Vulkan SC 1.0 Coming For "Safety Critical" Graphics / Compute

[Daktyl198]

Oh yay, can’t wait for 10 different Vulkan specs to start popping up like OpenGL used to have. That will TOTALLY make people want to use it over DX12!
/s

[Marc1n]

Does "safety critical" graphics really need to be fancy hardware accelerated eye candy gimmickry?

It seems such usage scenarios could do with something far more basic and simple, and therefore stable.

What if it's safety critical, but not eye-candy? Medical imaging devices for example might need a lot of GPU power, especially that some surgeries require real-time high quality visuals.
And your really don't want to eg. iMRI device to crash in the middle of a brain surgery. Same with 3d renders during ultrasound-guided surgery.
Often simple and basic solutions don't do the job we need them to do.

[geearf]

Why not add that scrictness to standard Vulkan? What is the need for a separate, even if not that separate, API?

[chocolate]

Why not add that scrictness to standard Vulkan? What is the need for a separate, even if not that separate, API?

A quorum must be reached on a subset of Vulkan for it to be worth certifying against a certain coding standard and set of rules such as MISRA. Also, some aspects concerning memory management. Certification is tedious and costly. Management-wise, the less you have to certify, the better. Engineering-wise, the more the ecosystem helps you in that direction, the better. There's not a lot more to it, it's just a palliative anyway.

Compare that with the world of video games, where software houses hack everything together one month before release (to be honest, sometimes it happens in embedded systems engineering as well, but shhh). Nobody wants an SC variant to be the default outside of, well, Safety-/Mission-/Whatever-Critical software engineering.

[geearf]

A quorum must be reached on a subset of Vulkan for it to be worth certifying against a certain coding standard and set of rules such as MISRA. Also, some aspects concerning memory management. Certification is tedious and costly. Management-wise, the less you have to certify, the better. Engineering-wise, the more the ecosystem helps you in that direction, the better. There's not a lot more to it, it's just a palliative anyway.

Compare that with the world of video games, where software houses hack everything together one month before release (to be honest, sometimes it happens in embedded systems engineering as well, but shhh). Nobody wants an SC variant to be the default outside of, well, Safety-/Mission-/Whatever-Critical software engineering.

Thank you!

[Markopolo]

What if it's safety critical, but not eye-candy? Medical imaging devices for example might need a lot of GPU power, especially that some surgeries require real-time high quality visuals.
And your really don't want to eg. iMRI device to crash in the middle of a brain surgery. Same with 3d renders during ultrasound-guided surgery.
Often simple and basic solutions don't do the job we need them to do.

I was going to post something similar but you said this part best.

the only thing I’d add is that ‘safety critical’ can mean very different things to different people but in the case of an API like Vulcan it seems the goal was so that you could use it in any safety critical system. Keep in
mind this is not magical - you’d still need to certify your hardware, particular Vulcan sc implementation and your own application code. But at least it’s possible to do with Vulcan SC now where as before it was not possible because the Vulcan APIs themselves were not certifiable (at the very least it had issues with MISRA C and unpredictable latency)

[skeevy420]

So it's will be like Debian Oldstable but in 3D world.

And that's a good thing. We don't necessarily want Arch in our Planes, Trains, and Automobiles.

[piorunz]

And that's a good thing. We don't necessarily want Arch in our Planes, Trains, and Automobiles.

Some may argue, that we may not even want broken software in our computers. For many people, computer is just as important as trains and automobiles. It must just work and get the job done. That's why there is Debian.

[fogger]

Why would you use anything other than PLCs in safety critical systems?

[bridgman]

Why would you use anything other than PLCs in safety critical systems?

Not sure I understand the question - PLCs have historically not been safety-critical either, relying on external logic ("safety relays") for the safety-critical part.

PLCs with built-in safety-critical logic ("safety PLCs") have become available recently - are you talking about something like that ?

Either way, if someone wanted to implement a PLC with graphical display they would want a safety-critical graphics API and implementation. This has usually been a subset of OpenGL that has been sufficiently cut down to allow safety certification, but now Vulkan will be available as an option alongside OpenGL once certified implementations become available.

I guess the question is whether PLCs and their programming environments are able to support the kind of safety-critical applications being conceived today, like autonomous vehicle control.