Wow, 4.5 kW of power draw max for a server with 512 MB of memory and the equivalent computer power of a single x86 core from about 25 years ago!
I’m pretty sure an Arduino could run circles around that thing, using a battery.
I’m pretty sure an Arduino could run circles around that thing, using a battery.
E.g.: Here's a drone reaching nearly 13 Gs, which would incapacitate or even injure a human: https://imgur.com/SHbDK2l
One recent change is that modern electronics uses mostly surface-mount components, which are not easily dislodged by g forces. Older electronics used large discrete components with wire legs (capacitors, inductors, etc...), which had a tendency to bend back and forth and eventually break off.
However, tender requirements aren't changed just because technology has moved on. The paperwork must be completed as specified! The vendors must show that their new technology that doesn't have these problems includes solutions for problems only present in old technology! Etc...
This results in 20-30 year old technology being used in new warplanes rolling off the production line today.
But rest assured, it's being done properly, with all the correct paperwork filled out! Everything in order.
Meanwhile, for the price of one of the two cooling fans on that monstrosity, I could buy half a dozen embedded processors made with silicon-on-insulator technology, which is also rad-hard, and have a 5-way system with majority voting and a hot spare... still running off a battery and cooled passively.
It's cool that your drone can do 13 G's, but the more realistic scenario is doing 5-6 Gs over and over and over and over and over again, sitting out in the sun for weeks, sitting on the deck of a carrier in the tropics, getting launched off a carrier, getting shaken around and exposed to salty air, sitting a few feet from a high power radar - you get the point.
Also, god forbid, but these things are designed to drop nukes, so that has to be taken into consideration as well.
Military hardware isn't magic unobtanium that exists in some parallel plane with unique physical conditions.
The choices being made for the F-35 have much more to do with government tender processes, military red tape, corruption, and general government incompetence than physics or engineering.
PS: It turns out I had misread the L3Harris info page. That spec isn't for the existing F-35 computers that are now about 20 years old, but for the TR-3 upgrade which will start rolling out in mid-2024!!! These new "flying supercomputers" will be 16x faster than before. This is equivalent to upgrading from a 1994 Intel Pentium to an Apple Watch Series 1 from 2015.
L3Harris is delivering just 1,000 units of these multi-kilowatt monstrosities for a low-low price of $300 million.
Say who? what were the competing solutions and how did F-35 go with the selection?
I've been on both sides of this tendering process, and it's absurd how horrifically inefficient it makes things. Small, agile, and capable vendors are literally excluded from the process and can't even begin to meet the criteria for acceptance. But... that's the point. It's not about cost-efficiency or product quality, but about keeping a few well-connected companies happy.
Try to look at things from the perspective of someone like Elon Musk, or a leader you admire from any FAANG or successful modern company.
Now tell them with a straight face that you're going to keep using computer hardware from the 1990s for 20+ years without an upgrade because... the paperwork is too hard.
You'd be laughed at, and then booted from the organisation for gross incompetence.
In the military, that's normal way things are done.
Imagine running Facebook, Google, or Amazon on 20 year old hardware. Picture that for a second.
Or if you want space-rated flight hardware, imagine SpaceX using something from the 1990s for Falcon 9 rockets because of red-tape.
You're rationalising nonsense. You're trying to tell me that it's good and proper that America and its allies are going to war with planes that have flight hardware that is equivalent to what I wear on my wrist to tell me the time, but uses 100,000x the power, takes up 500x the volume, and costs 1,000x as much.
We don't need ever more speed in processing, this isn't games.
Things like ultra reliable down to the nanosecond timing, etc. are important for these applications, not raw computing power.
You fundamentally don't understand some things about the application, therefore are presenting solutions that seem smart on the surface but are actually fundamentally flawed and non starters.
Different to what? Desktop PCs? Commercial avionics? Space-rated computers? Military fighter jet avionics from other planes?
The future upgrade to the F-35 is orders of magnitude worse at every metric than everything else in any of those categories.
For example, the latest F/18 has a computer that is several hundred times faster. It’s also a “networked” fighter jet with sensor integration, etc…
So what’s different about the F-35 avionics that justifies $300K per unit for the computer power of an outdated Apple Watch!?
From where I’m sitting it’s purely corruption and incompetence.
You’re saying it’s something else. Okay, what is it?
They're upgrading the 1994 level of computer power to 2015 level of computer power... in 2024, when they already have directly comparable fighter jet avionics that has a "giga" where this thing has a "mega".
The need to upgrade many of these computing systems is bottlenecked by the ability to upgrade the sensor systems that would necessitate upgrading the computing systems.
Also, you are talking about “mega” and “giga” when many modern US systems are “peta”. With all due respect, you might be working from an obsolete understanding of these systems.
It's fair to say, the engineering minds at work here have reasons. Reasons that would survive this armchair whataboutism.
In fact, there are many new articles bemoaning cost overruns, delays, poor performance, and more.
From what I've experienced first-hand with government tender processes, engineering considerations are a distant third on the priority list at best.
That's how you end up with the SLS using rocket engines from the Shuttle that cost more than an entire Falcon 9 rocket. Except that the the SLS uses 5 of these single-use engines, but the Falcon 9 is reusable dozens of times.
There really are situations that are this absurd, and the military industrial complex is notorious for them.
I would argue that the burden of proof is on you to demonstrate why F-35 jets are using hardware 100-1,000x worse in ever metric. Also please demonstrate why it uses bizarre non-standard CPUs instead of CotS hardware as seen in civilian avionics, SpaceX rockets, automotive ECUs, or pretty much anywhere else for that matter.
The F-35 planes are a success despite the avionics, not because of them.
The upgrade to Apple Watch levels of computing has notably been delayed by a year into mid-2024!
How is that a successful demonstration of computer engineering?
Space shuttle, bridges, tunnels, skyscrapers, planes, trains, on and on.
I'm comparing the F-35 to projects such as the Falcon 9, commercial avionics, or even just automotive ECUs. All of those outperform the F-35 in every metric, including reliability.
Please list the specific engineering hurdles that explain this monstrosity, because I just can't of think of any.
However, I can think of dozens of political- and corruption- related reasons for why this came to be. There are news articles and everything...
Apples and oranges.
Make any of those things you mention stealth, and armed, and the maintenance requirements will match that of the F-35.
Its the needs of the application that are driving many of the challenges you mention, not the implementation.
An example - the SR-71. Widely regarded as one of the great engineering feats of human history, a marvel of engineering for any time, but especially that time. It's by all accounts a maintenance nightmare. Not because it's not amazing, but because the specifics of the application demand so.
Its not easy to mass manufacture the most capable airframe in the world.
You're conflating the attributes of the plane with the attributes of the components that make it up.
The buttons in cockpit are not "stealth". You don't shoot the pilot seat out of the cannon. Etc...
Just because the plane is stealth and armed, doesn't impose any special requirements on the computers! They're on the inside of the plane, whereas the military nature of fighter jets is mostly relevant to the outside: stealth shape, stealth coatings, armor, etc...
Similarly, the special maintenance requirements of fighter jets are mostly to do with the high performance engines and physical stresses imposed on their airframe by high-G manoeuvrers. They're nothing to do with the computers, which have no moving parts and generally are serviced infrequently.
This whole discussion reminds me of how many people don't realise that many things are fungible and are designed to be fungible. There's this tendency to "label" things so as to break this mental model on purpose. The most common example is OPEX versus CAPEX -- it's all just money!
A more directly relevant example is a customer that was grumbling about how their email was slow to sync to their phones. They were blaming the phones, because they knew in their gut that "phones have slow CPUs", right? Meanwhile, the primary mail server had about 1/3rd the computer power of the phone it was synchronising to! Literally a fraction of the CPU power, memory capacity, and even disk capacity throughput. It was hilarious.
What had happened is that they "invested" a lot of money ($50K) on a big and important rack mount server... 15 years ago. The purpose of the server is critical to the business. They mentally labelled it as "production", "enterprise", "server grade", "highly available", and so forth. That's the box they put the thing into in their brain because that's the purpose the thing served. Never mind that after more than a decade, the phone in their pocket outperformed the server in every metric! They had mentally labelled those devices as "consumer", "gadget", "end user", etc...
These are just human labels. They don't affect the metrics, the objective measurements of reliability. The "enterprise grade" server had been crashing weekly. It had disk reliability issues (old mechanical drives!). It had capacity issues. Etc... It was labelled as reliable, but it was basically a worthless anchor on the organisation. If they had replaced it with a $500 laptop, they would have been measurably and objectively better off as a business! They just couldn't bring themselves to this realisation. It just wasn't "clicking" in their brain that it doesn't matter how much money they spent, or how flashy the sales presentation was, or what the purpose of it all is.
To me, this whole thread is that story all over again. Everyone here keeps going on about how the F-35 has "this" or "that" attribute that consumer electronics don't, while completely ignoring the fact that its computers are crazy slow and out-dated even when compared to directly comparable fighter jet aircraft such as the F-18!
PS: After doing even more research, I'm starting to suspect that the cause of the suckiness of the F-35 computers is that the tender requirements probably included "must be entirely manufactured in the US by US citizens", which is a common clause in military contracts. That would dramatically shrink the pool of options, reducing the performance and quality that is available. Similar issues occur with Russian military procurement, because their internal chip fab capability is much worse than what the US can produce.
It's specialised and more expensive, sure, but not that much. Having a cost or some spec be worse by a factor of 5x or 10x as a consequence is within reason. 100,000x... not so much.
I've seen this play out many times with government tender processes. They write the requirements months or even years before the tender officially begins. They write about established technologies at the start of this authoring process. Then it takes a year or more for a vendor to be selected, years more for development, years for acceptance testing, years for production and rollout, etc...
That's how you end up with a 10-20 year delay between the current-gen technology and what is contractually required, by a legal document literally mandating that the technology be hilariously outdated by the time it is actually used.
The more exponential the growth curve of some technology, the worse this is. If the Army is putting out a tender for uniforms, tents, or MREs, this is no big deal. If the Air Force is building "high-tech flying supercomputers", it is crazy bad.
Government agencies, especially big ones, simply have no concept of aiming for where the puck will be, not where the puck was.
Combine that with requirements that tie the hand of the vendor behind their back, such as mandating 100% US-made parts, then costs can go into the stratosphere while the technical capability lags behind by decades.
You can keep trying to rationalise and defend this madness, but it's your taxpayer dollars being heaped into a giant pile and set on fire.
Airliners and the like are triple, sometimes quad.
Designing a Dell server for high uptime with dual power supplies is one thing. Designing stuff where it falls out of the sky and crashes into a high-school if the computer signal from one sensor to one actuator takes 2 milliseconds too long to traverse.
I doubt to a high degree that an Arduino could handle anything that this computer can.