Yes,

Back in the day the power switch on the PSU was what you used to turn your PC on or off - essentially cutting power every time you did so.  The components handled it fine then, and modern hardware is considerably more resilient.

Unless the light timer is really badly designed, can't handle your power draw, or for some reason introduces big voltage spikes or sags, it will not harm your solid-state hardware.  HDDs, optical drives, or any other logic device with moving parts is a different story - you don't have any of these in your rig, so nothing to worry about.  If you did, you'd just want to make sure they were parked before cycling power.

Over a very long time, frequently cycling the power will cause wear, mostly on your PSU - eventually a capacitor will give up the ghost and pop.  Your machines will probably be retired long before that's anything to worry about.

A decent digital voltmeter will tell you if you're getting voltage spike/sag when the timer pops on.  If that's ok, then just use good surge suppressors on your rigs, don't overload your circuits, and you've got nothing to worry about.

With the 3.6GH in your sig, you've got more to worry about in your power panel box than in your PCs - I just broke the 3GH wall my self and I know I'm pulling a considerable amount of power (killed a 15A surge protector!).  If your rigs are not on their own isolated 20A+ circuits, it's a good idea to make that happen before going with this plan.

With all else equal, 6990.

1) resale value if you ever want to dump it.
2) higher hashrate on the 6990, a quick look at the hardware comparison shows ~100MH in favor of the 6990
    ~600 (5970) vs ~700 (6990) at stock clocks

I haven't looked at power consumption numbers to say if that makes enough of a difference to make up for 100MH, but it'd have to be significant.

Should be just fine, though if you use persistence or write logs, etc, you can still hit interrupted writes.

I PXE boot personally, the only moving parts in my rigs are the fans, and no persistent data.  I have zero worry about flicking off the power switch on a whim

The power in my neighborhood sucks (north Denver metro area) - I've lost power momentarily about 8 times in the last two months - considering I'd need a whole bank of UPSs to keep these things on, this seemed like the best option.  USB boot will work equally well, and removes dependency on a boot/nfs server.

Unless your miners are really unstable for some reason, I'd power cycle as infrequently as possible since you'll loose hashing time during each reboot.  I guess you'll probably want to try to balance your reboot frequency against the possibility they could be down for a significant period of time.  My guess is the short twice-daily cycles will add up to far more time in the long run.

Over a long time, probably.

I'd be more worried about corrupting your harddisk if you're using one, interrupting writes by cutting the power = lost data.

If your X server is running, and you've got all your cards initialized ( aticonfig --initial -f --adapter=all as root to fix these if not ), you probably forgot to set DISPLAY=:0

Try:

#>  DISPLAY=:0 poclbm.py

and see if it lists your GPUs.

This could just be a bad bios flash - if you're comfortable with doing so you can probably find / flash a stock BIOS image for the card and see if it helps.

I think this is becoming a common feature on newer motherboard, my ASRock mobos also have reset, power, and 'clear cmos' buttons built onto the board.  This is how I turn my rigs on without a case.

It's usually not too difficult to do if you've got an old case you can cannibalize.  You really only need the power switch.  You can also just pick up a momentary DC switch from RadioShack or such and wire the two leads to the power-button header, as that's all your power button really is.  It boils down to simply tripping the solenoid in your PSU to 'on'.

I've done it with an XFX 1GB 6950 - it allowed me to flash properly but didn't have any luck unlocking shaders.  From what I've read the Sapphires should allow flashing the BIOS the same way, but it's a gamble whether or not you'll get lucky and unlock shaders..

I just used a circuit-writer pen to mod a second card:





Allowed me to flash the card with no problem again, but no unlocked shaders on this one either.

I'm not certain if I'm going to bother modding my other two cards, the heatsinks only came off these two so I could re-do the thermal paste.  

[edit]

fancy smartphone camera makes an amazing magnifying glass

This is also removable with a cotton swab and a bit of acetone.

Time to try out water cooling

It's also a good idea to clarify your position when shooting someone else's down.  Either counter with stats that invalidate the point, or you're also spreading misinformation.

I used AS5 on one of my 6950s that was constantly > 90C, it's now running a nice chilly 80C overclocked slightly (885 gpu / 760 mem).

Applied properly, AS5 shouldn't give you any problems.  Applied improperly perhaps, but there shouldn't be anywhere to bridge gaps unless you get it on something outside of the GPU socket housing.

A 200hr cure time is ~8.3 days, to a miner that's not a long time.  I got cooler temps immediately, and they've gradually dropped to stabilize at 80C.  This was a huge improvement for me.

That's not to say there isn't better stuff than AS5, but when it's what you've got on hand it does do the job.

What kind of cards are in that 4x GPU system?  Is it really 8 GPUs on 4x 6990s?

If not, I don't think you should be pushing 15A unless you've got a bunch of other stuff on that circuit.

I ran two new dedicated 20A circuits for my rigs, right now I'm only using one for my miners - 4x 5830s and 4x 6950s, two AC high-velocity desk fans and a monitor, and I'm just pulling 13A at the socket.  The fans pull ~3A each I believe.

Seems like you should be coming in well under that, but again a bunch of other stuff on that circuit will reduce your available power.

5830's have a TDP ~150W, if you want to give yourself head room say 200W / card.

That's 1200W for the cards max, plus whatever power for your base system.  I think 1500W should do fine, but I don't know much about the particular PSU.  In reality I'd expect that system to come out ~950-1000W.

The bandwidths don't matter, assuming it doens't turn off any of the lanes when other slots are occupied it should work fine.

I'd go for extenders with molex power for at least two of the cards however; 6 cards would be pulling ~450W through the motherboard, better to let the PSU supply that directly.

I wouldn't run without the fan - if you want silent you'll want a much larger passive heatsink.

I've got Sempron 140s underclocked to 800mhz and undervolted, the CPU fan is basically inaudible.

It's certainly no where near the noise a desk fan and 6 GPU fans whirring at 100% make

The Corsair HX and TX series power supplies are awesome, I'd get them any time.

I have an HX750 (750W) powering two 6950s and two 5830s right now - it's right up at the max-fill line (740W at the wall), and it's definitely warm, but it's handling it with shining colors.

Note that this is bad to do long term, I'm adding Corsair CX430s to supplement soon.  At max, any PSU is going to produce a lot of heat and waste power, and likely shorten it's lifespan.

This is a common misconception - the 80% rating is an efficiency measurement only.

If you have a power supply rated for 750W, it should deliver 750W.  

The rating means it will supply that 750W with 80% efficiency.  This does not mean it can only deliver 600W to your PC.  This means it will be pulling up to 940W at the wall.  

80% of 940W is ~750W delivered.

At 100% efficiency, 750W at the wall would be 750W delivered.

At 60% efficiency, 1200W !! at the wall would deliver 750W.

With an 80+ Gold (87% minimum), ~860W at the wall would deliver 750W.

Make sense?


These numbers are all absolute minimums, real world efficiencies will trend higher usually.

All of that said, it's better to err on your side of the equation.  Loading a 750W PSU with only 600W means you won't blow the PSU or burn your house down

Fair warning - that's a huge, heavy 3-slot card.  Love the total overkill for a 5850, bet it clocks pretty well

The memory and GPU clocks are wholly independent, functionally and physically..  which makes it a real PITA that the mem clocks don't behave in linux. 

You can't hurt the card by tuning down the memory, however going too low may cause it to act a little weird.  Don't flash any values you're not comfortable with.

If you've got a Windows box handy it can live in for a couple of days, you should try to find your GPU / memory clock sweet-spot (where you get the most MH at comfortable temperatures and low memory clocks), then flash those settings for permanence.  After that you don't even need to mess with the clocks back in linux-land, it'll boot up where you want it.

I have noticed that a card flashed with lower clocks, when modified with aticonfig will fall back into the gpu - 125mhz bug regardless of the BIOS settings.