The ISS's Supercomputer is Ready for Service

AlphaAtlas

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
SpaceX sent an HP supercomputer to the ISS over a year ago, and now, after an extensive testing period, that supercomputer is available for astronauts to use. According to HP, performing calculations onboard the space station instead of on ground-based stations will save precious time and bandwidth. However, HP and NASA also view it as a testbed for future missions, where latency could range from seconds (on the Moon) to minutes (on Mars), and transmission windows and bandwidth could be spotty at best. The supercomputer is based on the aptly named Apollo p40 server, presumably with Nvidia Tesla GPUs and Intel Xeon CPUs, but has been modified with a water-cooling setup for space-based operation. Interestingly, as opposed to the traditional form of hardening used in most space equipment, HP "hardens" this system with special software designed to recover from errors that inevitably come with being bombarded by radiation. HP also posted a timeline of the system's history, for those who are interested.

Due to limited computing capabilities in space, many of the calculations that are necessary to complete research projects started in space are still processed on Earth. This approach is feasible for running research on the moon or in low earth orbits (LEO) between 400 and 1,000 miles above the Earth's surface, where communication can be in near real-time with Earth. However, larger communication latencies of up to 20 minutes both to and from Earth can occur when data is captured farther into space and closer to Mars. This reality makes any on-ground space exploration challenging and potentially dangerous if astronauts are faced with mission-critical scenarios that they are unable to solve themselves.
 
Last edited:
AlphaAtlas said:
SpaceX sent an HP supercomputer to the ISS over a year ago, and now, after an extensive testing period, that supercomputer is available for astronauts to use. According to HP, performing calculations onboard the space station instead of on ground-based stations will save precious time and bandwidth. However, HP an NASA also view it as a testbed for future missions, where latency could range from seconds (on the Moon) to minutes (on Mars), and transmission windows and bandwidth could be spotty at best. The supercomputer is based on the aptly named Apollo p40 server, presumably with Nvidia Tesla GPUs and Intel Xeon CPUs, but has been modified with a water-cooling setup for space-based operation. Interestingly, as opposed to the traditional form of hardening used in most space equipment, HP "hardens" this system with special software designed to recover from errors that inevitably come with being bombarded by radiation. HP also posted a timeline of the system's history, for those who are interested.

Due to limited computing capabilities in space, many of the calculations that are necessary to complete research projects started in space are still processed on Earth. This approach is feasible for running research on the moon or in low earth orbits (LEO) between 400 and 1,000 miles above the Earth's surface, where communication can be in near real-time with Earth. However, larger communication latencies of up to 20 minutes both to and from Earth can occur when data is captured farther into space and closer to Mars. This reality makes any on-ground space exploration challenging and potentially dangerous if astronauts are faced with mission-critical scenarios that they are unable to solve themselves.
Click to expand...

HP and NASA. Not HP an NASA.

The Apollo P40 is nice but I am not sure I would call it "super". Its 1 Teraflop so I guess if we still call anything that reaches a teraflop a supercomputer? I thought we had upped the ante with the massive improvements in compute capability.
 
kju1 said:
HP and NASA. Not HP an NASA.

.... Its 1 Teraflop so I guess if we still call anything that reaches a teraflop a supercomputer? I thought we had upped the ante with the massive improvements in compute capability.
Click to expand...
It's not its

I like this game.
I'm always a bit too chickenshit to start it because then people will highlight any minor errors I make and laugh at me.
Seems you're not bothered by that.
 
Laowai said:
It's not its

I like this game.
I'm always a bit too chickenshit to start it because then people will highlight any minor errors I make and laugh at me.
Seems you're not bothered by that.
Click to expand...

Nope I am not. But one is front page news the other is just some village idiot posting on a forum. Hint: I am not front page news ;).
 
But can it run Crysis?

AlphaAtlas said:
Interestingly, as opposed to the traditional form of hardening used in most space equipment, HP "hardens" this system with special software designed to recover from errors that inevitably come with being bombarded by radiation.
Click to expand...

The ISS is within Earth's magnetic field. I imagine the error rates are higher than normal but not drastically so. They use regular laptops up there.
 
What is the ISS power budget by the way? I guess they will have the crew compartment heating covered running that.
 
its not really a super computer.. as we think of it. its purely a test bed

https://www.nasa.gov/mission_pages/station/research/experiments/2304.html#overview


The research objectives of the Spaceborne Computer include a year-long experiment of operating high performance commercial off-the-shelf (COTS) computer systems on the ISS with its changing radiation climate. During high radiation events, the electrical power consumption and, therefore, the operating speeds of the computer systems are lowered in an attempt to determine if such systems can still operate correctly. Additionally, this is a long duration experiment, studying the practicality of running and managing COTS high performance computer systems in orbit from several months to one year. In summary, the objectives are: 1) run compute and data intensive applications in a changing radiation climate, 2) monitor power consumption and dynamically tune the power consumed, and 3) determine effects of solar radiation on the systems while running. In order to achieve these objectives, Hewlett Packard Enterprise (HPE) proposes a total of four identical high performance COTS computer systems. Two of the systems are installed aboard the ISS in a side-by-side EXPRESS locker within an ISS EXPRESS Rack. These two systems with the required networking are integrated at the HPE facility and turned over to the ISS Cargo Mission Contract (CMC) as required.
Click to expand...


so basically.. they will run 4 computers doing the EXACT same calculations. 2 up there, 2 on the ground and look for anomalies during high radiation events. then tune/tweak the setups to find optimal performance.

This is to get around the current hardening of computer chips that is needed to be used in space. by the time such hardening is done.. they are YEARS old and out of date, by our standards. They are hoping they will be able to overcome this so they can user newer, and faster computing equipment in space without it failing.



the TL;DR version.. they are looking to make computers stable so one in a long term ship doesnt get hit by radiation and become HAL9000

LOL
 
Hmm, I wonder how water cooling does in zero-g. I mean it's a forced flow pump so it should work, but it might have fluid settling in all the wrong ways.

Soon we'll have the cpu cooling tank from Sunshine... thats not zero grav tho.
 
Burticus said:
Hmm, I wonder how water cooling does in zero-g. I mean it's a forced flow pump so it should work, but it might have fluid settling in all the wrong ways.

Soon we'll have the cpu cooling tank from Sunshine... thats not zero grav tho.
Click to expand...

That’s ok because space isn’t zero g either.
 
kju1 said:
That’s ok because space isn’t zero g either.
Click to expand...
no, but you have no way to fight it so you go with the pull which effectively gives you a zero g like condition like those 45s free fall plane drops they do to train people for weightlessness.
 
kju1 said:
Its 1 Teraflop so I guess if we still call anything that reaches a teraflop a supercomputer? I thought we had upped the ante with the massive improvements in compute capability.
Click to expand...
I'm not sure where they are pulling their numbers from, but the two CPUs in that system alone are capable of more than 1 TFLOPs of FP32 compute.
Heck, each Tesla P100 GPU is capable of roughly 4.7 TFLOPs of FP32 computer, so maybe they were just measuring the CPU(s) with their proprietary software or something?
 
Red Falcon said:
I'm not sure where they are pulling their numbers from, but the two CPUs in that system alone are capable of more than 1 TFLOPs of FP32 compute.
Heck, each Tesla P100 GPU is capable of roughly 4.7 TFLOPs of FP32 computer, so maybe they were just measuring the CPU(s) with their proprietary software or something?
Click to expand...

*shrug* Or they just seriously under clocked every component? Either way I thought it was laughable they called this a super computer.
 
You must log in or register to reply here.
Back
Top