How many FLOP/s is there in a gigahash?

[mackminer]

In fact, what does gigahash mean exactly?

Also without getting into much technical detail can anyone explain what it means when you say you are generating a hash?

Thanks.

[DeathAndTaxes]

A "hash" is actually 2 SHA-256 hashing operations so.

Bitcoin hash = SHA-256(SHA-256(blockheader including nonce)

A Gigahash = 1 billion hashes.

To perform one bitcoin hash on an AMD GPU takes about 3500 ALU operations.

There is no direct conversion between FLOPS (floating point operations per second) and hashes because hashes involve integer operations (whole number math) and FLOPS involving obviously floating point math.  Any claim of Bticoin is x flops is simply an aproximation of the potential floating point power as the Btitcoin network uses 0.00 FLOPS.

Various sites come up with an estimate by on the number of FLOPS per hash by looking at how "fast" average hardware is on each.

Example:
AMD 5870 5970(stock clocks)
~ 700MH/s
~ 2700 GLOPS (single precision)
~ 5340 GFLOPS (double precision)

http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#PCIe_.28HD_5xxx.29

Using that you can get an estimated relationship between hashes and flops and then use networks hashing rate to estimate network's potential in TFLOPS.

[ngzhang]

approx．1300 32bit add operaions.

[finway]

According to bitcoinwatch.com

1 Hash = 12,697 FLOPs

[gmaxwell]

According to bitcoinwatch.com
1 Hash = 12,697 FLOPs

Thats a gibberish number. It is like saying that one mile per hour is equal to 14.285 pounds (because some car that weighs 2000 lbs can go 140 MPH).

There are no floating point operations involved in the calculation of a bitcoin hash. None at all. 0 FLOPS.

Ngzhang says above that there are about 1300 32-bit operations (in one bitcoin hash) which sounds about right. So one 1GH/s would be equal to 1,300,000,000,000 operations per second.

If you're looking to, e.g. compare bitcoin's speed with various supercomputers you can just figure out how fast the supercomputer could mine.  For example, one of the top machines is the Jaguar supercomputer which uses fairly standard opteron cores. I think my fairly boring rough numbers put it at about 522 GH/s.  (pshaw, who says CPU mining is useless? )

[DiabloD3]

According to bitcoinwatch.com
1 Hash = 12,697 FLOPs

Thats a gibberish number. It is like saying that one mile per hour is equal to 14.285 pounds (because some car that weighs 2000 lbs can go 140 MPH).

There are no floating point operations involved in the calculation of a bitcoin hash. None at all. 0 FLOPS.

Ngzhang says above that there are about 1300 32-bit operations (in one bitcoin hash) which sounds about right. So one 1GH/s would be equal to 1,300,000,000,000 operations per second.

If you're looking to, e.g. compare bitcoin's speed with various supercomputers you can just figure out how fast the supercomputer could mine.  For example, one of the top machines is the Jaguar supercomputer which uses fairly standard opteron cores. I think my fairly boring rough numbers put it at about 522 GH/s.  (pshaw, who says CPU mining is useless? )

I agree. Not only that, FLOPs aren't all made equal. A lot of people have done stuff like specifically state P4 flops, which has a strict conversion factor comparing something along the lines of P4s doing SSE MADs to the same instruction in integer math, etc etc etc.

I like my computational power metric better: number of 5850s at stock speed.

[ElectricMucus]

For integer operations there is the figure of MIPS and GIPS or GOPs, as for integer instructions.
Floating point performance is only related to the clock speed not integer performance.

[finway]

For example, one of the top machines is the Jaguar supercomputer which uses fairly standard opteron cores. I think my fairly boring rough numbers put it at about 522 GH/s.

So, how much FLOPs/s is it ?

[Matthew N. Wright]

I'm curious, how many giggadicks are in a dickaflop and how many pedophiles are in a pedoflop too?

[Jonathan Ryan Owens]

I'm curious, how many giggadicks are in a dickaflop and how many pedophiles are in a pedoflop too?

I can't find any information on #2, but I did find this:

(1) http://pastebin.com/RQe0S9Pi

[Matthew N. Wright]

I'm curious, how many giggadicks are in a dickaflop and how many pedophiles are in a pedoflop too?

I can't find any information on #2, but I did find this:

(1) http://pastebin.com/RQe0S9Pi

That is a very reputable source.

[BinaryMage]

A "hash" is actually 2 SHA-256 hashing operations so.

Bitcoin hash = SHA-256(SHA-256(blockheader including nonce)

A Gigahash = 1 billion hashes.

To perform one bitcoin hash on an AMD GPU takes about 3500 ALU operations.

There is no direct conversion between FLOPS (floating point operations per second) and hashes because hashes involve integer operations (whole number math) and FLOPS involving obviously floating point math.  Any claim of Bticoin is x flops is simply an aproximation of the potential floating point power as the Btitcoin network uses 0.00 FLOPS.

Various sites come up with an estimate by on the number of FLOPS per hash by looking at how "fast" average hardware is on each.

Example:
AMD 5870 (stock clocks)
~ 700MH/s
~ 2700 GLOPS (single precision)
~ 5340 GFLOPS (double precision)

http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#PCIe_.28HD_5xxx.29

Using that you can get an estimated relationship between hashes and flops and then use networks hashing rate to estimate network's potential in TFLOPS.

A HD 5870 puts out ~380 MH/s with stock clocks, not ~700. Where did you get that number from?

Yes, you can get a relationship, though it varies card-to-card. My suggestion would be to find the relationships for the 5870, 5850, and 5830, as I suspect those are the most common mining cards, though I could be wrong, and average those together. Then you can use the network hash rate to find the approximate FLOPS as stated above.

[DeathAndTaxes]

A HD 5870 puts out ~380 MH/s with stock clocks, not ~700. Where did you get that number from?

Typo brain said 5970 keyboard said 5870.