Network Hashrate variations

[aqrulesms]

Hello,

A trend I have been noticing for quite awhile is how much the total network hashrate varies.  It regularly jumps from ~11 Thash/s to 15 Thash/s and vice versa.

Why would people power off their rigs like that so often? I see no reason for it to vary that much especially since it seems stable over courses of days and then finally takes a sharp increase or drop, indicating that it is not due to "part time" miners.

Or is the reported hashrate simply jumping around due to the luck and variance of the rate of solved blocks, while there is actually a stable constant hashrate for the most part?

Just a question of wonder.

 

[twobitcoins]

Or is the reported hashrate simply jumping around due to the luck and variance of the rate of solved blocks, while there is actually a stable constant hashrate for the most part?

Yes.

[BCMan]

Or is the reported hashrate simply jumping around due to the luck and variance of the rate of solved blocks, while there is actually a stable constant hashrate for the most part?

Yes.

Thats one some crazy variation with 40% increase, dont you think?

[twobitcoins]

Thats one some crazy variation with 40% increase, dont you think?

Block generation is supposed to average 144 per day.  Playing with a poisson distribution calculator, it seems there is about a 10% chance of a day with <= 128 blocks and a 10% chance of a day with >= 160 blocks.  That's already a 25% difference.

If you are looking at peaks and valleys in the graph lasting significantly less than a day, say 6 hours, then the average is 36 blocks with about a 10% chance of <= 28 and a 10% chance of >= 44, which is a 57% percent difference.

So yes, I think you are mostly seeing random variance.  (Hopefully I got the math right.)

[aqrulesms]

Thats one some crazy variation with 40% increase, dont you think?

Block generation is supposed to average 144 per day.  Playing with a poisson distribution calculator, it seems there is about a 10% chance of a day with <= 128 blocks and a 10% chance of a day with >= 160 blocks.  That's already a 25% difference.

If you are looking at peaks and valleys in the graph lasting significantly less than a day, say 6 hours, then the average is 36 blocks with about a 10% chance of <= 28 and a 10% chance of >= 44, which is a 57% percent difference.

So yes, I think you are mostly seeing random variance.  (Hopefully I got the math right.)

So, in that case, the actual network hash rate is around 12 Thash I believe?

[lemonz]

We could be seeing test runs of some serious ASICs?  We know BFL will be making an announcement soon regarding their BitForce SC and third generation of hardware.

I wouldn't be too surprised if BFLs first batch of circuits are made and are being benchmarked under different conditions before being announced.

[aqrulesms]

We could be seeing test runs of some serious ASICs?  We know BFL will be making an announcement soon regarding their BitForce SC and third generation of hardware.

I wouldn't be too surprised if BFLs first batch of circuits are made and are being benchmarked under different conditions before being announced.

That could be a possibility but do they actually have the capacity to increase the network by ~4-5 Thash/s? That's a lot of power, I'm sure that even though they're ASICs/FPGAs than 5 Thash would be at least 200-300 kW or more.

 

[lemonz]

That could be a possibility but do they actually have the capacity to increase the network by ~4-5 Thash/s? That's a lot of power, I'm sure that even though they're ASICs/FPGAs than 5 Thash would be at least 200-300 kW or more.

 

If true, we're only talking ASICs (not FPGAs).  Again, I would expect to see a performance jump like we saw from CPU to GPU along with the efficiency jump we saw going from GPU to FPGA.  Just a 3x increase in performance with a 3x decrease in power consumption could yield a board running a GHash/s on about 10 Watts (or 10KW per THash/s).

This is speculation on my part of course, but it would have to be significant enough to make BFL offer a previous generation obsolescence protection package.

[aqrulesms]

That could be a possibility but do they actually have the capacity to increase the network by ~4-5 Thash/s? That's a lot of power, I'm sure that even though they're ASICs/FPGAs than 5 Thash would be at least 200-300 kW or more.

 

If true, we're only talking ASICs (not FPGAs).  Again, I would expect to see a performance jump like we saw from CPU to GPU along with the efficiency jump we saw going from GPU to FPGA.  Just a 3x increase in performance with a 3x decrease in power consumption could yield a board running a GHash/s on about 10 Watts (or 10KW per THash/s).

This is speculation on my part of course, but it would have to be significant enough to make BFL offer a previous generation obsolescence protection package.

That would be quite incredible if the ASICs could actually have that much efficiency.

The problem is, once efficiency goes up, the difficulty will go up as well, which will eventually even out the playing field for profitability no matter how advanced the technological advancements are.

(Just like how CPU mining used to be as profitable as GPU mining is now due to prior lower difficulties)

[lemonz]

The problem is, once efficiency goes up, the difficulty will go up as well, which will eventually even out the playing field for profitability no matter how advanced the technological advancements are.

(Just like how CPU mining used to be as profitable as GPU mining is now due to prior lower difficulties)

I absolutely agree.  But, if true, it will mean that ASIC mining will become the only competitive form of mining.  We were fortunate to see GPU mining still somewhat competitive (maybe not in running costs, but in acquisition costs) during FPGAs market penetration.  I don't expect either GPUs or FPGAs to remain competitive once ASICs start getting delivered.  (I hope I'm wrong though, and that FPGAs remain competitive).

[aqrulesms]

The problem is, once efficiency goes up, the difficulty will go up as well, which will eventually even out the playing field for profitability no matter how advanced the technological advancements are.

(Just like how CPU mining used to be as profitable as GPU mining is now due to prior lower difficulties)

I absolutely agree.  But, if true, it will mean that ASIC mining will become the only competitive form of mining.  We were fortunate to see GPU mining still somewhat competitive (maybe not in running costs, but in acquisition costs) during FPGAs market penetration.  I don't expect either GPUs or FPGAs to remain competitive once ASICs start getting delivered.  (I hope I'm wrong though, and that FPGAs remain competitive).

I'm sure FPGAs would still work out, due to their low energy consumption.  My prediction is that ASICs won't become widespread and viable until late 2013.