Startups are Using GPUs to Simulate Quantum Computers (reuters.com) 23
Lacking quantum computers, some startups "are developing a new breed of software inspired by algorithms used in quantum physics..." reports Reuters.
"Once too big for conventional computers, these algorithms are finally being put to work thanks to today's powerful artificial intelligence chips, industry executives told Reuters." QC Ware, a software startup that has raised more than $33 million and initially focused only on software that could run on quantum computers, said it needed to change tack and find a solution for clients today until the future quantum machines arrive. So QC Ware CEO Matt Johnson said it turned to Nvidia Corp's graphic processing units (GPU) to "figure out how can we get them something that is a big step change in performance ... and build a bridge to quantum processing in the future...."
This week, QC Ware is unveiling a quantum-inspired software platform called Promethium that will simulate chemical molecules — to see how they interact with things like protein — on a traditional computer using GPUs. The software can cut simulation time from hours to minutes for molecules of 100 atoms, and months to hours for molecules of up to 2000 atoms, compared with existing software solutions, said QC Ware's head of quantum chemistry Robert Parrish...
In the past 18 months, quantum software startups including SandBoxAQ — an Alphabet spinoff — raised about $1 billion, according to data firm PitchBook. To be sure, development of this technology is nascent and these startups must work hard to convince some prospective clients. SandBoxAQ CEO Jack Hidary said it was only 24 months ago that AI chips became powerful enough to simulate hundreds of thousands of chemical interactions simultaneously. It developed a quantum-inspired algorithm for biopharma simulation on Google's AI chip called a Tensor Processing Unit (TPU)...
"Once too big for conventional computers, these algorithms are finally being put to work thanks to today's powerful artificial intelligence chips, industry executives told Reuters." QC Ware, a software startup that has raised more than $33 million and initially focused only on software that could run on quantum computers, said it needed to change tack and find a solution for clients today until the future quantum machines arrive. So QC Ware CEO Matt Johnson said it turned to Nvidia Corp's graphic processing units (GPU) to "figure out how can we get them something that is a big step change in performance ... and build a bridge to quantum processing in the future...."
This week, QC Ware is unveiling a quantum-inspired software platform called Promethium that will simulate chemical molecules — to see how they interact with things like protein — on a traditional computer using GPUs. The software can cut simulation time from hours to minutes for molecules of 100 atoms, and months to hours for molecules of up to 2000 atoms, compared with existing software solutions, said QC Ware's head of quantum chemistry Robert Parrish...
In the past 18 months, quantum software startups including SandBoxAQ — an Alphabet spinoff — raised about $1 billion, according to data firm PitchBook. To be sure, development of this technology is nascent and these startups must work hard to convince some prospective clients. SandBoxAQ CEO Jack Hidary said it was only 24 months ago that AI chips became powerful enough to simulate hundreds of thousands of chemical interactions simultaneously. It developed a quantum-inspired algorithm for biopharma simulation on Google's AI chip called a Tensor Processing Unit (TPU)...
BS Computers (Score:2)
Re: BS Computers (Score:2)
I find it unlikely that we will need quantum computers capable of many qubits simultaneously to accomplish what people are hoping from QC's at this point.
IBM made one with 127 qubits 'the first capable of calculating 3 digit numbers.'
at my level of ignorance, it seems like it should be possible to use fewer qubits in combination with traditional cpu/GPU, in order to calculate large numbers.
Re: (Score:2)
IBM made one with 127 qubits 'the first capable of calculating 3 digit numbers.'
Someone published a paper on the state of the art in quantum cryptanalysis recently where they pointed out that the same could be achieved with a dog trained to bark three times. For some reason Fido the Quantum Wonder-dog is less newsworthy than quantum something-or-other.
Re:BS Computers (Score:4, Informative)
You can. People were writing quantum simulators in the sixties. There are Richard Feynman stories about them. Google "quantum circuit simulator in 27 lines of Python" and you can have your own.
The idea with quantum computers is to build hardware that makes them run faster.
Re: (Score:2)
Re: (Score:2)
I think you might want to read more carefully.
Re: (Score:2)
> one should be able to simulate the quantum entanglement with a regular computer.
That's like simulating a modern Intel processor with a cluster of 386's. You could string together enough compute power and fake out 64 bit operations but a modern CPU is going to be faster and more efficient.
You could also use oplocks and data transmission to fake out quantum entanglement, but the point is quantum entanglement is going to be faster and more efficient.
Re: (Score:3)
Re: (Score:2)
Solving problems using simulated quantum computers on top of real computers is slower than natively solving the problems on the computers themselves.
This is all a parallel exercise to develop the software and algorithms for when the hardware does exist. I'm sure that isn't over your head?
Re: (Score:2)
Many worlds or many cores? (Score:2)
The simulation runs on GPUs all in this world. Where is the quantum Magic if actual entanglement is not involved in the computation?
What is the technology below the hype? (Score:3)
This is about GPU accelerated Quantum Molecular Dynamics I assume, all the bullshit in the article just being hype for particularly idiotic VCs?
Re: (Score:2)
all the bullshit in the article just being hype for particularly idiotic VCs?
That seems to describe the vast majority of stories I see in the tech press nowadays.
Re: (Score:2)
Approximate infinity? (Score:4)
Every time I try to understand QC my head explodes in the old universe and we enter a new one; but I was given to understand that simulating a QC on ordinary hardware isn't possible because the probability of a q-bit state is a "smooth" probability between 0 and 1 before you measure it.
Either I'm misunderstanding again, or a few trillion states between 0 and 1 is enough to approximate a q-bit. If it is possible to simulate a q-bit without actually building one, it probably involves more expense than building an actual physical q-bit for now, otherwise it seems like all the quantum computing projects would be doing it.
Am I right, or is it time for my head to explode again? If so, see you in the next universe. I think. Or not.
Re:Approximate infinity? (Score:4, Insightful)
Either I'm misunderstanding again, or a few trillion states between 0 and 1 is enough to approximate a q-bit. If it is possible to simulate a q-bit without actually building one, it probably involves more expense than building an actual physical q-bit for now, otherwise it seems like all the quantum computing projects would be doing it.
Am I right, or is it time for my head to explode again?
I mean, yes... and no. (Appropriate for a quantum-related question, eh?)
The hope is that actual physical q-bits will become cheaper once a reliable process allows for economies of scale. Abacuses and slide rules were cheaper than complicated vacuum tube orchestrations, so it was perfectly possible to simulate an ENIAC without actually building one; and given enough abacuses, slide rules, and nerds to tinker on them you could have done (and people did do) all the missile trajectory calculation stuff for likely about the same overall cost as the R&D, fabrication, and sustained operation of ENIAC. But it's not always about the short-term, and focusing too much on immediate profit/loss is a phenomenally unhealthy way to run an R&D department.
Re: (Score:2)
Am I right, or is it time for my head to explode again?
It's easier if you go at it from the wave mechanics side of things.
A quantum mechanical state (of which a "qubit" is an example) can be calculated like a superposition of other states, pretty much like any periodic signal can be regarded as a superposition of other periodic signals (a.k.a. Fourier-transform). There's nothing magical about it.
You can put infinitely many states on top of one another, but if I'm not mistaken, the "qubit" part specifically takes the smallest possible definition -- i.e. a superp
BS meter pegged. (Score:3)
Re: (Score:2)
Looking at their website, I'd say the article is clueless garbage. They've developed a program for running density functional theory (a standard, widely used algorithm for simulating molecules) on GPUs. That's it. They say their GPU implementation is a lot faster than existing codes that run on CPUs, which is nice but has nothing at all to do with quantum computing.
Re: (Score:2)
using methods based on quantum algorithm development. They therefore continuously improve along with advances in the field of quantum computing without relying on quantum hardware.
So yes, they are pushing the quantum BS. In production use of quantum computing without quantum hardware, not for simulation...
News? (Score:2)
Re: (Score:2)