HP's Intel and Nvidia Powered ISS Supercomputer is Stuck in Space

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
HP and SpaceX sent some servers with off-the-shelf Intel Xeon CPUs and Nvidia Tesla GPUs to the ISS in 2017, and as of 2018, those servers were fully operational. But, according to a recent BBC report, HP's "Spaceborne Computer" was schedules to come back down to Earth about 3 months ago, and those plans were reportedly scrapped after a Soyuz rocket exploded with ISS crew members on board.

The BBC report says that the computers have been operational without any sort of maintenance for 530 days, and that they use radiators to transfer hot air from the server to the ISS's internal liquid cooling systems in a "closed air loop." That's impressive enough by itself, but a recent podcast from HP dives into just how difficult the conditions for these machines really are. The power that comes from the ISS's solar panels and batteries, for example, is somewhat unstable, and exhibits different characteristics than your typical power grid on Earth. The ISS's 92 minute day/night cycle reportedly subjects the system to temperature extremes, while alpha radiation tends to randomly flip bits in the silicon itself. HP's software package has to deal with all these issues, as well as more mundane ones like power regulation, without any human intervention, and according to HP, it's done a phenomenal job so far. The company seems to think they can apply some of these software innovations to computers on Earth, making them significantly more fault tolerant, but the ultimate goal of the program is to harden cutting edge computers for a Mars journey by 2030.

Check out the podcast here.

Traveling to Mars won't just be a physiological challenge for astronauts, but also one for the technology they carry on their missions. When an SOS transmission from deep space could mean life or death, every minute counts. It's today's research of space travel and the cosmos that will make a manned journey to Mars a reality.
 
Jeez.

And that's in low earth orbit where Alpha radiation is mUCH more limited than out in space.

Can you imagine what would be required for a Mars mission?
 
Jeez.

And that's in low earth orbit where Alpha radiation is mUCH more limited than out in space.

Can you imagine what would be required for a Mars mission?

Perhaps large geometry, slow-clocked, but massively parallel processors could work better out there. With tons of error checking. Maybe also with custom instructions geared toward the needed computation, rather than general purpose. Seems like the race to the smallest down here may be counter-productive up there. Of course that probably needs more power which is probably of equal concern to radiation.

Of course, maybe having some centralized computing resources in places that are better shielded for super-computer type workloads (say lunar installation and/or Mars for example) and remote interfaces and smaller machines for daily tasks on board space craft/stations might be a better solution. We need a better space communications network.
 
Last edited by a moderator:
Let's be realistic here......We need 100 more years of technological advancement before anything exciting happens in space travel...this!
 
The radiation thing is a legit concern. It's not something that hasn't been seen a lot here though - analog transmission required a lot of error checking, for instance... so it's not unheard of. Just something we don't consider at all levels of a computer. That part, the error detection and correction - legit impressive.

Everything else = meh. HPE just puffing up their chests.

There are power quality issues all over the world, I'm sure in Zamboozlestan the power grid is worse than ISS. Even inside an typical automobile you have electrical power issues - just watch what happens to voltage when you turn over the starter.

You have to consider heating/cooling anywhere you go, and there are more challenging environments than Outer Space without going up through the atmosphere or getting too exotic - trying running a server in Antarctica, or in a submersible, or in Mojave.

Those aren't the same challenges as what HPE is talking about, but still challenges none the less. And those have all been done, successfully. I'm not impressed by running on solar panels, or by being hooked into a coolant loop. I bet a lot of you guys here have more impressive cooling on your rigs than what HPE has going on up there.
 
let's face it. we're not leaving earth with our fancy full-tower custom liquid cooled bare-die overlocked love crafts. space is a disappoint.

scratch earth... i mean basement
 
Imagine that.. a spacecraft rolling along in space, with astronauts fraging inside in doom... hilarious!
 
It'll have to be the original. There's no such thing as dedicated servers anymore. :p :mad:
 
while alpha radiation tends to randomly flip bits in the silicon itself.

Does anyone know where these computers are placed that they interact with alpha particles?! That's He(2+) if someone doesn't know, and has a very small penetration depth to solid materials due to its size/mass and coulomb forces (microns for penetration depth; https://sciencedemonstrations.fas.harvard.edu/presentations/α-β-γ-penetration-and-shielding). They don't penetrate the skin. On the other hand, they'd certainly screw up a computer chip being so charged, but the most basic of shielding will block alpha particles. Either that was misstated (and meant another high energy particle that typically weakly interacts with matter) or they have their chips directly exposed to space.

I'm a bit confused by that detail.
 
Does anyone know where these computers are placed that they interact with alpha particles?! That's He(2+) if someone doesn't know, and has a very small penetration depth to solid materials due to its size/mass and coulomb forces (microns for penetration depth; https://sciencedemonstrations.fas.harvard.edu/presentations/α-β-γ-penetration-and-shielding). They don't penetrate the skin. On the other hand, they'd certainly screw up a computer chip being so charged, but the most basic of shielding will block alpha particles. Either that was misstated (and meant another high energy particle that typically weakly interacts with matter) or they have their chips directly exposed to space.

I'm a bit confused by that detail.

Secondary scattering?
 
Interesting article. For the Mars mimssion I'd expect them to get what they need for a nuclear reactor up in space and when they are ready to go they turn that on. It should provide stable power for the mission and beyond, give them a power source for doing things like charging the hull to help isolate the ship from external radiation and particulate. (Sorry maybe that's not a thing yet. But it should be! If the earth is naturally protected by a manetosphere why are we not protecting our deep space ships with the same? Probably power.. hence building a reactor in space.)

I'm thinking they won't do this... but it would be awwwwesoooome if they did.
 
For anyone who's ever asked why NASA uses Pentium 1/2 era chips still in their multi-million dollar devices? This is why.

NASA Hardens all their computers to prevent the radiation from utterly wrecking all their systems. Can you imagine the headache if Opportunity was sent to mars, only to have a corrupted disk drive when it landed?
 
For anyone who's ever asked why NASA uses Pentium 1/2 era chips still in their multi-million dollar devices? This is why.

NASA Hardens all their computers to prevent the radiation from utterly wrecking all their systems. Can you imagine the headache if Opportunity was sent to mars, only to have a corrupted disk drive when it landed?

As long as they're not calculating trajectories with a floating point bug. :D
 
The radiation thing is a legit concern. It's not something that hasn't been seen a lot here though - analog transmission required a lot of error checking, for instance... so it's not unheard of. Just something we don't consider at all levels of a computer. That part, the error detection and correction - legit impressive.

Everything else = meh. HPE just puffing up their chests.

There are power quality issues all over the world, I'm sure in Zamboozlestan the power grid is worse than ISS. Even inside an typical automobile you have electrical power issues - just watch what happens to voltage when you turn over the starter.

You have to consider heating/cooling anywhere you go, and there are more challenging environments than Outer Space without going up through the atmosphere or getting too exotic - trying running a server in Antarctica, or in a submersible, or in Mojave.

Those aren't the same challenges as what HPE is talking about, but still challenges none the less. And those have all been done, successfully. I'm not impressed by running on solar panels, or by being hooked into a coolant loop. I bet a lot of you guys here have more impressive cooling on your rigs than what HPE has going on up there.

The idea of changing power and temperatures so drastically and consistently and often, with no maintenance, seems to be unparalleled to me. Unless you can name a place on Earth that swings the temps and power that quickly and dramatically on Earth.. (I'm from Michigan, the land of "Hey, I bet you can't change difference in high temps 50 degrees in one day/here, hold my faygo" and honestly it doesn't matter all because powerful computers are *always* indoors, in a temperature controlled environment... they have to do that in space...)
 
For anyone who's ever asked why NASA uses Pentium 1/2 era chips still in their multi-million dollar devices? This is why.

NASA Hardens all their computers to prevent the radiation from utterly wrecking all their systems. Can you imagine the headache if Opportunity was sent to mars, only to have a corrupted disk drive when it landed?

Don't quote me absolutely, but there's been a number of cubesats built without hardened hardware (but with redundant hardware and software to error check constantly), which provides a natural experiment on what the best way forward is. Multiple non-hardened COTS systems greatly reduces prices and doesn't lock hardware early in the design process, while the multiplicity provides different forms of redundancy. Rad hardening eats into your weight budget as well, so you end up net-neutral by duplication vs hardening. We'll see how it evolves!
 
Yeah, but because it's a full PPC 750 design, it's out-of-order and superscalar, so performance is similar per clock to a Celeron 266 (Cacheless PII). The only thing it's missing is the huge L2 cache you saw on the Apple G3 systems.

Compared to earlier rad-hardened designs it's like a luxury car.
 
alpha radiation won't affect a Babbage's Analytical Engine. Brass and wood, baby. Brass and wood...
 
we humans can't seem to solve our problems here on earth; might as well venture out into space and see how things work out for us there. We're already well on our way when it comes to starting a junk yard business on the moon :LOL:
 
Back
Top