10

I was reading this question about CPU's used in space and that got me thinking about the practical applications of GPUs for space-bound vessels. I'm wondering if we ever sent a radiation-hardened GPU to Mars or something, seeing as GPUs are extremely good at multi-threading processes. Anyone have any Insight on this (bad pun intended)?

Upon a precusory search (the results went over my knowledge threshold quickly)...

The only thing I could really find is this: https://www.sbir.gov/sbirsearch/detail/1482621

And this: https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20170006038.pdf

Which is kind of making me think we haven't perfected radiation hardened GPUs...

Regardless, the question is:

  • Have we used GPUs for practical applications in space-bound vessels?
  • What for? Or, if the answer is no, what is preventing us from doing so?

Other related topics (provided by @uhoh):

Magic Octopus Urn
  • 9,387
  • 3
  • 31
  • 82
  • 1
    No helpful hits here: SAR image processing or here: Machine Learning and Classical Navigation, Guidance and Control in Space Exploration – uhoh Dec 21 '18 at 15:48
  • 1
    @uhoh good lord... you know what I didn't do? Search the SE... I googled it instead. Whoops. Lucky those aren't directly related to GPUs in space. Still really neat topics though. Sort of touches on the problems with energy usage in GPUs in the ML topic too, didn't think about that. – Magic Octopus Urn Dec 21 '18 at 15:50
  • 2
    IIRC many radiation-hardened ICs still use 250nm process. The current GPU design may not be readily implemented in those old process. – Mys_721tx Dec 21 '18 at 16:15
  • you can google the HP "Apollo" HPC system that's on the ISS to see if it's all CPU or has some GPU power as well, and also look at the SpaceCube project 1, 2, 3 – uhoh Dec 21 '18 at 16:36
  • 1
    Regarding the HP Enterprise "Apollo" on ISS that @uhoh mentioned, I asked HPE last year if that one has GPUs, and was told that it does not. IT has standard HPE Apollo 40 parts with "HPE Gen-9 compute nodes featuring Haswell/Broadwell class processors, commonly used in HPC deployments." – Rick 0xfff Dec 22 '18 at 23:54
  • @Rick0xfff What's up with the "Apollo" computer on the ISS? How is it being used? feel free to edit the question (or suggest edits) if you think it can be made more interesting. – uhoh Dec 23 '18 at 00:05
  • @MagicOctopusUrn Those two links you found are very interesting. A nit: I wouldn't consider GPUs to be multi-threaded; but they are SIMD with a really big "M". :-) The NEPP paper looks like an overview for a project; the NSF SBIR page is amusing; it is a copy of a request from DoD MDA, asking for someone to develop such tech, implying they don't have it. – Rick 0xfff Dec 23 '18 at 00:06
  • why GPUs; we are still flat-lining the 2cores per cpu on intel 8/16 laptops and amd >=16/32 desktops that the manufacturers seldom mention: with ssd virtual memory and adjoint monte carlo methods we have little time to reflect on the answers we do get on commodity machines! – thomas m jordan May 02 '22 at 22:07

2 Answers2

10

This is a little bit of a cop-out answer, but I have some pertinent experience. There are GPUs in use on the ISS ... in the laptops. The astronauts on the ISS receive briefings before EVAs in a "3D walkthrough" form. This uses NASA's EDGE renderer and a super-accurate 10mil poly model of the exterior of the Station. They also stay up to date on SAFER procedure training using an Oculus Rift (VR headset). The JSC VR Lab had to bypass significant portions of Oculus' software in order to optimize the VR to be useable on the (radiation-tested) HP Z-book 15 Gen 2 laptops they have there.

For general purpose rendering use, the extra radiation hasn't had a noticeable impact. This could change once outside of the Van Allen belts.

CourageousPotato
  • 2,947
  • 11
  • 34
  • Wow, thanks for the actual names of the technology being used. Any answer is better than no answer, and this one seems better than many I've seen! Also-- I may ask and link a question about the VR headset. They brought it up just for training purposes or did Oculus contract them to test stuff in space with it? – Magic Octopus Urn May 15 '19 at 12:47
  • Do they use EDGE now instead of DOUG? – Organic Marble May 15 '19 at 16:55
  • They brought up the Oculus for training to replace the previous VR solution: a laptop strapped to the head. They chose Oculus for its light weight and they needed to bypass some of its sensors because of the zero gravity and inside-out operation. – CourageousPotato May 15 '19 at 18:32
  • 1
    EDGE stands for Engineering DOUG Graphics for Exploration: https://software.nasa.gov/featuredsoftware/edge. It’s a continuation of DOUG. – CourageousPotato May 15 '19 at 18:34
  • "A laptop strapped to the head" YES! It's a practical-ish solution, though; your neck won't get tired holding it up like it would on Earth, but turning your head would be difficult and control wouldn't be great. In addition, I'm guessing the pixel density of the laptop is not designed for that, meaning that the image would look really terrible. Oculus probably uses really high pixel densities for small screens and lenses; it's purpose built. –  May 16 '19 at 15:54
  • Actually, since the laptop has a 1080p screen, assuming the lenses are done well, the resulting image would be similar to a first-generation Gear VR setup in resolution/clarity. Low refresh rate and no asynchronous reprojection though. – CourageousPotato May 16 '19 at 18:11
  • 2
    Edit to my previous post: DOUG is a somewhat minimal real-time rendering engine (pass it models, shaders and a real-time data stream and it outputs video). EDGE is DOUG packaged with sim setup and UI/interaction tools. DOUG is still used on its own for some things. – CourageousPotato May 22 '19 at 23:28
6

The Mars helicopter Ingenuity uses a Qualcomm Snapdragon 801 System-on-Chip which is well known from smartphones and includes an Adreno 330 integrated GPU.

The MOCI satellite will have a NVIDIA Jetson GPU module for 3D reconstruction of Earth surface features:

The Multiview Onboard Computational Imager (MOCI) is a 3U cube satellite designed to convert high resolution imagery, 4K images at 8m Ground Sample Distance (GSD), into useful end data products in near real time. The primary data products that MOCI seeks to provide are a 3D terrain models of the surface of Earth that can be directly compared to the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) v3 global Digital Elevation Model (DEM). MOCI utilizes a Nvidia TX2 Graphic Processing Unit (GPU)/System on a Chip (SoC) to perform the complex calculations required for such a task

"A Near Real Time Space Based Computer Vision System for Accurate Terrain Mapping" by Caleb Adams

KarlKastor
  • 2,163
  • 20
  • 27