A neuromorphic supercomputer now has one-million processor cores.
The Spiking Neural Network Architecture (SpiNNaker) machine can perform 200 trillion actions per second. Each chip has 100 million parts.
The goal is to continue to scale up by 1000 times and simulate one billion biological neurons in real time. A mouse brain has about 100 million neurons and the human brain is 1000 times bigger than that.
They are simulating a section of the brain called the Basal Ganglia. This is the area affected in Parkinson’s disease.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Idk, Can it run minecraft?
Still, it’s just general purpose ARM9 CPUs with a very parallel interconnect fabric. An FPGA can be anything. You can easily implement interconnected ARM9 logic within an FPGA but at a lot higher efficiency. Also, the modern FPGAs have very high bandwidth interconnects so you can chain them for a larger combined logic.I would say FPGAs mimic a real brain better than general purpose CPUs because the core logic is restructured fundamentally just like the brain is physically altered when trained.
Still, it’s just general purpose ARM9 CPUs with a very parallel interconnect fabric. An FPGA can be anything. You can easily implement interconnected ARM9 logic within an FPGA but at a lot higher efficiency. Also, the modern FPGAs have very high bandwidth interconnects so you can chain them for a larger combined logic.
I would say FPGAs mimic a real brain better than general purpose CPUs because the core logic is restructured fundamentally just like the brain is physically altered when trained.
There is difference in just running common neural networks.As width FPGA’s or GPU’s vs emulating real biological neural networks.The FPGA/GPU are more commonly used networks in computers but they dont try to mimic the firing pattern of real biological neural networks. Instead neural newtorks on FPGA or CUDA just use something that works best in electronics and our math understanding. To really understand Parkinson a brain signaling disease, it wont help simply just tensorflow on the latest FPGA or GPU. The best signaling emulations have to be used, in the most strange configurations (not as logical as in voice to text)
AFAIK, Fortnight doesn’t require a particularly strong computer. Back in the day, Crysis was notorious for requiring very high-end hardware, and even then pushing it to its limits. Or at least, that’s the impression I got at the time. The version of Doom that I think GoatGuy was referring to is even older still.
Roughly equivalent to a mouse brain. It’s far from any human brain simulation.
There is difference in just running common neural networks.
As width FPGA’s or GPU’s vs emulating real biological neural networks.
The FPGA/GPU are more commonly used networks in computers but they dont try to mimic the firing pattern of real biological neural networks. Instead neural newtorks on FPGA or CUDA just use something that works best in electronics and our math understanding. To really understand Parkinson a brain signaling disease, it wont help simply just tensorflow on the latest FPGA or GPU. The best signaling emulations have to be used, in the most strange configurations (not as logical as in voice to text)
AFAIK, Fortnight doesn’t require a particularly strong computer. Back in the day, Crysis was notorious for requiring very high-end hardware, and even then pushing it to its limits. Or at least, that’s the impression I got at the time. The version of Doom that I think GoatGuy was referring to is even older still.
Sorry Michael, Crysis hasn’t had a new release in 5 years. Try “Fortnight”. To Goatguy’s point – a million of anything is formidable, but it still needs good coordination and consolidation of results. A million soldiers without any command and control is going to get you random results. If we don’t know the structure of the brain, a very fast bad simulation is no better than a slow bad simulation.
Right. Even if the hardware is there, it’s meaningless without the right software, and software development most assuredly does not follow Moore’s Law.
Are you exercising? .
I Googled a bit on that machine.It uses a whole bunch of ARM9 cores interconnected with modest hardware in a very parallel fashion. Total energy consumption is 90 kW.They don’t publish inference performance for any standard benchmarks that I can find, which is a pity. They have spent a long time (~10 years) developing this machine so I have a feeling they are maybe leapfrogged by other solutions like the recent XILINX/AMD FPGA monster.Google on “Alveo U250 world record” to see what we use in 2018 for inference.
But can it simulate a part of the brain in a meaningful way? Just recently, some scientists managed to analyze a part of the brain, and they concluded that there were 68 different types of brain cells (!). In addition, it may be that the glia-cells (which outnumber the brain cells 10-to-1) have some unknown affect on the neurological operation. On top of this… To make a meaningful simulation, you need information about how the neurons are connected to each other, i.e. an approximate connection map.I don’t think they are anywhere near this knowledge, so I would guess that the simulation will yield almost completely meaningless results if your goal is to learn about the human brain…
I was diagnosed 2 years ago at age 63. Symptoms were tremor in right leg, loss of handwriting ability, and soft voice. I also have difficulty rising from a seated position and have balance issues. I started out taking only Azilect, then Mirapex, and 6 months ago Sinemet. Several months ago I started falling frequently, hence the reason for Sinemet. I tried every shots available but nothing worked. In June 2018, my neurologist and I decided to go with natural treatment and was introduced to Natural Herbal Gardens natural organic Parkinson’s Herbal formula
These days, I think the saying is “But can it run Crysis?”.
Sorry Michael, Crysis hasn’t had a new release in 5 years. Try “Fortnight”.
To Goatguy’s point – a million of anything is formidable, but it still needs good coordination and consolidation of results. A million soldiers without any command and control is going to get you random results.
If we don’t know the structure of the brain, a very fast bad simulation is no better than a slow bad simulation.
Right. Even if the hardware is there, it’s meaningless without the right software, and software development most assuredly does not follow Moore’s Law.
Are you exercising? https://www.parkinson.org/Understanding-Parkinsons/Treatment/Exercise .
I Googled a bit on that machine.
It uses a whole bunch of ARM9 cores interconnected with modest hardware in a very parallel fashion. Total energy consumption is 90 kW.
They don’t publish inference performance for any standard benchmarks that I can find, which is a pity. They have spent a long time (~10 years) developing this machine so I have a feeling they are maybe leapfrogged by other solutions like the recent XILINX/AMD FPGA monster.
Google on “Alveo U250 world record” to see what we use in 2018 for inference.
But can it simulate a part of the brain in a meaningful way? Just recently, some scientists managed to analyze a part of the brain, and they concluded that there were 68 different types of brain cells (!). In addition, it may be that the glia-cells (which outnumber the brain cells 10-to-1) have some unknown affect on the neurological operation. On top of this… To make a meaningful simulation, you need information about how the neurons are connected to each other, i.e. an approximate connection map.
I don’t think they are anywhere near this knowledge, so I would guess that the simulation will yield almost completely meaningless results if your goal is to learn about the human brain…
I was diagnosed 2 years ago at age 63. Symptoms were tremor in right leg, loss of handwriting ability, and soft voice. I also have difficulty rising from a seated position and have balance issues. I started out taking only Azilect, then Mirapex, and 6 months ago Sinemet. Several months ago I started falling frequently, hence the reason for Sinemet. I tried every shots available but nothing worked. In June 2018, my neurologist and I decided to go with natural treatment and was introduced to Natural Herbal Gardens natural organic Parkinson’s Herbal formula
These days, I think the saying is “But can it run Crysis?”.
Or as we used to say, “Yah? But how fast does it play Doom?”I know, bad joke. Thing is, that a million-of-anything computing-wise is a formidable computing device. Heck, it could be a blôody digital Turing Machine, and it’d still be an impressive computing device. In this case, a million-plus specialized processors are working together in a neurocomputational network, to infer, as quick as possible, the relationships between time-varying inputs. Quite a computer. Just saying,GoatGuy
Or as we used to say, “Yah? But how fast does it play Doom?”
I know, bad joke.
Thing is, that a million-of-anything computing-wise is a formidable computing device. Heck, it could be a blôody digital Turing Machine, and it’d still be an impressive computing device. In this case, a million-plus specialized processors are working together in a neurocomputational network, to infer, as quick as possible, the relationships between time-varying inputs. Quite a computer.
Just saying,
GoatGuy