildly variable depending on, among other things, the workload. To really push into seeing that the GPU is capable of in terms of sustained boost clocks, you need to saturate it with the heaviest workloads imaginable. 4K with all the trimmings is one such avenue, and why I recommend Quake RTX as it's fully path traced and will bring any RTX capable card to it's knees sustained clocks/power draw wise. It's more nuanced than any GPU I've ever owned before in this way and that for example two different games could show 99% GPU utilisation, but one pull 270w and another pull 350w of TDP when settings a locked max clock/voltage, which is a bit different to OP using the core clock offset slider.

But at the end of the day I can appreciate that if these are the games OP plays, and he is getting 2100+ in those games, I mean that ain't bad by any means. My methodology is more to test and set profiles around my worst case scenario, rather than best case scenario. It does make me wonder however about my example of 270w vs 350w, if I take the game that only wants 270w from the card, I can probably tweak a profile with much higher clocks just for that game.

Actually the heaviest workload is not meant to test the ASIC of the card, it only test how far PL allows you to go . Meaning as you increase the PL, your scores improve accordingly.

OP's 3080 chip is definitely meant to go into the Strix OC version, which much higher PL so that it can stretch its legs better.

Think of it like this, for an average 3080, a 400W PL is enough for its maximum clock of 2050-2070mhz, however OP's can reach 2160mhz which requires 500W PL to sustain that clocks in all type of workload.

Nvidia was really clever to use Power Limit as the way to standardize performance, with Maxwell era and before, you could have chips that clock 100-200mhz apart, making performance between models wildly different. Vendors (EVGA specifically) was selling higher ASIC chip for way more money.
Now with a tight PL in place, higher or lower ASIC make little difference, they would generally end up around the same clocks under the same PL.

 

[wolf]

Actually the heaviest workload is not meant to test the ASIC of the card, it only test how far PL allows you to go.

And this is more the angle/point I'm making anyway, given RTX3080's are held back by far the most by the PL, we try and maximise what we can get our of our silicone lottery 'win' within that ~375w PL, the lighter the workload the higher his clocks will be which can potentially give a false impression of an amazing ASIC quality.

The ASIC quality might be amazing, I'm not the judge of that, but in a normalised workload between our individual cards, restricted by an identical PL, It doesn't seem like it will hold clocks much higher. Golden sample? maybe, I suppose that's not my call to make, you seem pretty convinced. Unless he's willing to do hard mods like a shunt mod, we might not ever truly find out because the card can't draw an unlimited amount of power to see.