This week, another article came from a Washington DC think tank – that China is going to out-compute America – with catastrophic results.
America will have to adapt all its technology to Chinese-set standards.
The Chinese will flood the market with cheap A.I. systems driving U.S. companies out of business.
Autonomous weapons – like drone swarms – where the Chinese already have a lead, will out weaponize America.
The only thing the article did not include is Ghostbusters – “dogs and cats, sleeping together…..”
However, this week, any chance of that Chinese dream was set back a long way – when Fractal moved the contents of several data center server clusters, to a single Apple Mac Studio – about the size of a small shoe box – proving intelligence at the edge, without a data center is now possible.
Life without scores of new data centers is demonstrable at massive scale – and if you are on our growing early morning Zoom sessions, you are experiencing the future – as we demonstrate systems that previously required data center server clusters – running on a single Apple Mac Studio plugged into the wall.
These applications run faster – using almost no energy – compared to a data center server cluster.
Let’s tie these thoughts together.
The first concept is Chinese dominance in A.I.
To dominate in A.I. you must solve BOTH the energy problem and the edge problem.
The energy problem, if you haven’t noticed, is well presented by guys like Oracle’s Larry Ellison who want a half trillion dollars of new data centers dotting the farmland.
The little secret is that money – all of the money on the planet, cannot solve the energy problem because you cannot build data centers where there are not extraordinary amounts of energy to run them.
Nor can you develop that energy any time soon – nuclear reactors and new energy plants take years to build. Many years – even for the Chinese.
You can throw all the money in the world at the problem, but that solution is years, perhaps decades away.
Whoever solves the energy problem – without taking years – probably wins. (If you do not want to read the entire article, Fractal just did it this week on an Apple Studio).
The edge means A.I. literally everywhere.
You cannot dominate the edge if you need an energy-consuming data center at every edge point.
The edge is best defined as “everywhere BUT in the data center.”
Your wrist is the edge. A soldier’s backpack is the edge. A satellite is the edge. A drone is the edge. A Palantir or Oracle data center is not the edge – it is the problem – we’ll get back to that.
A.I. is useless if it isn’t everywhere.
Well, not useless, but he who delivers A.I. – intelligence – everywhere – wins.
And, to tie these two concepts together – YOU CANNOT DELIVER A.I. AT THE EDGE IF YOU NEED DATA CENTERS EVERYWHERE TO DO IT.
OK, so we have three postulates:
- Whoever solves the immediate energy problem can dominate A.I. because they develop A.I. without its single largest constraint – energy.
- Whoever delivers A.I. everywhere – at every edge – wins.
- Whoever does both simultaneously – sets the standards for A.I. going forward.
The Fractal team believes you must solve both the energy challenge and the edge challenge simultaneously – to dominate A.I.
We have been quietly showing this for some time – but this week we did it on our new favorite hardware platforms – the Apple Mac Studio and Apple Mac Mini.
Since we are now deeply engaged in almost constant demonstrations to people who want to see the future, we are reviewing it for you here – and we are happy to demonstrate these capabilities any time, anywhere, on any size data.
Here’s how this project started…….
The Fractal team was asked by a Department of War vendor – the guys doing the drones operating as a flock of birds – to benchmark different hardware platforms in support of battlefield super-computing.
Their engineers understand Fractal’s distributed capabilities enable MESH computing – where a central data center – is replaced by the aggregation of scores, hundreds or thousands of devices, on a battlefield – coming together virtually – not physically – delivering super-compute – without a data center single point of failure.
If you haven’t read the piece, here is how we do it:
So here is the Apple Mac Studio test – we are now demonstrating on weekly Zoom calls to our growing list of software partners:
- Voter rolls for 26 states – each with millions of records
- Each state uses a multi-million dollar-a-year data center to process, store, search and manage these rolls
- 5 States also use a cloud to handle the extra work
- Fractal moved all 26 state systems – the entire voter roll, voter history, everything – to a single Apple Mac Studio – the size of a shoebox
26 state systems – each needing a data center to run – moved to an Apple Mac Studio the size of a shoe box.
Not only did it work, but it ran faster than the systems in the data centers.
Then the engineers upped the game a bit.
Why not take those voter rolls and multiply them by 20 – let’s test this “scale” thing.
Fractal applications often perform time series analysis on historical data – thus we take multiple copies of every state voter roll and compare each copy to its predecessor to detect changes.
In Pennsylvania, the Fractal system doesn’t have one copy of Pennsylvania’s about 9 million voters, it has 77.
Pennsylvania’s government cannot manage 77 copies of its voter rolls on different dates – using a data center and all of the energy that it consumes.
We can – on an Apple Mac Studio with the power consumption of a Black and Decker electric drill.
We aren’t running 26 data center applications, we are running 26 data center-sized applications and data – times 20.
The times 20 is because we keep so many historical copies for comparison.
For some time now, we have been drawing your attention to the fact that compute is fast – but I/O (input/output) data movement is slow – and it is this slowness of data movement that drives the need for data centers.
Today’s CPUs (compute) are blindingly fast – they have been engineered so with almost a trillion dollars in cumulative research and development investment over the last 50 years.
What Fractal delivers is an architecture that unleashes those little chip beasts to compute at full speed.
For 50 years software was made faster because chips were made smaller and faster.
Now, chips cannot get much smaller, and enterprise software does not fully utilize the vast capabilities of modern CPUs – and hasn’t for over a decade because enterprise software does not do a good job of optimizing I/O.
Fractal leverages the massive investment in chip design to unleash that speed – which no relational database can do.
Since we are one of a handful of companies doing this right now, you may want to pay attention to us here on Substack and subscribe.
The point is that if you optimize I/O throughout the entire software stack, there is no need for a new data center – unless of course, you are in the business of building data centers.
Data centers are “solving” the I/O problem with brute force – and huge amounts of money and energy consumption. Fractal solves the I/O problem with innovation in the software stack itself – and that software stack can run anywhere and does not require a data center.
Remember the Fractal equation:
99% of what a computer or data center does is I/O.
If you dramatically reduce and/or optimize I/O, applications can run 1,000 times faster.
If an application runs 1,000 times faster, it needs 1/1,000th the hardware for the same work.
Thus it needs 1/1,000th the energy to make that hardware run.
Thus, no new data center.
So, when we read these articles that China is going to rule the A.I. world, we have the following observations:
What we saw this week is there is a computer hardware company (Apple), that is already at the edge – on watches, phones, tablets, computers, servers, numbering in the billions.
Frankly, Apple is the edge.
We then proved that data center sized applications can run on them today – both on individual Apple Mac Studios and Mac Minis and in a company’s entire Apple compute MESH – with lower cost, less energy use, and better performance.
We think that’s pretty cool and we are showing it to people like you every week.
