RTX 3090 Boost Clock

[Elric82]

3090FE stock out the box see it peak around 1995 Mhz then drop to around 1965 Mhz as it warms up. Temp wise it tops out around 68 degrees or so with fan profile at 43% so very quiet FE cooler.

Day to day, I have it at 2100 Mhz and +800Mhz on memory profile with a slight undervolt using the voltage curve. Temp wise is the same as above amusingly as its set to use less voltage then at stock. Card can peek at 2205 Mhz and I have a saved at 2175 Mhz but see no difference really between 2100mhz and 2175mhz.

Will hold this in everything bar Cyberpunk 2077 with RT on and benchmarks. This is partly as with CP2077 RT / benchmarks, game will smash into the 400w power limit. So there using a undervolt / overclock works best I find.



Normal of all of Ampere, 15mhz increments that is. Though some other people seem to mention in between numbers but not seen that on Ampere GPUs I have used and never seen in between steps. Can select in between steps in likes of MSI AB voltage curve, but it rounds up or down in my experience.

Could you show us your curve please? (that sounds so weird out of context)

 

[Radox-0]

Could you show us your curve please? (that sounds so weird out of context)

Sure below is what I use day to day bar heavy RT games, basically CP which hits power limit:

Spoiler

Its under water now so fan bit is moot. For 2100Mhz @ 1.0v its actually 2115 MHz on the curve. As it heats up it will drop and settle at the 2100Mhz around 40 degrees. normally at stock card chugged down around 1.075v

 

[oldestgregg]

I'm sure there's something up with mine. I have the TUF non OC version and the fans are up to 75% to keep it at 70 degrees. I've undervolted it to 885mv and it boosts anywhere up to 1950-ish. It's weirdly game dependant though. For instance, I'm playing Resi evil now and while the tomb raider bench will max those clocks, Resi hovers around the 1770mhz it's advertised at. Cyberpunk seemed to run at the highest clock speed available so could it be Resident evil?

Mine does seem overly warm though having to run it at 75% fans speed to keep 70 degrees.

I've actually just realised that I upgraded to the 500X Corsair case last week and the fan profile is pretty tame for the six fans. Set the case fans to kick to 50% when the CPU hits 50 degrees so it should help when gaming.

 

[Elric82]

Oh yeah underw water is another story

i'd still love to keep my GPU cooler and especially more silent when gaming
My temps are usually around 65 to 70 on core, around 82 to 90 on memory depending on game.

those are good temps but i would love to have less noise the fans are almost at max speed to keep the gpu cool...


I was interested in undervolting but if I do so would I lose a lot of FPS knowing i'm running +100 on core and clocking usually around 2070 to 2100mhz in game?

I've tried it for 1mv for 2100 but games crashed..

 

[Radox-0]

Oh yeah underw water is another story

i'd still love to keep my GPU cooler and especially more silent when gaming
My temps are usually around 65 to 70 on core, around 82 to 90 on memory depending on game.

those are good temps but i would love to have less noise the fans are almost at max speed to keep the gpu cool...


I was interested in undervolting but if I do so would I lose a lot of FPS knowing i'm running +100 on core and clocking usually around 2070 to 2100mhz in game?

I've tried it for 1mv for 2100 but games crashed..

Performance drop is not always massive. Depending on game, dropping down to 0.85v - 0.9v region, you will save a decent amount of power but likely need to drop 100 Mhz on the clock. In actual performance may mean a couple of %, but I expect the drop in power which will mean the card is cooler and in turn less noise if that's something your trying to achieve will work out better the the couple of % drop in speed.

Will learn your chip behaviour pretty well, seeing what clock speeds stable at each voltage.

 

[Elric82]

Which clock should i go for at 0.9 for instance?
I'm trying that, will do some test https://www.casimages.com/i/210521075224786781.jpg.html

Is timespy for instance a good test?

 

[Grim5]

Which clock should i go for at 0.9 for instance?
I'm trying that, will do some test https://www.casimages.com/i/210521075224786781.jpg.html

Is timespy for instance a good test?

Timespy is fairly good at detecting unstable overclocks, Sonia unigene heaven

 

[Raiden85]

3090 Strix OC here and when completely stock I find the boost clocks can vary quite a lot depending on the game, If I up the power limit to 480w and up the GPU clock I have no issues getting it to maintain 2ghz with decent GPU and VRAM temps, but at stock it can vary quite a bit.

In AC Valhalla the GPU has no issues auto boosting and maintaining 1930Mhz/1980Mhz while the GPU temp stabilises at around 67c, VRAM tops out at 88c. Same temps in The Division 2 the GPU likes to run between 1730Mhz/1830Mhz and in Metro Exodus enhanced with all the ray tracing features turned on and maxed out, same temps it can go as low 1650Mhz to 1730Mhz, even Quake 2 RTX does the same thing, once the card is making heavy use of RTX features the clocks can drop quite a bit at stock even when there is plenty of temperature headroom.

I have run the card at 2.1GHz for timespy runs (does fluctuate between 2Ghz/2.1Ghz) so no issues there and my scores are just fine. Strange why some games seem to put such large loads on the card that it needs to drop the clocks by large amounts.

https://www.3dmark.com/spy/17632860

 

[Radox-0]

Which clock should i go for at 0.9 for instance?
I'm trying that, will do some test https://www.casimages.com/i/210521075224786781.jpg.html

Is timespy for instance a good test?

It’s going to vary and as above depends on game / task, time spy and similar use a lot of power compared to a non-rt game. I think therefore its a good idea to try and target both things like timespy but more typical games you play a lot which are heavy but Non-RT. As you have a card with lower power limit 375w, something like Timespy means you smash into power limit pretty quick and clock speeds drop so in effect the power will limit things.

Personally I set up a couple of profiles, Non-RT so profile above and then a RT profile and learn behaviour of chip at each step. Can take some time, took me a day or so to get an idea of my chip and the refine it as I played more games.

 

[Elric82]

Tring 1965 for 0.9, seems stable for now.

Playing AC valhalla, getting 60c on core and around 76c on memory with a less agressive fan curve than I had before : https://www.casimages.com/i/210522065359201274.jpg.htm

I played some Days Gone before but this game seems extremely heavy on memory for some reason and it dropped a bit but remains around 82c.

I haven't touched the memory not sure if it worths it?

My temps are fine even without UV but I really hate the noisy fans actually and i want to have a bit more silence

 

[Radox-0]

3090 Strix OC here and when completely stock I find the boost clocks can vary quite a lot depending on the game, If I up the power limit to 480w and up the GPU clock I have no issues getting it to maintain 2ghz with decent GPU and VRAM temps, but at stock it can vary quite a bit.

In AC Valhalla the GPU has no issues auto boosting and maintaining 1930Mhz/1980Mhz while the GPU temp stabilises at around 67c, VRAM tops out at 88c. Same temps in The Division 2 the GPU likes to run between 1730Mhz/1830Mhz and in Metro Exodus enhanced with all the ray tracing features turned on and maxed out, same temps it can go as low 1650Mhz to 1730Mhz, even Quake 2 RTX does the same thing, once the card is making heavy use of RTX features the clocks can drop quite a bit at stock even when there is plenty of temperature headroom.

I have run the card at 2.1GHz for timespy runs (does fluctuate between 2Ghz/2.1Ghz) so no issues there and my scores are just fine. Strange why some games seem to put such large loads on the card that it needs to drop the clocks by large amounts.

https://www.3dmark.com/spy/17632860

Am curious have you tried undervolting also in some games? I know you have a strix so higher power BIOS, but wonder if shoving so much power and voltage as the card gets hotter is not contributing to voltage / power leak so actually making things less stable if you try to hit higher clocks. More so as the card gets hotter with more power shoving through it etc.

Tring 1965 for 0.9, seems stable for now.

Playing AC valhalla, getting 60c on core and around 76c on memory with a less agressive fan curve than I had before : https://www.casimages.com/i/210522065359201274.jpg.htm

I played some Days Gone before but this game seems extremely heavy on memory for some reason and it dropped a bit but remains around 82c.

I haven't touched the memory not sure if it worths it?

My temps are fine even without UV but I really hate the noisy fans actually and i want to have a bit more silence

Some games can benefit from a memory overclock, others not so much. Can increase memory clock, but keep an eye on temps I think.

 

[Grim5]

It's a bit random as to who which games benefit from higher power

I have a 3090 Strix and I find just by changing power draw from stock 390w to 480w without changing clocks will improve performance in AC Valhalla by 1-2% but in The Medium with Ray Tracing on doing the same nets 12%

 

[James Mackenzie]

This is completely worthless. Every game/application will put a different load on the card meaning some applications you will get a higher or lower clock. If you actually want to compare core clocks all use the same benchmark.

I would recommend comparing 3d mark scores for a better understanding of gpu performance. If anyone is interested I scored 14,269 in port royal if anyone is interested (the card is overclocked).

 

[Purgatory]

This is completely worthless. Every game/application will put a different load on the card meaning some applications you will get a higher or lower clock. If you actually want to compare core clocks all use the same benchmark.

I would recommend comparing 3d mark scores for a better understanding of gpu performance. If anyone is interested I scored 14,269 in port royal if anyone is interested (the card is overclocked).

Finally someone that understands how a GPU works and how to correctly test any computer part against another..

Also some of the temps stated by some are either on water or they live in an igloo.. When comparing temps state also what cooling you are using on the gpu and for your case, is the card running stock or OCed etc etc, or the numbers mean nothing.

 

[Samhawkes86]

What is your clock rate when under 90/100% gpu load?

 

[RipGroove]

FWIW then:

3090 FE (No Overclock)
1695-1710Mhz (Reported via Afterburner)
73-74°C @ 50% fan rpm (Custom fan curve, see below)
97-98% GPU Usage
Unigine Heaven 10 mins (temp would probably creep up over a longer duration)

Ideally I'd like lower temps but so far this is a compromise between temps and fan noise.

And TIME SPY results: https://www.3dmark.com/spy/21019586 (Average GPU Temp of 68°C)

Default settings with G-Sync Off, assume it's 1440p because it does not support 3440x1440 Ultrawide.

At what temps do these GPU's start to throttle at? Do my GPU clocks look right in my TIME SPY results? It seems to run at 1695-1905MHz while gaming and fluctuates wildly? (Down to the game I expect and the fact I'm bouncing off of the games hard FPS cap of 200) Would FurMark not be a better comparison?

 

[RipGroove]

Would FurMark not be a better comparison?

Probably not as I just ran it and my GPU was around 900MHz average so probably not the best test.

 

[KillBoY_UK]

2000-2050mhz 48-50c @365w on a Watercooled XC3 (stock clocks0 and very slow fans) a 100mhz or so less @350w 2-3c less on the temps