momaka wrote on Yesterday, 19:11:

To do that, step 1 would be to just get rid of that stock cooler and install something much bigger - dual slot preferred. Or maybe one of the large passive coolers and put a fan on it. Otherwise with the stock cooler, you can expect your full load temperatures to easily

I've decided to get a Zalman VF900 for my card as I heard that it's compatible with 7900gt series and keeps them cool.

momaka wrote on Yesterday, 19:11:

In reality with these bumpgate era cards, you should keep it at 55C or less at full load if you want it to last at all.

This is just an arbitrary number though, is not it?

Actual critical temperature is 70C as far as i am aware. And that's surface temperature, not hotspot or anything. And while it makes sense to include some safety margin, i do not think using such arbitrary numbers is a good idea. Because then people add some safety margin to your safety margin and end up with completely absurd numbers/solutions which can easily be harmful, not helpful.

DrLucienSanchez wrote on Yesterday, 16:09:

Are any of you using a 7 series card or G71 based quadro able to control the cards fan via software?

I'm using a Quadro FX 1500 with the 82.69 drivers in 98SE but unable to control the cooler fan via Riva Tuner, so it's running at 100% fan all the time, not ideal, fine in XP though. It allows me to chaange the fan percantage, save, but does not apply the setting.

The fan control works for me with all 7800/900 cards. Win98SE/XP. After the driver is loaded, it dies down, RivaTuner can also control it. Under Win11, however, there is a secondary problem, 100% speed when inactive, I need to find a solution for this. There is a driver for Win11, but because of the two NVidias I can only activate one. GTX1650 is the main card, there are no problems with it, its own control is also ok, the fan stops at low load.

Archer57 wrote on Today, 02:43:

This is just an arbitrary number though, is not it?

Actual critical temperature is 70C as far as i am aware. And that's surface temperature, not hotspot or anything. And while it makes sense to include some safety margin, i do not think using such arbitrary numbers is a good idea. Because then people add some safety margin to your safety margin and end up with completely absurd numbers/solutions which can easily be harmful, not helpful.

Not arbitrary.

But with safety margin included, like you said.

No, most gaming forums I've browsed, people almost never include a safety margin when doing temp calcs... that is, for the few that even look at temperatures at all. Most people's thoughts are usually along the lines of, "oh, according to nVidia, it's fine for the GPU temps to go up to XX, so my hardware should be fine." NO!

My number is based on actual experiments I've done my own hardware. You want that low temp, because as I've mentioned in other threads in regards to this issue, the lower the temperature differential between high and low temps, the less expansion and contraction there will be between materials in the GPU. The underfill on these bumpgated cards already looses much of its properties at 70C and is essentially useless. Even at 60-65C, where older PS3's often run under load, we still see lots of failures. And I have boards of my own with these bumpgated chips that ran around 60C max and still failed.

So yes, 55C is very much a realistic number to aim for... though in reality, anything under 60C can be considered decent already.

Archer57 wrote on Today, 02:43:

Because then people add some safety margin to your safety margin and end up with completely absurd numbers/solutions which can easily be harmful, not helpful.

I've seen that happen far far less (probably can count the number of times on my fingers if I try to go back to old forums posts) compared to the people who just don't give a shit or simply put a bigger fan in the case + change the thermal compound and think that solves the problem.

Overkill cooling IS good. The only issue you may get from it is if you put a really heavy cooler and don't properly support the card afterwards. On that note, even stock coolers on higher-end cards can be heavy enough to warp the card and break the BGA over time, so nothing new here.

MrSegfault wrote on Today, 02:30:

I've decided to get a Zalman VF900 for my card as I heard that it's compatible with 7900gt series and keeps them cool.

Sounds good.
Some 7900 cards actually came with that from the factory (very few more premium models.)
The VF900 is a pretty good cooler for up to around 70W TDP (if you want temps to stay below 60C), so should fit the bill with these cards. Temps will probably be in the mid to high 50's under full load at average room temperatures (~23-25C) and with a well-ventilated case.

DarthSun wrote on Today, 03:42:

Under Win11, however, there is a secondary problem, 100% speed when inactive, I need to find a solution for this. There is a driver for Win11, but because of the two NVidias I can only activate one. GTX1650 is the main card, there are no problems with it, its own control is also ok, the fan stops at low load.

Yeah, that would be the problem here - the fan speed on these old nVidia cards is driver-controlled. So when the card is inactive, I suspect Win10/11 disable or suspend the card's driver, making the card default to whatever fan speed is programmed in BIOS for text mode / no driver mode. On most nVidia cards I have seen from that era, this is typically 100%... though on a few different brands it's not. Most likely you will need to play with modifying the BIOS with Nibitor, then reflash the card. I know a guy on another forum that would do that on a lot of his card to keep them quiet when in text mode or idle on the desktop.
I personally tend to shy away from software/BIOS mods and rather just slap a big cooler on the card with a big fan to deal with the noise.

momaka wrote on Today, 11:20:

Not arbitrary.

But with safety margin included, like you said.

So there is statistical data available with significant number of samples which shows failure rates for different temperatures and, for example, at 65C it is significantly higher than at 55C? I suspect not...

Why margin has to be 15C? Not 5C, 7C, 10C, whatever?

Is there any data at all, showing that lower temperature even helps defective chips, assuming in any case Tg is not reached?

momaka wrote on Today, 11:20:

Overkill cooling IS good. The only issue you may get from it is if you put a really heavy cooler and don't properly support the card afterwards. On that note, even stock coolers on higher-end cards can be heavy enough to warp the card and break the BGA over time, so nothing new here.

Sufficient cooling is good. Overkill anything is bad. A waste of effort at best, potential cause for different failure at worst. Any modifications can have unintended consequences which are hard to foresee and i've seen plenty of hardware murdered by attempts at "overkill cooling".

Overall my opinion:

- All hardware manufactured in this way is doomed to fail at some point. Design itself implies that, certain wear is happening no matter what. Just like any ball bearing will fail eventually.

- Hardware from bumpgate era is doomed to fail faster. There are multiple reasons, not just underfill. This is unavoidable.

- There is nothing but logical conclusions based on unverified assumptions to show that lower temperature at any cost = better. There are multiple factors omitted from this conclusions, like rates of temperature change (up or down), mounting pressure/mechanical stresses, etc.

- There is a reason to be careful with any modifications to cooling system, as such modifications can have unintended consequences and it is highly arrogant to think you can foresee everything. Simple example - someone does not like stuff idling at 55C with fans off and disables fan stop. Now it idles at 35C with fans on. Nice? Seems so. Except temperature variation between load and idle was just dramatically increased, say instead of 55->65C you now get 35->65C. How would that affect longevity?

- So there is no reason to get obsessed with maintaining temperature below certain arbitrary value or even "as low as possible", assuming it somehow helps. Making sure that it does not reach underfill Tg (70C) in case of bumpgate affected HW is sufficient. So set fan curve to max out at 65C, if that's too loud or even can not be maintained then there is a reason to consider swapping the cooler. But on most cards it should be fine and simply decreasing ambient (improving case airflow) should provide reasonably good results.

And just to be clear - this is nothing more than my opinion. I just highly dislike the idea of "overkill cool everything" some people get obsessed with, because in my experience i never seen anything come out of this other than wasted money and dead hardware. One funny example of how "overkill cooling" can spectacularly backfire are SSDs and whole story with water blocks for them...