Nvidia minimum fan speed - why does this keep happening?

[Rawr]

I'm in the process of returning my 5090 Solid OC for pretty much this reason. The sound of it 30% and upward kind of sucks, but it is the start/stop that drove me spare. I was able to stop it happening as much but tweaking the slope with Fan Control, but it can't be avoided 100%. Even just the start sound when beginning and ending a game sounds dreadful imo.
The worst part though is that the fans drift in and out of sync constantly doing that "whoom-whoom" effect. (the thing that the Noctua/GN video explains)
Spent days trying to offset them, but they will not hold a set speed well enough to make it matter anyway.
I have a family member who has just strapped 3x 92mm Noctua fans to their 4090 and it is in a different league. Any % you want, any curve you fancy, and so much quieter than any stock card I've heard. I'm a bit jealous, but I can't be bothered with the whole thing, especially after spending a couple of grand.

I've necro'd this post I know, but it came up while searching to see if any 5090s have fan speeds that go below 30% lol. Seems the answer is no, so hopefully whatever one I do get at least doesn't make the cat jump when it kicks in.

Yeah I went with the 92mm fan hah. I've just got it sitting at the top of the right end of the card blowing air in. It's largely solved the problem outside of some rare situations on hot days.

Shouldn't have had to do it though, the quiet bios should have been an actual quiet bios that sorts this kinda thing out.

 

[Dave2150]

Welcome to my ranting thread about unacceptable minimum (30%) fan speed on Nvidia cards.
You are especially welcomed if you are similarly noise-conscious about selecting PC components as am I.

So: I just got Inno3D RTX 5080 and was quite annoyed / almost shocked when I noticed that it has minimum fan speed of 1400rpm. Just to clarify - it of course supports fanstop,
where during idle the fans have 0 rpm. However the problem is, that when you are running 4K @ 120Hz (not to say 240Hz) and doing light GPU work (playing video etc.),
or just simply running the desktop apps (when working), it's enough to have the fans occasionally spin up even during this so called "idle" to cool the GPU back down.
(And this is already happening during winter - just wait for the summer!)

And of course it can not be done with some reasonable rpm like around 600, which would be more than adequate for the mild GPU load.
No, it has to be either 0% rpm or sudden 1400 rpm spikes. Which is miles above noise floor of any semi-decently noise tuned PC in idle and it grabs unwanted attention to its self.
Thus the real "idle" rpm for this particular card is not 0 but the superposition of 0 and 1400 rpm.

More generally: as far as I know all Nvidia GPUs for some time now have their minimum rpm set to 30% of the max rpm.
Therefore all modern Nvidia GPUs have this problem - it has actually reminded me why I replaced stock cooling solution on my previous RTX 3080 SUPRIM
with two Noctua NF-A12 fans attached to the heatsink. And AFAIK there is no good software workaround for this. At least I didn't find it in case of my RTX 3080.
(i.e. seems to me, that when certain conditions are met, firmware on the card will spin the fans and overrule any fan profile you might have set in Afterburner or FanControl etc.)

Now you might think that the higher end models don't suffer with this, however if you look at the techpowerup reviews of RTX 5080 / 5090 (section "Temperature & Fan noise")
you will see that with the exception of Astral models all the others have minimum fan rpm at ~1100 or higher. Which is completely suboptimal tradeoff of noise vs. cooling in idle.
Also with "higher-end" models you will typically run in more problems with coil whine, therefore if you want card with acoustically acceptable profile it's very difficult to find one.

To illustrate how absurd this is: if you look here https://www.techpowerup.com/review/msi-geforce-rtx-5090-suprim/39.html
the first (quiet) tested BIOS shows that the card is so efficient at cooling that it only requires maximum rpm of 1093 under load.
However the same card has its 30% rpm equal to around 1000 rpm. Which means during idle it will behave as a superposition of silent and also almost equally as loud as during full load.
On which planet is this considered a good design choice?

Now I would be interested if there are more people who are similarly annoyed with this, especially since it's going on for so long and nothing seems to be changing.
Does it mean everyone is "OK" with it? I definitely found some other forums where people were complaining about it, but maybe it's not part of public awareness
enough for a change to be implemented. Which sux.

Then maybe what are the software workarounds for it? I.e. I remember flashing BIOS to my old GTX 580 which lowered this stupid minimum rpm limit which together with Arctic freezer triple fan aftermarket cooler provided very good noise profile. Is there some lower than 30% bios that can be flashed to the Inno3D card? I assume I will lose warranty doing so?

Now I will definitely try attaching Noctua fan to the exposed heatsink part of the card and see if it helps.
However it's very stupid that generation after generation you just can not grab an Nvidia card knowing it will not produce annoying noise spikes during idle. (not to mention coil whine during load)
And have to come up with DIY solutions for this completely unnecessary problem, while most of the DYI solutions will void the warranty.
Maybe someone can explain to me why was 30% selected for minimum fan speed (which on most cards mean >= 1100rpm) Why not for example 20% or 15%
Why is it 30% on cards where it means 1400rpm as well as on cards where 30% means 600rpm (Astral)

Why does Jensen insists on screwing up our "10K gaming centers" with this low quality stupid nonsensical design?

My gaming rig has a 5090. I use it for gaming, the rest of the time it's shutdown. A 5080 or 5090 is a bad choice for a system that's going to be on the desktop or web browser for much of it's time - they are horribly inefficient at low loads.

For work, media and everything else, I use a 4060 which allows 4k 120hz for desktop while being completely passive. It's got a 13900K heavily clocked/optimised for productivity, sips power at idle. This system I keep on 24/7, as no silly 12V-2x6 or AMD CPU high idle power usage to worry about.

You're right that Nvidia should fix this, but they clearly don't care so best to come up with a solution instead of just whining about it.

 

[brendan.mclaugh]

A while back I got a second hand M1 Mac Mini for all non gaming stuff. The 4090 system idling at 140w for browsing and watching videos was a total waste, whereas the mac uses under 10w most of the time. The thing will pay for itself in a couple of years, and the reduction in "wear and tear" on the fancy stuff is a bonus too.

 

[Dave2150]

A while back I got a second hand M1 Mac Mini for all non gaming stuff. The 4090 system idling at 140w for browsing and watching videos was a total waste, whereas the mac uses under 10w most of the time. The thing will pay for itself in a couple of years, and the reduction in "wear and tear" on the fancy stuff is a bonus too.

Yeah makes perfect sense. Years ago you could get away with using a 1080ti for example for your "all around" PC, work, gaming etc. Now it's just not feasible in terms of wasted power, heat, noise and the 12V-2x6 safety issue.

 

[SuperElectric]

A while back I got a second hand M1 Mac Mini for all non gaming stuff. The 4090 system idling at 140w for browsing and watching videos was a total waste, whereas the mac uses under 10w most of the time. The thing will pay for itself in a couple of years, and the reduction in "wear and tear" on the fancy stuff is a bonus too.

...How? 130watt difference at 6 hours a day, 250 days a year (assuming you're gaming on the remaining days) is 195kWh per year. At 25.3p per kWh (your tariff may vary) that's £49.34 per year.

A second system is always going to be more expensive, unless you already have one for some other reason.

Although, I am curious to know exactly how much power my system uses, I might get one of those power meters

 

[brendan.mclaugh]

It was £219 off eBay with a voucher. The 4090 system was on (I kid you not) 16-18 hours a day, 7 days a week as it was used for everything basically. Browsing, TV, streaming, music, etc etc. It saves me about a tenner a month going by the electric meter/bill.
Those meters are great, cheap too. Though my dad bought a load and now logs everything, driving himself mad, silly old bugger.

-edit-
Yeah I guess it would be saving 12 hours a day roughly, so £11 a day according to an online calculator. Not had my coffee yet so I won't embarrass myself doing it mentally

 

[LuckyBenski]

Yeah I guess it would be saving 12 hours a day roughly, so £11 a day according to an online calculator. Not had my coffee yet so I won't embarrass myself doing it mentally

£11 for 12 hours suggests you're doing about £1/hr which would be a 4kW computer at 25p/kWh, so that's miles off.

 

[Eddie99]

£11 for 12 hours suggests you're doing about £1/hr which would be a 4kW computer at 25p/kWh, so that's miles off.

Saving 130w for 12 hours a day would be £11.86 per month at 25p/kWh. I don't have a 4090 but 140w system idle power use seems unusually high.

 

[SuperElectric]

It was £219 off eBay with a voucher. The 4090 system was on (I kid you not) 16-18 hours a day, 7 days a week as it was used for everything basically.

Well, that makes more sense, an (almost) always on device should be as low power as possible.

 

[Eddie99]

A while back I got a second hand M1 Mac Mini for all non gaming stuff. The 4090 system idling at 140w for browsing and watching videos was a total waste, whereas the mac uses under 10w most of the time. The thing will pay for itself in a couple of years, and the reduction in "wear and tear" on the fancy stuff is a bonus too.

I measured my 7800X3D + 9070 XT power consumption at the wall, it's 62w at idle, around 100w browsing and 110w when playing a 4K 60FPS YouTube video. This is with 3 NVMe SSDs and a RAM overclock (6000 C30 Buildzoid timings). Does a 4090 really consume that much more power at idle than a 9070 XT?

 

[brendan.mclaugh]

I measured my 7800X3D + 9070 XT power consumption at the wall, it's 62w at idle, around 100w browsing and 110w when playing a 4K 60FPS YouTube video. This is with 3 NVMe SSDs and a RAM overclock (6000 C30 Buildzoid timings). Does a 4090 really consume that much more power at idle than a 9070 XT?

I guess when I said "idle" I was really meaning "not gaming".
(if I'd known you lot were going to ask me to show my all workings, I'd have been more careful, lol )
From memory, actual idle, without the TV on, so no video output, was 100w. This bit I'm 100% sure on as I saw it many times on the plug monitor thing while rooting around the sockets for whatever reason.
However, the vast majority of the time it was browsing and watching youtube though (so much youtube, it's ridiculous actually), and that was 110w and 140w respectively.
(2x gen 5 SSDs, 1x gen 4 SSD. 2x 10TB HDDs, 32GB RAM, 9800X3D, 7x fans)

The other bit I'm 100% on is that the Mac saves me a tenner a month. My electricity use is incredibly consistent so I know exactly how much to put on the meter and since I got the thing it's a tenner less.
Can't say I like anything else about it though, but it's a means to an end.

(sorry for totally derailing your thread op )

 

[trilby]

In addition to everything already discussed in this thread, my 5090 FE seems to have an issue that when it reaches the 52C temperature threshold to take the fans from 0% to 30%, sometimes one of the fans will hugely overshoot and go up to 2000 RPM for a few seconds before settling at 1100 RPM which is where it should be at 30%. HWinfo shows both fans at 30% even while one of them is at 2000 RPM and one is at 1100 RPM.

Obviously this creates much more noise than usual, as 1100 RPM is barely audible but 2000 RPM is pretty loud. It only happens less than 10% of the time the fans spin up, but it's pretty annoying when it does. Just posting here in case anyone else is experiencing this and found this thread while Googling like I did!

 

[GigaFanRPM]

[...]

So I am also looking for a solution to the problem. So far placing the 92 mm fan on the backplate venting area sounds promising (fortunately the card is long enough so that the D15 doesn't interfere), but I am also afraid of negative impact to noise during higher load since the Asus Astral has a fourth fan which makes it louder instead of more silent (reviews have tested this by stopping it forcefully).

[...]

Since nothing else helped I also did the 92 mm fan trick. I just added a custom fan curve for it in Fan Control so that it disables itself after 55°C GPU temp so that it doesn't interfere with the other fans on heavy load. Only a few °C difference in stress test benchmarks, but a difference is a difference

 

[Ilstar57]

My gaming rig has a 5090. I use it for gaming, the rest of the time it's shutdown. A 5080 or 5090 is a bad choice for a system that's going to be on the desktop or web browser for much of it's time - they are horribly inefficient at low loads.

For work, media and everything else, I use a 4060 which allows 4k 120hz for desktop while being completely passive. It's got a 13900K heavily clocked/optimised for productivity, sips power at idle. This system I keep on 24/7, as no silly 12V-2x6 or AMD CPU high idle power usage to worry about.

You're right that Nvidia should fix this, but they clearly don't care so best to come up with a solution instead of just whining about it.

From what I've read when researching GPUs at idle and low loads, the 5080 is actually no slouch, usually beating out even the 5070 Ti in efficiency. The 5090 is a power hog, though. It wants to draw 30-50W just sitting at idle. With a light load, that could cause the fans to have to ramp up quicker.

OP has the 5080 Inno3D. This actually comes in two variants - the OC and non-OC. There is a difference, since the OC has a much beefier cooler (same as on 5090) than the non-OC (they call it the X3 model). The non-OC uses the same cooler that's used for the lower-tier cards, all the way down to a 5060 Ti! As you might guess, that cooler will easily get soaked with heat at a 5080 performance level and this will cause the fans to have to run more. Inno3D seems to have among the highest minimum fanspeed at ~1300-1400 RPM of any brand that makes the 5000 series (Zotac has similar). Again, this is due to nVidia's specified 30% minimum. Inno3D's fans can spin up to higher speeds than most other brands, so that 30% is high. I assume OP has the non-OC, since this issue appears to affect them a lot.

For a quieter build during light tasks, I'd recommend two things. First, get a beefier cooler. The beefier the cooler, the more heat it can soak up passively without having to resort to running the fans. Second, watch out for the brand. Asus has the lowest minimum fanspeeds at ~600 RPM. Most other brands, like PNY, MSI and Gigabyte, tend to have them at around 1000RPM. Palit and their sister brand Gainward are reported as having ~1150RPM, the Founders' Editions (at least for the 5090) was reported at ~1200RPM, Zotac have them at around 1300 RPM and then Inno3D gets to about 1400RPM, some users reporting even 1500, though I can't verify and seem to find that 1400 is the most often quoted number. It's important to pay attention to the brand in this respect, as you can't really do anything yourself (cannot be set to 1-29% through any kind of software or setting), aside from deshrouding the GPU, as several others in this thread have already done. Not something I'd recommend to the general, as it might void the warranty, though.

Whether or not that bothers you is personal preference. In an enclosed case that's not too close to the user, or a case with generally louder case fans, it's not going to really be an issue. I've had case fans before that ran at similar RPMs and, though audible, I got through with them for several years. Now that I've invested in a quieter build, though, that would probably annoy me. Also, it matters what tier of card you have. I'd wager their 5060 Ti, using that exact same cooler and fans, will probably be spinning them up a good deal less than a 5070 Ti or a 5080, simply due to the lower power draw.

I do hope that Inno3D do something about this. Nobody is going to be running ~4000RPM fans, so we don't need them to go that high and having a lower maximum RPM would allow for a lower 30% minimum, too. Inno3D fans from previous gens have had some issues, and while we don't have any bad takes on the 5000 series from them so far, I'd really like to see them investing just a bit more in getting high-quality fans. It really is the only major concern with their products. Besides this, they tend to be very functional and some people like their industrial-esque design. Not to mention they're typically among the best-priced. And nVidia can perhaps stop with this arbitary requirement. AMD cards have no such issues. Older nVidia cards have no such issues. Why the 30%? To make the FEs look good or what? I really don't get the logic.

PS. It'd be interesting to hear from Inno3d iChill owners. The iChill sub-brand has typically been associated with Inno3D's high-end products, and users are generally quite happy with them. I've only managed to find info on the iChill line for the 4000 series, where one user reported ~950 RPM as the minimum. The 4000 series has the same 30% restriction as the 5000. 950 would be a major improvement, sending Inno3D from the last spot to the second best in this 'race'.