Mars Perseverance rover uses 'same processor' as original iMac

[HEXUS]

The RAD750 SBC is a radiation / temperature hardened device using a PowerPC 750 CPU.

Read more.

[Irien]

Whilst I get that performance isn't the key metric here, I would have thought efficiency would still matter, especially within limited power budgets on a rover. 5w for that level of performance is pretty dire, although granted I have no idea on the overheads introduced by the hardening - it'd be facinating to learn more about how that is achieved.

For a non-hardened system, a SoC ARM board (from a few years ago) would typically use half (or less) the quoted power consumption, and would be sub 1W for the level of performance mentioned here. I guess the hardening process really is the key - does anyone know of any ARM designs in that area?

[Glyce]

Its probably a US government contract, you know,"one of those".
I'm 100% sure SpaceX will make 10 times faster, 10 times more efficient, more resilient and stable, and all based on a $30 Rasp Pi or similar.

All sides of this story are fascinating.

[spacein_vader]

Don't forget this mission was first proposed in 2012 and the rover itself is based on the Curiosity rover which launched in 2012 and was designed some years before that. The CPU is the same model used in Curiosity and I suspect the appeal of a known good CPU (Curiosity still works!) with a proven codebase to build upon was an appealing one to NASA, allowing more resources to be spent on other elements of the rover.

If it ain't broke...

[CAT-THE-FIFTH]

A few things. Firstly Perseverance is actually based on Curiosity,with some improvements and the existing hardware is tried and tested. The more changes you make with new untested hardware,the more issues you could have during a space mission which could last 10 years. Basically it was cheaper to re-use tried and tested hardware. It was also known back in 2017 the computer systems would be re-used:
https://www.reddit.com/r/space/comme...ame_computing/
https://mars.nasa.gov/msl/spacecraft/rover/brains/
https://mars.nasa.gov/mars2020/spacecraft/rover/brains/

There is also a good reason why you don't want to rush out and use the latest and greatest hardware,even if it is new radiation hardened lines. How well errata have been documented.You really don't want to use something too new,as not all errata are well documented. Also some of the older SOI process nodes are very radiation tolerant compared to newer bulk nodes. For example NASA tested Llano which was made on GF 32NM SOI,and it was suprisingly radiation tolerant compared to the Intel equivalents(but the motherboards were the major limiting factor):

https://nepp.nasa.gov/files/24982/NE..._LaBel_AMD.pdf

Don't forget this mission was first proposed in 2012 and the rover itself is based on the Curiosity rover which launched in 2012 and was designed some years before that. The CPU is the same model used in Curiosity and I suspect the appeal of a known good CPU (Curiosity still works!) with a proven codebase to build upon was an appealing one to NASA, allowing more resources to be spent on other elements of the rover.

If it ain't broke...

Exactly. Its even the case for defence projects,ie,the F22 is one of the most sophisticated fighters ever made,but used Intel i960MX CPUs from the 1980s:
https://www.militaryaerospace.com/co...uture-upgrades

Then during the end of production run,Intel stopped making it meaning an alternative CPU had to be integrated. It's because it took decades for the F22 as a weapons system to be developed. So that is another consideration,the hardware specs for many aerospace and military projects can be locked 10 or maybe 20 years beforehand.

[Gentle Viking]

Embarrassing sending a rover with such a poor hashrate to another planet

[Rubarb]

Explains why it cost so much to make then if it used apple parts

[Wirko]

Embarrassing sending a rover with such a poor hashrate to another planet

Doubly embarrassing, knowing that the economy on said planet is crypto-based.

[GlasWolf]

Maybe this was the most recent CPU they could find in stock anywhere.

[DanceswithUnix]

If it ain't broke...

Quite. You don't want the next Mars rocket to be "To switch the rover off and on again"

[maverick77_uk]

It may also be to do with the width of the circuits. 150-250nm is probably much more resistant to everything from random radiation, temperature etc. etc. than the latest and greatest 5nm, being much thinner and consequently more "fragile" or at least affected by extreme conditions.

As others have said, you want something tried and tested albeit a bit slower, then the latest and greatest. Not exactly like you can do a recall and stick a new CPU in it

[cheesemp]

It may also be to do with the width of the circuits. 150-250nm is probably much more resistant to everything from random radiation, temperature etc. etc. than the latest and greatest 5nm, being much thinner and consequently more "fragile" or at least affected by extreme conditions.

As others have said, you want something tried and tested albeit a bit slower, then the latest and greatest. Not exactly like you can do a recall and stick a new CPU in it

I was going to say the same. I guess with 5nm lower voltage/faster speeds a stray bit of radiation from space can easily cause a lot of bit flips. What's better a tried and tested in the environment chip or a new ARM which needs Kgs of lead shielding (that won't be 100% affective at blocking it). Hopefully by the time they start the next project a new hardened ARM with redundant cores/Advanced EEC RAM will be available.

[raygdunn]

Well at least it wasn't a Zilog Z80.

I was amazed to find they still make them .

[philehidiot]

Of perhaps interest, I did a huge project recently looking at the vulnerability of new Vs old pacemakers and implanted defibs to high radiation environments. The current accepted practice for people undergoing radiotherapy is from 1993 and tests then showed that radiation / stray neutrons weren't a huge problem, but EM interference was. Now, on modern devices, the position is inverted. Newer devices are far more vulnerable to bit flipping and so on (although they do have more advanced error correction deelies) but way better at filtering EMI.

Most of this appears to be just due to the shrinking size of components, especially NAND memory. The error correction routines can fix most things but there's a chance the device will need to do a soft reset on itself or even require external input to reload the firmware and programming.

More transistors, etc simply means more to go wrong and more ways an error can propoage through a device undetected.

[CAT-THE-FIFTH]

Also companies are moving to different kinds of process technology for cutting edge nodes. Apparently older SOI based process nodes have more favourable characteristics:
https://www.gsaglobal.org/forums/fd-...-applications/

ATM,the best SOI based process node is made by GF and is 22NM,but they are aiming for 12NM in a few years:
https://www.eenewseurope.com/news/gl...soi-holds-12nm