Quantum volume: One number to benchmark a quantum computer

Quantum computing benchmark looks at fundamental limit of computer, not speed.

Read the whole story

 

[gamerk2]

Correct me if I'm wrong, but if you have a 1% error rate (for example), then wouldn't quantum computers need to be at least 2x as powerful as current ones, as you need to perform each computation at least twice to confirm the end result is correct? Because speaking as a SW Engineer: 1% error rate is enough to make sure your PC doesn't even boot, let alone get you to your desktop.

 

[niwax]

Correct me if I'm wrong, but if you have a 1% error rate (for example), then wouldn't quantum computers need to be at least 2x as powerful as current ones, as you need to perform each computation at least twice to confirm the end result is correct? Because speaking as a SW Engineer: 1% error rate is enough to make sure your PC doesn't even boot, let alone get you to your desktop.

This shouldn't play much of a role considering the way quantum computers are used. They are very efficient at certain tasks that traditional computers are bad at. That especially includes calculations with easy to check results. A popular example would be factorizing a number. It's incredibly hard to do on a traditional computer but the result is ridiculously easy to check - just multiply them. This is also the reason why certain crypto applications can be broken by quantum computers: The assumption that a calculation is very fast one way but really hard the other way doesn't hold.

 

[steinah]

I agree with Niwax. It seems like future traditional computers will have quantum add-ons (sort of like graphics cards) for these type of computations. You wouldn't want to rely on them for loading the whole OS or dealing with sensitive files, where even a 0.001% error rate could be catastrophic.

 

[nexusofjoseph]

Can anyone comment as to why the chip has the control wires going in very non-direct paths to the qubits? I'm guessing they are designed to be the same length in order to better manage the synchronisation of instructions to the qubits, but I would hope someone could say more about it than I can.

 

[ytene]

I can certainly understand the observation that understanding the relative performance of different quantum computers can be difficult.

However, this is something that conventional computing solved quite some time ago: it's called a benchmark.

Extrapolating, could we not develop a small selection of Q-computing-relevant problems and have each Q-computer tackle them? I appreciate that single problem is unreasonable given the anticipation of slightly different design priorities. However, a range of [for example] cryptographic problems to factor might present us with a workable model.

Ultimately, the value of a Q-computer to the world at large will be derived from the ability of that computer to solve world-relevant problems. A generic benchmark seems like a workable approach.

Am I missing something?

 

[Starbuck_13]

I can certainly understand the observation that understanding the relative performance of different quantum computers can be difficult.

However, this is something that conventional computing solved quite some time ago: it's called a benchmark.

Extrapolating, could we not develop a small selection of Q-computing-relevant problems and have each Q-computer tackle them? I appreciate that single problem is unreasonable given the anticipation of slightly different design priorities. However, a range of [for example] cryptographic problems to factor might present us with a workable model.

Ultimately, the value of a Q-computer to the world at large will be derived from the ability of that computer to solve world-relevant problems. A generic benchmark seems like a workable approach.

Am I missing something?

I could be wrong, but given that there's the inherent possibility that a quantum computer will get the solution to a problem wrong, I don't know that a standard benchmark could work. I mean, we have a benchmark for what CPUs are capable of because we have the basic understanding that it's always going to be correct, it's just a question of how long it takes to get there. With quantum computing introducing the variable of incorrect answers, you have to take that into account. Maybe you can run the benchmark you're talking about 100 times and average the results, but I think this idea of quantum volume is a simpler - or at least faster - way of looking at capability.

 

[Pubert]

I'll be the first to admit that I think quantum computing exists solely for the purpose of making my brain hurt.

Some bar graphs illustrating the projected limits of computer speed would've been nice.

 

[BinaryMuppet]

I agree with Niwax. It seems like future traditional computers will have quantum add-ons (sort of like graphics cards) for these type of computations. You wouldn't want to rely on them for loading the whole OS or dealing with sensitive files, where even a 0.001% error rate could be catastrophic.

I wonder if they'll every make a hybrid board that supports the traditional CPU along with a built-in Quantum chip rather than a PCI slot?

 

[niwax]

I'll be the first to admit that I think quantum computing exists solely for the purpose of making my brain hurt.

Some bar graphs illustrating the projected limits of computer speed would've been nice.

There are some lovely easy to understand aspects though. My favorite is this one: An interesting property of ice is that its molecules are almost perfectly optimally ordered due to the gradual effects during the cooling process that allow smaller and smaller optimizations. By applying the same gradually smaller steps to traditional optimization problems, simulated annealing was created. We have now come full circle and implement those algorithms by wiring a quantum computer to represent the problem and actually cooling it down.

 

[tripodal]

"There is insufficient quantum volume to complete this task" sounds like something out of a terrible sci-fi movie.

I love it.

 

[tie]

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

 

[matthewslyman]

I have a traditional computer science degree, and I'm confused by some of this article. I do believe that IBM is doing worthwhile research here that will eventually produce interesting commercial applications. Some parts of this article make heuristic sense, but others either don't parse correctly, or trip my "scientific snake-oil" detector. Examples:

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

Unfortunately, if you compare the calculated error rate, the number of qubits and the quantum volume, the results are inconsistent. We've reached out to IBM and will update when they respond.

At least I have more confidence in what IBM is doing in the "quantum" space, than I have in what D-Wave is doing (largely because IBM are not letting their marketing department off their leash just yet!) Yet in some of my earliest comments on Ars Technica about quantum computing, I questioned the idea that quantum computers could somehow magically break free from the fundamental mathematical constraints that classical computers are bound by… I have always believed that probabilistic errors in quantum computing will essentially constrain quantum computing to a similar technological trajectories as we find in classical computing — where the quantum computing industry would have its equivalent of Moore's Law, and its equivalents of O complexity. (We might compare what I consider to be the false promise, or marketing hype, of quantum computing; with the false promise of nuclear energy as sold to the tax-paying public: "unlimited pollution-free energy". Yet despite knowing this has always been a lie in the past, we keep returning to this marketing theme, as we are doing now with fusion power technology…)

It's one thing to believe in quantum computing breaking cryptography overnight, because some equations & algorithms tell you this might happen. But in the practical process of researching how to make a quantum computer, we should not be surprised to find facets of the problem that we had not fully anticipated — or, that "the devil is in the details". We should not be surprised to find that quantum computing is "merely" a useful technology beyond the CMOS end-point, with potentially far greater computational density and power (representing "only" a high-proportional improvement over CMOS — with a special bonus for "small" calculations requiring only a few qubits)…

 

[cse84]

I'll be the first to admit that I think quantum computing exists solely for the purpose of making my brain hurt.

Some bar graphs illustrating the projected limits of computer speed would've been nice.

There are some lovely easy to understand aspects though. My favorite is this one: An interesting property of ice is that its molecules are almost perfectly optimally ordered due to the gradual effects during the cooling process that allow smaller and smaller optimizations. By applying the same gradually smaller steps to traditional optimization problems, simulated annealing was created. We have now come full circle and implement those algorithms by wiring a quantum computer to represent the problem and actually cooling it down.

Simulated annealing is a very nice concept, but simulated annealing is what D-Wave's pseudo-quantum-computer is doing. A general purpose quantum computer is more powerful and does not rely on annealing. As far as I understand it, this new IBM system is a general purpose quantum computer.

 

[niwax]

I'll be the first to admit that I think quantum computing exists solely for the purpose of making my brain hurt.

Some bar graphs illustrating the projected limits of computer speed would've been nice.

There are some lovely easy to understand aspects though. My favorite is this one: An interesting property of ice is that its molecules are almost perfectly optimally ordered due to the gradual effects during the cooling process that allow smaller and smaller optimizations. By applying the same gradually smaller steps to traditional optimization problems, simulated annealing was created. We have now come full circle and implement those algorithms by wiring a quantum computer to represent the problem and actually cooling it down.

Simulated annealing is a very nice concept, but simulated annealing is what D-Wave's pseudo-quantum-computer is doing. A general purpose quantum computer is more powerful and does not rely on annealing. As far as I understand it, this new IBM system is a general purpose quantum computer.

Yes, that's another reason why they're so hard to compare and benchmark

 

[cse84]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.

Edit: spelling

 

[Statistical]

I agree with Niwax. It seems like future traditional computers will have quantum add-ons (sort of like graphics cards) for these type of computations. You wouldn't want to rely on them for loading the whole OS or dealing with sensitive files, where even a 0.001% error rate could be catastrophic.

I wonder if they'll every make a hybrid board that supports the traditional CPU along with a built-in Quantum chip rather than a PCI slot?

Well given that all quantum processors operate at near absolute zero I doubt it. It doesn't take much thermal noise to flip a qubit. The idea that you will have a general PC with a quantum coprocessor is doubtful.

To give you an idea the tiny chip in the article photo is inside the giant cylinder in this photo. Most of that volume is insulation and successive layers of cryocoolers to bring the temp at the chip to near absolute zero. That entire rack of equipment to the right is the additional equipment needed to run the series of cryocoolers.

You may someday have your always online computer (even one that fits in your pocket) send a request off to a quantum computer in the cloud. Even that would require costs to come down by probably six orders of magnitude.

 

[Statistical]

I can certainly understand the observation that understanding the relative performance of different quantum computers can be difficult.

However, this is something that conventional computing solved quite some time ago: it's called a benchmark.

Extrapolating, could we not develop a small selection of Q-computing-relevant problems and have each Q-computer tackle them? I appreciate that single problem is unreasonable given the anticipation of slightly different design priorities. However, a range of [for example] cryptographic problems to factor might present us with a workable model.

Ultimately, the value of a Q-computer to the world at large will be derived from the ability of that computer to solve world-relevant problems. A generic benchmark seems like a workable approach.

Am I missing something?

Because quantum computer isn't classic computing. Speed is kinda irrelevant. Assuming error is kept under control and you have a large enough qubits you will solve problems no matter the complexity essentially instantly and never be able to solve problems beyond the limits of the system no matter how much time is given.

As an example IBM's 16 qubit computer (assumming error is reasonably low) could implement Shor's algorithm to factor the product of 16 bit primes essentially instantly. However it could never factor the product of 17 bit primes no matter how much time is given.

 

[bldkcstark]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.

Edit: spelling

Also, sometimes it just isn't necessary to include the units. I do however want to volunteer "Qberts" as the standardized unit for this measurement.

I think a 30 qubit system that attains a score of 900 Qberts is the goal here.

 

[Statistical]

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

 

[tricorn]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are a lot of Dimensionless Quantities that have real meaning.

 

[matthewslyman]

cse84[/url]":332mkoxh]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.
Edit: spelling

I'm afraid that your explanation is partly beyond me, but based on the part I understand (and the studies I've done), I might suggest that even unitless quantities are not always dimensionless, directionless and scalar. For example, even if you divide seconds by seconds, you still end up with a proportion of time — and it can still be useful to retain that conceptual label on a quantity that at face value, represents nothing more than a raw number!

So even with your explanation, this is still tripping my "scientific snake-oil" detector — since too many quantities in this article are essentially unitless, or heuristically assigned some nebulous conceptual unit…

Some of this gap in understanding comes down to a lack of necessary education on my part. However, I do believe that a lot of progress in mathematics and physical sciences may be made through reforms that emphasize the application of units and unit-like dimensional and directional (perhaps i-j-k style or polar-equivalent) annotations…

 

[Lacedaemon]

edit: ninja'd

 

[cse84]

cse84[/url]":2z1e3h34]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.
Edit: spelling

I'm afraid that your explanation is partly beyond me, but based on the part I understand (and the studies I've done), I might suggest that even unitless quantities are not always dimensionless, directionless and scalar. For example, even if you divide seconds by seconds, you still end up with a proportion of time — and it can still be useful to retain that conceptual label on a quantity that at face value, represents nothing more than a raw number!

So even with your explanation, this is still tripping my "scientific snake-oil" detector — since too many quantities in this article are essentially unitless, or heuristically assigned some nebulous conceptual unit…

Some of this gap in understanding comes down to a lack of necessary education on my part. However, I do believe that a lot of progress in mathematics and physical sciences may be made through reforms that emphasize the application of units and unit-like dimensional and directional (perhaps i-j-k style or polar-equivalent) annotations…

I agree that a dimensionless number still has a meaning that is not interchangeable with the meaning of other dimensionless numbers. Unfortunately many theoretical physicists leave out units and even many constant factors because of the convenience of having a shorter, easier to read (for an expert in the field at least) formula - you might call that laziness instead of convenience, if you like. It might also have to do with the fact that all of our measurement systems choose their units more or less arbitrarily.

By the way, being scalar or non-scalar is a concept that is different from having a unit of measurement or having none. You can have scalar quantities with or without units and you can have non-scalar quantities (i.e. vector/tensor/whatever quantities) with or without units. It is unfortunate that the term "dimensionless" is often used instead of "without measurement unit", because it can easily be confused with "scalar".

 

[matthewslyman]

tricorn[/url]":hpqljkj6]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are a lot of Dimensionless Quantities that have real meaning.

Perhaps. But with many such quantities, it could be argued that they are only "dimensionless" because our current system of units & dimensions is inadequate… Interestingly, each of these "dimensionless quantities" has a name — one might even use this name effectively as a unit, in a system which has rules for combining them (where the rules might be governed partly by consideration of these units as "compound units"). The adequacy or inadequacy of a system of units, might be determined by the requirements of the application…

cse84[/url]":hpqljkj6]

cse84[/url]":hpqljkj6]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.
Edit: spelling

I'm afraid that your explanation is partly beyond me, but based on the part I understand (and the studies I've done), I might suggest that even unitless quantities are not always dimensionless, directionless and scalar. For example, even if you divide seconds by seconds, you still end up with a proportion of time — and it can still be useful to retain that conceptual label on a quantity that at face value, represents nothing more than a raw number!

So even with your explanation, this is still tripping my "scientific snake-oil" detector — since too many quantities in this article are essentially unitless, or heuristically assigned some nebulous conceptual unit…

Some of this gap in understanding comes down to a lack of necessary education on my part. However, I do believe that a lot of progress in mathematics and physical sciences may be made through reforms that emphasize the application of units and unit-like dimensional and directional (perhaps i-j-k style or polar-equivalent) annotations…

I agree that a dimensionless number still has a meaning that is not interchangeable with the meaning of other dimensionless numbers. Unfortunately many theoretical physicists leave out units and even many constant factors because of the convenience of having a shorter, easier to read (for an expert in the field at least) formula - you might call that laziness instead of convenience, if you like. It might also have to do with the fact that all of our measurement systems choose their units more or less arbitrarily.

By the way, being scalar or non-scalar is a concept that is different from having a unit of measurement or having none. You can have scalar quantities with or without units and you can have non-scalar quantities (i.e. vector/tensor/whatever quantities) with or without units. It is unfortunate that the term "dimensionless" is often used instead of "without measurement unit", because it can easily be confused with "scalar".

Agreed — in part, this was my point — that "unitless" does not imply "scalar", nor does it imply interchangeability with other "unitless" quantities.

 

[ssiu]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.

Edit: spelling

Also, sometimes it just isn't necessary to include the units. I do however want to volunteer "Qberts" as the standardized unit for this measurement.

I think a 30 qubit system that attains a score of 900 Qberts is the goal here.

It's got to be spelled Q*berts instead of Qberts.

Also correlates nicely with the sentiment "I don't @!*?@! understand what this unit represent". Q*bert will approve

 

[tie]

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless. The gates are not 99% accurate like this article says. They seem to be closer to 90-95% accurate (or maybe the measurements are only 85% accurate). It's really bad. Fun, but you can't do much.

 

[Statistical]

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless.

https://www.research.ibm.com/ibm-q/

Two new processors
IBM Q has successfully built and tested two of its most powerful universal quantum computing processors to date: 16 qubits for public use and a 17 qubit prototype commercial processor.

 

[tie]

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless.

https://www.research.ibm.com/ibm-q/

Two new processors
IBM Q has successfully built and tested two of its most powerful universal quantum computing processors to date: 16 qubits for public use and a 17 qubit prototype commercial processor.

Thank you.

 

[Rotoars]

Excellent article Chris!

 

[BrainsOnToast]

I agree with Niwax. It seems like future traditional computers will have quantum add-ons (sort of like graphics cards) for these type of computations. You wouldn't want to rely on them for loading the whole OS or dealing with sensitive files, where even a 0.001% error rate could be catastrophic.

I wonder if they'll every make a hybrid board that supports the traditional CPU along with a built-in Quantum chip rather than a PCI slot?

Well given that all quantum processors operate at near absolute zero I doubt it. It doesn't take much thermal noise to flip a qubit. The idea that you will have a general PC with a quantum coprocessor is doubtful.

To give you an idea the tiny chip in the article photo is inside the giant cylinder in this photo. Most of that volume is insulation and successive layers of cryocoolers to bring the temp at the chip to near absolute zero. That entire rack of equipment to the right is the additional equipment needed to run the series of cryocoolers.

You may someday have your always online computer (even one that fits in your pocket) send a request off to a quantum computer in the cloud. Even that would require costs to come down by probably six orders of magnitude.

I know it's apocryphal, but there's a world market for maybe 5 computers?

https://en.wikipedia.org/wiki/Thomas_J. ... ttribution

Which I just looked up on the tiny computer that lives in my pocket off a battery that lasts all day.

 

[VisualSeed]

I agree with Niwax. It seems like future traditional computers will have quantum add-ons (sort of like graphics cards) for these type of computations. You wouldn't want to rely on them for loading the whole OS or dealing with sensitive files, where even a 0.001% error rate could be catastrophic.

I think you would just have a system with a quantum co-processor (or probably early versions would be an external rig) like we used to have math coprocessors. That is basically what the original Cray super computers were. Big, expensive, external math co-processors.

 

[Ibmresearch]

The upgraded 16 qubit chip is available via our SDK: The details are here: https://developer.ibm.com/code/2017/05/ ... iskit-sdk/

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless.

https://www.research.ibm.com/ibm-q/

Two new processors
IBM Q has successfully built and tested two of its most powerful universal quantum computing processors to date: 16 qubits for public use and a 17 qubit prototype commercial processor.

Thank you.

 

[Ibmresearch]

Access to the 16 qubit chip is available via the SDK. Details are here: https://developer.ibm.com/code/2017/05/ ... iskit-sdk/

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless.

https://www.research.ibm.com/ibm-q/

Two new processors
IBM Q has successfully built and tested two of its most powerful universal quantum computing processors to date: 16 qubits for public use and a 17 qubit prototype commercial processor.

 

[Ibmresearch]

Sure, benchmarks are standard, but consider what has happened to the TOP500 benchmark. While it initially provided great value, now teams who submit to it have rigged their systems to do very well, but not necessarily for real world applications. In addition, it ignored speed to result and energy efficiency to result. More can be read here: https://www.ibm.com/blogs/research/2014 ... fficiency/

I can certainly understand the observation that understanding the relative performance of different quantum computers can be difficult.

However, this is something that conventional computing solved quite some time ago: it's called a benchmark.

Extrapolating, could we not develop a small selection of Q-computing-relevant problems and have each Q-computer tackle them? I appreciate that single problem is unreasonable given the anticipation of slightly different design priorities. However, a range of [for example] cryptographic problems to factor might present us with a workable model.

Ultimately, the value of a Q-computer to the world at large will be derived from the ability of that computer to solve world-relevant problems. A generic benchmark seems like a workable approach.

Am I missing something?

Because quantum computer isn't classic computing. Speed is kinda irrelevant. Assuming error is kept under control and you have a large enough qubits you will solve problems no matter the complexity essentially instantly and never be able to solve problems beyond the limits of the system no matter how much time is given.

As an example IBM's 16 qubit computer (assumming error is reasonably low) could implement Shor's algorithm to factor the product of 16 bit primes essentially instantly. However it could never factor the product of 17 bit primes no matter how much time is given.

 

[eideticex]

I can certainly understand the observation that understanding the relative performance of different quantum computers can be difficult.

However, this is something that conventional computing solved quite some time ago: it's called a benchmark.

Extrapolating, could we not develop a small selection of Q-computing-relevant problems and have each Q-computer tackle them? I appreciate that single problem is unreasonable given the anticipation of slightly different design priorities. However, a range of [for example] cryptographic problems to factor might present us with a workable model.

Ultimately, the value of a Q-computer to the world at large will be derived from the ability of that computer to solve world-relevant problems. A generic benchmark seems like a workable approach.

Am I missing something?

That was kind of my thought and developed further down the line while reading the article. What came to mind was the choose function (also known as ab function). It's an interesting mathematical function for a few reasons but the most notable being it's very easy to derive Pascal's triangle from. It works a bit like this:

Definition: (A + B)^N

N=1 - (A + B)^1 = 1A + 1B
N=2 - (A + B)^2 = (A + B) * (A + B) = A^2 + AB^2 + BA^2 + B^2 = 1A^2 + 2AB^2 + 1B^2

If you look at the multipliers, you get Pascal's triangle. It doubles in complexity for each successive iteration of N. Results are very easy to glance for correctness but can be automatically checked with a transistor based computer.

But I think we are both missing something important as to why this is a bad approach to benchmark these computers.

 

[Pewmaymen]

tricorn[/url]":18vaveay]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are a lot of Dimensionless Quantities that have real meaning.

Perhaps. But with many such quantities, it could be argued that they are only "dimensionless" because our current system of units & dimensions is inadequate… Interestingly, each of these "dimensionless quantities" has a name — one might even use this name effectively as a unit, in a system which has rules for combining them (where the rules might be governed partly by consideration of these units as "compound units"). The adequacy or inadequacy of a system of units, might be determined by the requirements of the application…

cse84[/url]":18vaveay]

cse84[/url]":18vaveay]

…a 30-qubit system with no gate errors has a quantum volume of 900 (no units for this).

— A measurement without units is, in my opinion, a measurement we don't understand. It's frequently a fudge-factor, or, an excuse for not fully understanding the shape & size, or the dimensions of the problem we are trying to solve.

There are quantities that are truly unitless, because units can be multiplied and divided like numbers and for some quantities the units simply cancel out. For example the fine-structure constant is unitless (which is very helpful - it will have the same numerical value in all measurement systems). Also angles are unitless (length of arc divided by length of radius).

two-qubit operation

— What is an "operation" in quantum-computing land? Where is this explained? This could just be my ignorance here…

It is basically multiplication of a complex vector that represents the internal state of the quantum computer by a complex matrix (a unitary matrix, if i remember correctly). That's what I recollect from the introductory quantum computing course I took in university.
Edit: spelling

I'm afraid that your explanation is partly beyond me, but based on the part I understand (and the studies I've done), I might suggest that even unitless quantities are not always dimensionless, directionless and scalar. For example, even if you divide seconds by seconds, you still end up with a proportion of time — and it can still be useful to retain that conceptual label on a quantity that at face value, represents nothing more than a raw number!

So even with your explanation, this is still tripping my "scientific snake-oil" detector — since too many quantities in this article are essentially unitless, or heuristically assigned some nebulous conceptual unit…

Some of this gap in understanding comes down to a lack of necessary education on my part. However, I do believe that a lot of progress in mathematics and physical sciences may be made through reforms that emphasize the application of units and unit-like dimensional and directional (perhaps i-j-k style or polar-equivalent) annotations…

I agree that a dimensionless number still has a meaning that is not interchangeable with the meaning of other dimensionless numbers. Unfortunately many theoretical physicists leave out units and even many constant factors because of the convenience of having a shorter, easier to read (for an expert in the field at least) formula - you might call that laziness instead of convenience, if you like. It might also have to do with the fact that all of our measurement systems choose their units more or less arbitrarily.

By the way, being scalar or non-scalar is a concept that is different from having a unit of measurement or having none. You can have scalar quantities with or without units and you can have non-scalar quantities (i.e. vector/tensor/whatever quantities) with or without units. It is unfortunate that the term "dimensionless" is often used instead of "without measurement unit", because it can easily be confused with "scalar".

Agreed — in part, this was my point — that "unitless" does not imply "scalar", nor does it imply interchangeability with other "unitless" quantities.

I think it isn't saying there isn't a unit per se. I think it is saying that there isn't a defined unit YET. That is, this is a proposal for a method that attempts to create a way to encompass a quantum computer into a single integer value so that you can compare two different computers. Currently, the "unit" is a single output number. If the process becomes somewhat accepted, the groups that accept it can propose to and agree upon a name for the unit (such as the afore-mentioned "Q*bert"). I mean, we call it a Joule, but what did Joule call it?

 

[kaleberg]

So, the damned thing is going to run and give us an answer of 42 with a probability of 99.9999%. As an answer for life, the universe and everything, that just isn't good enough.

 

[Noreastern]

The upgraded 16 qubit chip is available via our SDK: The details are here: https://developer.ibm.com/code/2017/05/ ... iskit-sdk/

Where is this "newly announced" 17-qubit device? Why is there no link? Google isn't finding any announcements. On IBM's website, I don't see any news posted, and you are still limited to 5-qubit experiments.

At about 64 qubits current cryptology is dead. At 100x100 qubit arrays who know what will happen. AI? I certainly have no idea.

Maybe because you searched for 17-qubit not 16?

https://phys.org/news/2017-05-ibm-power ... ssors.html

No, I saw that page. It also doesn't link to anything from IBM.

And if you go to IBM's page, they still only give you a five-qubit machine to play with, not 16. I think it is fairly useless.

https://www.research.ibm.com/ibm-q/

Two new processors
IBM Q has successfully built and tested two of its most powerful universal quantum computing processors to date: 16 qubits for public use and a 17 qubit prototype commercial processor.

Thank you.

 

[Noreastern]

No one at this point knows where quantum computing is headed. The most powerful truly functional quantum computer today is the 16 bit IBM prototype. Given the mathematics underpinning quantum computing their processing power goes up by nearly an order of magnitude with every doubling of the number of qubits. God only knows where large qubit arrays will take us.