RTX3090 not performing much better on water?

Grim5

Grim5

Grim5

Soldato
Joined
6 Feb 2019
Posts
21,269
So this is confusing for me.

Temps in the loop are now great.

When my 3090 strix was on air, I run the card with a 420w power limit, it would reach 73c at 80% fan speed. I was able to get 21200 graphics score in Timespy, it would run at 2050mhz

Now the card is under water, I've pushed it to 480w power limit, it maxes out at 51c. Timespy score has improved to just 21500.

When I look at the Timespy benchmark thread, others with their RTX3090 on water are getting clock speeds between 2100 and 2200mhz, mine under water won't go over 2070mhz even though I've given it 480w and just 51c tempreture.
 
Last edited:
Ok I think my card is just not as good as the others, low temps won't help it.

The reason I only see 2070mhz in Timespy is because I'm limited to how much clock offset I can add to it, if I go above +80mhz on the core, Timespy will just keep crashing half way - this happened when the card was on air and still happens on water.

But in Port Royal I can do +150mhz easy, so the card runs at 2150mhz in that benchmark and gives me a good score improvement - I may be able to go even higher, thats just the most I've tried so far
 
Pushing the power limit up should get you some more boost clocks if you're staying under 50. iirc that's about the point that boost throttling starts to take effect these days, have you checked what gpu-z reports as the performance cap reason?

When you mention others getting faster clock speeds are you comparing like for like on the exact card model? If so then it's just down to the silicone lottery, if not then there's a whole mess of factors which could influence that max number.

Realistically, other than epeen... do you really need to push the clock speed on a 3090 by such a tiny margin? I doubt it would gain you more than a couple of FPS max and unless you're gaming at 8k you're probably not going to make that card struggle too much!
 
Is it on the power limit the whole time during the benchmark?

There can still be value in tuning different points of the curve. My 3090 FE does +155 rock solid in everything, yet that means 1890 at 912 mV, but I've been slowly testing different points on the curve, working from 850 mV up and making sure each is super stable. I played for about 5 hours yesterday and 1980 at 912 mV was stable. So a pretty decent jump.

Port Royal seems a poor stability test. I think I got to +225 and it was all good. Not stable in games though. I've found that Fenyx Immortals Rising has been good at finding random issues.

But yeah, maybe try manually tuning the curve. Or just play games and enjoy :D
 
SirConfused said:
Pushing the power limit up should get you some more boost clocks if you're staying under 50. iirc that's about the point that boost throttling starts to take effect these days, have you checked what gpu-z reports as the performance cap reason?

When you mention others getting faster clock speeds are you comparing like for like on the exact card model? If so then it's just down to the silicone lottery, if not then there's a whole mess of factors which could influence that max number.

Realistically, other than epeen... do you really need to push the clock speed on a 3090 by such a tiny margin? I doubt it would gain you more than a couple of FPS max and unless you're gaming at 8k you're probably not going to make that card struggle too much!
Click to expand...

Ah didnt know gpu-z had that function. While running timespy and port royal, it reports the performance cap is "Power"

TT158 said:
Is it on the power limit the whole time during the benchmark?
Click to expand...

In port royal Gpuz reports performance limit is Power all the time. In Timespy, it reports Voltage in the first test, Power in the 2nd and 3rd, Idle in the 4th. The 1st and 4th tests are CPU limited so thats why, 2nd and 3rd are GPU limited. Generally in Port Royal and the 2nd and 3rd tests in Timespy the card will be pulling between 440w and 460w with few deviations from that. The clock frequency graph in the results page looks very flat
 
Last edited:
Ampere just don't overclock very well and even if you win the silicon lottery and can push clocks really high it doesn't translate into hardly any fps in games.
 
you nothing earth shattering tbh for me either but tbf does hold boost clock at 2055mhz consistently

that and rig is a lot quieter and looks winrar extractable
 
3t3P said:
you nothing earth shattering tbh for me either but tbf does hold boost clock at 2055mhz consistently

that and rig is a lot quieter and looks winrar extractable
Click to expand...

Yeah that's where I'm seeing the most benefit from water, it hasn't improved by maximum clock speed sadly, but it's made the clocks a lot more stable (on air the card would reach up to 2100mhz at the start of the benchmark when the card is under 50c, but then also drops as low as 1930mhz on the same benchmark run once the card heated up to mid 70s) -and now it never goes under 2000mhz and I'd say 98% of the time it's at 2050mhz or higher.

I suppose it's less impressive to me because the Strix was a damn good air cooled card to begin with, I'm even more impressed at how well the card cools after seeing what expensive water does - air cooling has come a long way and it nearly makes water cooling worthless. If I had a cheaper slower card (I know some 3090's will run at low 1800s 95% of the time) then the performance jump would have been much bigger.

But other than the GPU performance, there are other benefits - noise is a major one, I've dropped at least 10db and it's now much easier on the ears and not distracting at all (I dont use headphones). Measuring 1 inch from the side panel, I get 49db which is great, when air cooled and the gpu was hitting high temps I'd see 60 to 63db
 
Last edited:
Watercooling was mainly to avoid 350W+ heat being dumped inside the case and the noisy fans.
As side effect, the boost lasts longer and the sustained boost is slightly higher.
Overclocking would benefit more than in my case, but then many other variants gets into play: power limit and chip limitation. Temperature shouldn't be a problem.
 
I noticed MSi has new monitors to tell you what is hitting your performance. For me it's always voltage - when i run a benchmark nearly all the time the voltage limit is capping the performance

How do I raise the voltage??
 
Hwinfo would tell if the limit is thermal or voltage.
You can raise to a point using supplied software or afterburn. Problem is, the limit how much the cables as the slot PCI-E can supply.
A dual 8-pin card wouldn't allow much over the 350W, even if you flash a BIOS of a 400+W card. Limitation of the 8-pin.
 
Drumroll said:
Hwinfo would tell if the limit is thermal or voltage.
You can raise to a point using supplied software or afterburn. Problem is, the limit how much the cables as the slot PCI-E can supply.
A dual 8-pin card wouldn't allow much over the 350W, even if you flash a BIOS of a 400+W card. Limitation of the 8-pin.
Click to expand...

Hwinfo says voltage limited. I've got MSI afterburner, there is no voltage control - raising the voltage slider does not actually change the voltage
 
Grim5 said:
I noticed MSi has new monitors to tell you what is hitting your performance. For me it's always voltage - when i run a benchmark nearly all the time the voltage limit is capping the performance

How do I raise the voltage??
Click to expand...

Other than maxing the voltage slider in Afterburner or some such tool, you can't. I think the truly hardcore have special bios or other mods, but you're at the mercy of nvidia / AMD otherwise.
 
You could play with the voltage curve and clock speeds and force it to max volts to give you a specific clock speed. I'm pretty sure Optimum tech did a video on it and how to set it etc.

I tweaked mine slightly so it would hold 2000mhz at lower voltages since i run my gpu really cool below 40c.
 
WCing cards now is never really about the performance like say a cpu is. That’s mainly because temperature isn’t the limiting factor.

You’re only talking a few steps back of the nvidia boost curve 50-100mhz max between a watercooled and air cooled card. The main benefit is noise. And cooler card/components.

Years back you could mod your own bios for extra voltage and thus extra performance which would require liquid cooling to even sustain. Today there isn’t really such a thing as nvidia locked the bios to the average user.

980Ti for example you could put a 1.3v bios on. Stock was 1.18v to 1.23v. Today’s cards your talking the difference between 1.2v and 1.21v maximum adjustment.
 
tyler_jrb said:
WCing cards now is never really about the performance like say a cpu is. That’s mainly because temperature isn’t the limiting factor.

You’re only talking a few steps back of the nvidia boost curve 50-100mhz max between a watercooled and air cooled card. The main benefit is noise. And cooler card/components.

Years back you could mod your own bios for extra voltage and thus extra performance which would require liquid cooling to even sustain. Today there isn’t really such a thing as nvidia locked the bios to the average user.

980Ti for example you could put a 1.3v bios on. Stock was 1.18v to 1.23v. Today’s cards your talking the difference between 1.2v and 1.21v maximum adjustment.
Click to expand...
This. Unless your case's airflow us "cooking" the GPU, or fans aren't spinning fast enough, no much room.
Mine, which is limited by the 2 X 8-pin, could boost slightly higher and sustain a higher boost for longer. If I had the fans at 80% +, the difference would be very small. But the noise...
Another benefit of watercooling is that the other components in the case or more specifically next to the GPU (even when the airflow is good), would benefit from most of the heat being directed far from them.
Happened in the past with a Vega GPU where my M.2 was thermal throttling.
And even watercooling, I can feel the heating around the backplate. Can only imagine yours at 450W +.
 
Jay85 said:
You could play with the voltage curve and clock speeds and force it to max volts to give you a specific clock speed. I'm pretty sure Optimum tech did a video on it and how to set it etc.

I tweaked mine slightly so it would hold 2000mhz at lower voltages since i run my gpu really cool below 40c.
Click to expand...

I think this only works if you undervolt, but I want to overvolt

I did try using ctrl+l in afterburner to increase the voltage, but the card didn't actually use a higher voltage regardless of where you set the voltage cap, only reducing voltage has any affect
 
Out of interest - if you keep the power limit at 420w what do you get - could be an issue at the higher power levels holding you back - long shot, but might be worth a go?
 
A good way to test max clocks is to run a game, I went with doom eternal @1440 then cap fps to say 50fps to limit GPU utilisation to around 30% then see how high you can push clocks before it crashes.

If you can 2170 is about average with 2200 good and 2240+ golden atleast on the 3080.
 
Joxeon said:
A good way to test max clocks is to run a game, I went with doom eternal @1440 then cap fps to say 50fps to limit GPU utilisation to around 30% then see how high you can push clocks before it crashes.

If you can 2170 is about average with 2200 good and 2240+ golden atleast on the 3080.
Click to expand...

I tried this with warzone, GPU util at 30% resulted in 1700mhz

GPU util at 50% resulted in 2160mhz (this is with +150mhz overclock applied and gpu using 250w) - any higher = crash

GPU util at 100% resulted in 2070mhz (this is with +100mhz overclock applied and gpu using 440w) - any higher = crash
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom