A cautionary tail

[WizWom]

Anyone for a googolplex of gigabytes?
And just how big is that anyway? How many bytes?

2^30*10^10^10
~1.07x10^109 bytes.

And another question, "How long would it take for the fastest processor to write googolplex to high speed memory?" That is, without using scientific notation.

Assuming you could get memory as fast as L1 cache, each byte would take 1 cycle to write.
Even at an overclocked 4 GHz, and writing 8 bytes a cycle, that brings you to 3.1x10^88 seconds. There are 3.1x10^7 seconds in a year. That gives 1x10^81 years.

[krenshala]

Anyone for a googolplex of gigabytes?
And just how big is that anyway? How many bytes?

2^30*10^10^10
~1.07x10^109 bytes.

And another question, "How long would it take for the fastest processor to write googolplex to high speed memory?" That is, without using scientific notation.

Assuming you could get memory as fast as L1 cache, each byte would take 1 cycle to write.
Even at an overclocked 4 GHz, and writing 8 bytes a cycle, that brings you to 3.1x10^88 seconds. There are 3.1x10^7 seconds in a year. That gives 1x10^81 years.

That is a LOT of years ...

[pfrit]

Anyone for a googolplex of gigabytes?
And just how big is that anyway? How many bytes?

2^30*10^10^10
~1.07x10^109 bytes.

And another question, "How long would it take for the fastest processor to write googolplex to high speed memory?" That is, without using scientific notation.

Assuming you could get memory as fast as L1 cache, each byte would take 1 cycle to write.
Even at an overclocked 4 GHz, and writing 8 bytes a cycle, that brings you to 3.1x10^88 seconds. There are 3.1x10^7 seconds in a year. That gives 1x10^81 years.

I believe that that is incorrect. You should be able to store max int (system dependent) each cycle. Or you could set it to zero (should be one cycle with most archetectures) and subtract 1 (another cycle). But I could be wrong. My days of low level memory work are very far behind me. On some old systems, you were limited to the bus size for storing bits, but I think that those kind of systems are gone now. The real question is why would you want to do this?

[WizWom]

Anyone for a googolplex of gigabytes?
And just how big is that anyway? How many bytes?

2^30*10^10^10
~1.07x10^109 bytes.

And another question, "How long would it take for the fastest processor to write googolplex to high speed memory?" That is, without using scientific notation.

Assuming you could get memory as fast as L1 cache, each byte would take 1 cycle to write.
Even at an overclocked 4 GHz, and writing 8 bytes a cycle, that brings you to 3.1x10^88 seconds. There are 3.1x10^7 seconds in a year. That gives 1x10^81 years.

I believe that that is incorrect. You should be able to store max int (system dependent) each cycle. Or you could set it to zero (should be one cycle with most archetectures) and subtract 1 (another cycle). But I could be wrong. My days of low level memory work are very far behind me. On some old systems, you were limited to the bus size for storing bits, but I think that those kind of systems are gone now. The real question is why would you want to do this?

Yeah, it's "one word per cycle" and that's where 8 bytes comes in; that's a 64 bit processor, the current PC word.

[Soylent]

Yeah, it's "one word per cycle" and that's where 8 bytes comes in; that's a 64 bit processor, the current PC word.

That's an unwarranted assumption.

Modern processors are super-scalar(can process multiple instructions in parallel if they are independent).

Modern processors are pipelined(most instructions have a latency higher than 1 clock, but multiple instructions in different stages of completion can be in-flight at the same time).

Modern processors sometimes have ALUs that run at double the clock frequency of the rest of the chip(although this has gone somewhat out of fashion lately due to power concerns).

Modern processors have at least two CPU cores.

Processors have had 128-bit SIMD registers(e.g. SSE) since the late nineties and 64-bit SIMD registers(MMX) since the mid nineties.

The RAM and CPU are running at a different clock multiplier even in those cases where they run off the same clock.

RAM is arranged into multiple channels, commonly two channels for 128 bit wide data path, sometimes three channels for 192 bit wide data path, rarely 4(octacore server systems). Memory bandwidth is generally a great deal lower than what would be required to serve all cores if they were doing something extremely simple like filling the memory with zeroes. The CPU itself is not the limit; the limit is the bandwidth of the RAM which is about 20 to 40 GB/s.

Typical graphics cards have over 100 GB/s of bandwidth, with the high end dual chip cards having somewhere above 200 GB/s.

[Skipjack]

Typical graphics cards have over 100 GB/s of bandwidth, with the high end dual chip cards having somewhere above 200 GB/s.

That depends on how you look at it. It is still only 100 GB/s per chip.

[WizWom]

BS

You just like to argue, don't you?
Even 200GB/s going into a multiplexed RAM would leave you just a factor of 50 faster, or 2x10^79 years.