Huh?  Looks like you're the one who can't read.

Forget it.  A 2.8 GHz Pentium 4 will only achieve around 1.12 million hashes per second.  All 45 computers, together, will be around 50 million per second.  You're not using spare CPU time; you'll instead be running all of those computers at full power 24/7, which is roughly an extra 70 watts each.  So 3150 watts to produce 50 million hashes per second.

At the current bitcoin difficulty, it would take you around 12 weeks to generate 50 bitcoins (and that's assuming difficulty doesn't increase, which is unlikely).  12 weeks * 3150 watts = 6300 kilowatt hours of electicity.  At average US residential rates (~$0.11/kilowatt hour), that's around $700 of electricity to generate 50 bitcoins which may be worth roughly $40.  (Plus don't forget about the added wear and tear on the computers and the extra heat they'll be generating, and all of the fans that will be running at full speed all the time.)

By comparison, a single 5970 video card will use 300 watts to generate roughly 600 million hashes per second.  That is 12 times faster with 10.5 times less power, or 126 times more efficient.

So ... no.  

There is this one, right?

I was using a 5970 until I got this 6990 on Monday, and I upgraded my drivers every month, through 11.3, on Win 7 64.  No crashes.  The only problem I had (which started long ago) was that with GPU acceleration turned on in Flash, the image on my second monitor would jump up momentarily when a video started (actually not the entire image; there would be a black bar about 1/4 of the way up the screen about 50 pixels tall, and the image above that bar would jump up and back down).  Other people had the same problem.  I simply turned off GPU acceleration months ago and forgot about it.  

Edit: The 6990 does not have that problem, which is nice.

It seems to be every driver version, and GPU acceleration in Flash makes no difference (tried both ways).  But it's not Flash-related--it's a problem with any video overlay. There's the same problem playing video in any other applications (Windows Media Player, Media Player Classic, etc.).

See this thread on the Bitcoin Forum where it has been discussed extensively, along with the official "GPU lockup if running OpenCL and playing video at the same time" thread on the AMD OpenCL forum.  It was a problem a year ago with different cards and different drivers, and then resurfaced again now with the 6900 series.

Sounds perfect. The current drivers will also be very happy with that configuration.

If you're using the 6990 by itself, it will work fine for mining.  I'd leave the 5970 in your gaming rig (and do bitcoin mining in the background) since it won't lock up your system while mining if you watch a YouTube video, etc., like the 6990 will.

Damn, I should have that printed on a T-shirt after my 6990 upgrade fiasco.   

The 6990 is not supported under Linux yet, even with the 11.4b hotfix.

Probably will be fairly soon; all other 6900 series cards are.

I've got that exact configuration and it doesn't work at all I'm afraid.  At least not with 64-bit Windows 7 and Catalyst 11.3 or Catalyst 11.4 (beta), and the 2.4 Stream SDK.

What happens is that if you have both cards enabled in Windows (i.e. with monitors connected, and your desktop expanded onto them), any OpenCL program you try to launch will simply freeze when it tries to initialize the OpenCL subsystem.

If you disable the 5970, then you are able to start OpenCL programs but they can only see the 6990.  AMD says they know about the problem and are "looking into it" but I'm not hopeful for a resolution anytime soon.

The 6990 is also frankly a bit annoying for everyday use, because if a program is using the first GPU for OpenCL calculations and you watch a video (YouTube, Windows Media Player, etc.) the video driver freezes and you have to reboot. This is also a "known issue," although I'm more hopeful for this one to be fixed.

But as far as the two cards working together ... I wouldn't hold out hope.  If you can figure out a way to make it work, I'd love to know how you did it.  :-)

Exactly!

And even with that very conservative estimate of a 10% change per two weeks, difficulty increases by a factor of 12 (1.10²⁶ = 11.92) over the course of a year.  By the end of that year, a rig that generated 50 BTC per week initially becomes one that takes an entire three months to generate 50 BTC!

Take a look at this graph.  It shows difficulty versus time on a logarithmic scale.  (There's also a linear scale version available here.)  You can see that difficulty has increased from roughly 200 to 80000 (i.e. by a factor of 400) since the middle of July 2010.  That's 9 months, or 39 weeks.  Difficulty adjusts every two weeks, so that is 19 adjustment periods.  That means an average of roughly a 37% increase in difficulty every two weeks over this time period, since 1.37¹⁹ ~= 400.

You can work this into your equations easily.  You start with the equation:

factor(rate, numPeriods) = (1 + rate - (1 + rate)^-numPeriods) / rate

In PHP, that would be:


You need to allow the user to enter in a rate, which is how much they expect the difficulty to increase per two week period.  You could default this to 0.37 for example, or to a more conservative value like 0.20 if you think the number of people mining bitcoins isn't going to keep increasing at the historical rate.

For your specific code, you'd change it to be something like this:


As an example, your current page with the default values shows 2676.80 coins per year (you say blocks, but mean coins). That's 103 per two week period.  We then calculate "factor" as above, assuming a rate of 0.20:


For a final answer of 613.88, rather than 2676.80.  This is a huge difference, but is very real.  If you use a rate of 0.37, you get a factor of 3.70, for a final answer of 381.1 coins, which is seven times less.

I would also suggest that you allow people to enter the number of periods, rather than one year.

The difficulty changes every 2016 blocks, which on average will be once every two weeks.  It can change by up to a factor of four in either direction, but lately has been much less than that.  Right now, the difficulty is 82347, and best estimates say it will increase to 93094 at the next change in 717 blocks (+13%).

Suppose you had a configuration that could produce 580 million hashes per second (roughly a pair of 6950's).  That means it takes 7 days, 1 hour, 23 minutes to generate 50 bitcoins on average right now.  That's just a hair under 50 bitcoins per week.  Let's say 49.35, which is a number I happened to do the math with earlier today.

As a simple example, consider the case where the difficulty continued to increase by 13% every two weeks for a year.  We'll assume the difficulty changes exactly every two weeks for simplicity.  That means the following would happen:


The current exchange rate is USD 0.92 for one BTC.  (Please note that I am not considering exchange rate fluctuations; that is a different topic, and if you want to make money on exchange rate speculation, you can simply buy the bitcoins rather than generate them.)

Each 6950 consumes 200 watts at full load, so that is 400 watts total.  A high-end power supply is approximately 85% efficient, so that means they will consume 470 watts of power.  Per day, that is 11.28 kilowatt hours, or 78.96 per week.  According to the Department of Energy the average power cost in the US right now is 11.04 US cents per kilowatt hour, which means a cost of USD 8.71 per week.

That means that after week 30, you would actually be losing money, so you would stop then.

Up to week 30, you will have made 660 bitcoins, which at the current exchange rate is USD 607.20.  Your electricity costs will have been USD 261.30.  This gives you a net profit of USD 345.90.

Considering that a 6950 sells for USD 240 right now, this does not pay for the cards (let alone any associated computer) before mining with them is no longer profitable.  You will need to sell the cards to see a profit.  The days of making a fortune in the bitcoin market by buying hardware are likely numbered.

Nice!  Also, I forgot to say thanks for looking into the weirdness in pool block 2861 (i.e. block 116884) and taking care of it and distributing the earnings appropriately.  Greatly appreciate your responsiveness!

Is there any way to see which blocks a particular worker found? I noticed that one of mine found one today and purely out of curiosity I'm wondering which block it was. I'd take a certain amount of pride in being able to say, "Oh, yeah, block 117551, that was me!"

What voltages do you use--in both situations--for the GPUs and memory?

If you're really serious about a cluster, you should use a PCIe expansion chassis designed for GPUs, which can hold 16 cards rather than a bunch of separate computers.  For example, the Dell PowerEdge C410x.  It works with ATI/AMD cards as well as NVIDIA cards, and can plug right into any computer to add 16 PCIe slots.  In a 42U rack you can easily fit three computers, nine of those, and a power distribution unit and you'll find yourself with 144 6990's--which is somewhere around 100 billion hashes per second.  All for a cost of under 140 000 BTC.

Yes, 11.4 is beta.  I do believe you can manually set the fan speed by creating custom profiles (editing them yourself or with a tool), but it's not a feature provided in the standard Catalyst Control Panel, and since I'm just doing this for fun, I don't really want to mess with something that might result in my card bursting into flames.  :-)

Having run 11.4 for about 24 hours now, I can now update my previous observation.  The fan isn't always louder.  It seems to be loud for an hour or so at a time and then can be quiet for many hours, then eventually gets loud again.  Not sure what is up with that.

I'd imagine there is probably some sort of tool I could use to monitor and record the fan RPM and the GPU temperatures.  However, I won't be going back to an older version of the drivers (since I'm upgrading to a 6990 on Monday), so this data would be in isolation as I wouldn't be able to collect similar data using the old drivers to see what the difference is.

Another difference is that the ambient temperature may be changing, since the weather is also starting to get warmer.  So it may not be due to 11.4 but because we've had a 10°C change in the outdoor temperature over the past three days.  The computer is in my basement, where the temperature should be relatively constant, but perhaps even a couple degrees would make enough of a difference to bring the fan to the next higher speed.  Who knows.  All I know is I can hear it upstairs, whereas I didn't used to be able to.

On my 5970 on 64-bit Windows 7, there is no difference performance-wise (higher or lower) compared to 2.2.

However, I also upgraded my drivers from 11.2 to 11.4 at the same time.  There is one noticeable difference and that is that the fan noise is TERRIBLY louder than it used to be.  The temperature is now pegged at 88 degrees C when it used to be pegged at 89.

Strange indeed.  Thanks for the prompt reply.  Sounds about like my luck; I just start on this and within an hour bugs start popping up.  ;-)

And I have no idea what JackRabiit is talking about, although it sounds like he thinks that block being marked invalid had something to do with him specifically, as a result of him clicking "Stop" and "Start" quickly in succession?