Topic: Efficiency of 7970(youre gonna like this) (Read 4161 times)

My cards are in a data center.  A few of the earliest ones were run at home, and a few were purchased used with who knows what history, but many of the failures were new cards run exclusively in the data center.

My cards are in cases and densely packed with something jammed between them to create small gaps (similar to your blog post but probably even smaller gaps due to the case).  Perhaps heat is the issue, or perhaps the restricted air flow makes the fans work harder.  They are also 5970s which may be worse than 6990s.

Dust is probably the root cause of your fan failures (are you mining at home?). My miners run in a datacenter with filtered air. When I have compared my low fan failure rate to other 5970 miners, dust has always turned out to be the main difference between my environment and theirs. Your fans would have likely failed just as spectacularly even when running at lower speed...

Also, most of my fan failures affect disproportionately 5970s, not 6990s, even when taking into account the fact that my 5970s are generally older than 6990s (I have 5970s whose fan failed despite having been in operation for less time than most of my 6990s). I would venture that the fan on 5970s is of lower quality than 6990s.

On high-density 5970s miners, at 25C ambient, the GPU temp diff between the fan at 50% and 100% is a full 10C or more. Not negligible!

As I'm only buying used cards, I consider fans to be already a little worn out. I've had a 5970 fan failure and the end result (after several weeks of tinkering) was a dead board. There are 2 kinds of fan failure I saw symptoms off : fan blocked (bearing dead or full of garbage/dust), fan axis off-center (very noisy with resonances at some fan speed amplifying the problem and ends with fan blocked too but with noise as a warning before doing so).

For the first one you probably want to avoid dust, heat and too much speed. Dust is obvious, heat may cook the lubricant (theoretically... I'm not convinced it can), too much speed may push the lubricant out of the bearing on some kinds of fan (if you practiced roller skating or any sports using ball bearings for example, you probably had to clean and re-oil them).
For the second one, I believe it happens on some kinds of bearings : the more friction, the more risks the axis will be worn out and get off-center. More speed -> more friction. Don't put your boards at weird angles either...

I don't think that all board vendors have designed their cooling systems to run at 100% : some may have, but this is not the kind of use the cards are sold for. So while it may work for some, I prefer to avoid running fans at 100% (I limit most of them to 85% and try to setup my rigs to get under 70% if I can).

There's another reason to limit fan speed : power efficiency.
The cooler the GPU runs at, the less power it requires but the faster the fan spins the more it requires. And from what I could see, there's not much fan power usage difference between 0 and 70% (given the heat inertia of the boards, if you monitor your power usage and change the fan speed quickly, you can see the power usage change before the GPU temperature changes and get a good estimation of the fan power usage). That's not true between 70 and 100% with the boards I tested (Asus 5870, Sapphire ref 5870 and Sapphire 7970). I don't know why but it's how they behave. Today, after installing a remote power usage monitor, I checked this and tried to find a sweet spot for my boards.

In my case I found out that being cautious cost me some kWh and to lower my power consumption I had to lower the temperature targets in cgminer, which raised the fan speed. But while experimenting with different temperature targets, when fans reached 70-80% I couldn't gain anything more and trying to reach 50 or 55°C on some boards was even raising my overall power usage.

So given there are some unknown risks to run at 100% and no benefit I can think of, I encourage everyone else to do so... after all the more fan failures in other rigs, the more BTC for me

My experience is different.  I too have been mining on dozens of 5970s, continuously, with the fan speed manually forced at 100%, with my oldest cards mining for 18+ months.  I would estimate my fan failure rate is at least 25%.  I don't know the reason for the difference, but if I were starting over now (say with 7990s) I would definitely try to take it easier on the fans.

Well, not that high, but close. I've tested "benefits" of full fans speed at winter with 22C room temp and had 3C lesser temps at max. Anyway my cards temps are below 70C, so theres no benefits to run fans faster, than their current auto speed - 40-50%.

Maybe when the ambient temperature is 35°C.. yeah.


FML

 Don't see any reason to run fans @ 80 or 100%, than current 40-50%, for extra 2-3C cooling.

Well said.

Running fans at 100% for a year on 8 cards and they still turning like a boss.

I am a PC Technician in UK and i see a lot of fans, from my experience it's all about the quality of the fan.

Good quality fan can run for many years.

Low quality fan will create friction as it turns, this will create considerable heat, and the heat will accelerate the degradation of the fan.

Thus there is a good reason why one fan costs $1 and another fan costs $10.

As a farm operator who is mining on dozens of 5970s and 6990, continuously, with the fan speed manually forced at 100%, and with my oldest cards mining for as long as 18 months, let me tell you this claim is rubbish! (My fan bearing mechanical failure rate is 6%, which is low and expected.)

The claim that "running fans harder than 80% kills them" is spread by paranoid teenager geeks who covet their expensive gaming hardware without an understanding of the electronic and mechanical engineering work to control failure rates.

When a fan datasheet says it is designed for a lifetime of X years of continuous operation, it assumes a 100% duty cycle. In other words, it assumes the fan running at full speed. Fan manufacturers don't build fans and say "don't run it at full speed, or else..."

For example, if the datasheet says 2 years, running the fan continuously means it should last at least 2 years; running it every other day means you can prolong its lifetime to 4 years; etc. At 50% duty cycle ("50% fan speed"), the PWM voltage alternates between 0 and the nominal input voltage, the fan should last somewhere between 2 and 4 years (but much closer to 2 years because the bearings are still moving when the voltage is 0, just not quite as fast as full speed.) At 80% duty cycle, the fan should last between 2 and 2.5 years (but closer to 2 years).

To significantly improve the operating lifetime of a fan, you would have to run it at such a low duty cycle that it would be a bad idea for high-end graphics cards, because the cooling would be so much reduced that the high temperature would start seriously reducing the operating lifetime of all the other components on the PCB (capacitors, VRMs, GPU ASICs, etc).

Bottom line, fans are designed to run at 100% duty cycle. And there is no significant difference between 80 and 100%. Some people choose to run at 80% because it does not significantly decrease cooling (for the same reason that it does not significantly increase lifetime), but because the fan consumes 20% less power (and the fan is just a tiny fraction of the overall power consumed by the card).

(I choose to run at 100% because I pay a fixed price per electrical circuit, and the tiny extra power consumption costs me nothing.)

Take for a example a box fan or a desk fan with 3 speed settings. Running it 24/7 at the fastest setting, it will definitely have a shorter life than one running 24/7 at a lower setting. The motor will heat up due to the increased power and the air between the rotor and the magnets will also be hotter due to friction at high speeds. The airflow created by the fan won't be much help cooling the fan motor. Add to that the dust collected inside and you will soon have a dead fan in your hands.

If unlucky, with a dead fan, you will also end up with a dead card. Why would anyone submit themselves 24/7 to the torture of GPU fans at 100%. I even get a headache listening to several fans at 70%.

I don't see why it'd kill them more than, uh, 25% faster than running them at 80% would..

I don't know of any presentable evidence but if you ask any GPU farm operator, they will tell you that running fans harder than 75-80% kills them.

He posted a screenshot of using AB 2.2.0.

how did you set the voltage?

what are the default voltages for level0 and level1 ?

That sounds almost competitive with BFLs single, price/performance speaking.
Nice.

Yes, but the question posed was that does increasing fan speed actually decrease fan life? No one was asking "Does undervolting increase fan life?"

Exactly. Running fans at 100% is actually beneficial in the long run and much better than 80% say.

Find it very funny that most of the miners say otherwise so that other miners have their fans failing and less difficulty for the misinforming miners

OK, so longer fan life is a result of keeping the bearing temperatures down. Again, undervolting can accomplish this.