Re: Is Intel exceptionally unsuccessful as an architecture designer?

David Brown wrote:

On 19/09/2024 09:44, Niklas Holsti wrote:

On 2024-09-19 2:47, Lawrence D'Oliveiro wrote:

On Wed, 18 Sep 2024 20:09:53 GMT, Anton Ertl wrote:
>

He mentioned that several physics breakthroughs
are needed for quantum computing to become useful.

>
The biggest one would be getting around the fundamental problem that you
can’t get something for nothing.

>
>
Stupid argument. Look at the effort and tech it takes to make quantum computers... that is not "nothing".
>
>

The promise of an exponential increase in computing power for a linear
increase in the number of processing elements sounds very much like
“something for nothing” under another name, wouldn’t you say?

>
>
No, it is exploiting the very non-intuitive nature of quantum entanglement to create an exponential number of collective states of a linear number of elements. Medieval arguments about "nothing" vs "something" don't work there.
>

 Quantum computing certainly gives you some tricks that are hard to replicate with classical computers.  (And of course some quantum effects are impossible to replicate classically, but those are not actually computations.)
 But it is still ultimately limited in many ways.  Landauer's principle about the minimal energy costs of calculations applies equally to quantum calculations.
 The practical limitations for quantum computers are far more significant.  Roughly speaking, when you entangle more states at once, you need tighter tolerances to maintain coherence, which translates to lower temperatures, higher energy costs, and lower times to do your calculations.  And to be useful, you need large numbers of qubits, which again makes maintaining coherence increasingly difficult.
 I'm sure that there will be breakthroughs that improve some of this, but I am not holding my breath - I don't believe quantum computers will ever be cost-effective for anything but a few very niche problems.  Currently they have only beat classical computers in tasks that involve simulating some quantum effects.  That's a bit like noticing that soap bubble computers are really good at solving 2D minimal energy surface problems.
 Remember, the current record for Shor's algorithm is factorising 21 into 3 x 7.  Factorising 35 is still beyond current engineering levels.
 

 From my recent reading, it seems like factoring 21 (5 bits) requires at least 5+10=15 bits all staying entangled, plus a number of additional bits for error correction. I'm guessing you also need some extra bits/redundancy in order to successfully read out the results?
Getting to at the very least 3K entangled bits in order to speed up RSA 1024 decryption will certainly be out of the question for the remainder of my professional career, and most probably also the rest of my life.
Terje
--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"