Are armhf CUDA accelerated Linux viewers a possibility?

[Yakov Oh]

I have a 32-bit Tegra-based linux computer with 192 CUDA cores. Could I expect to see a compatible viewer?

[Jenni Darkwatch]

No. Linux is on the way out for SL. You'll be lucky to even get a Linux viewer in the near future.

[Nova Convair]

There are plenty of viewers for Linux.

But your Tegra is an arm chip I guess that will run android, chrome os or a special linux. The linux viewers will not run on that.

It can probably use SLGO but that you need to try out yourself.

[Callum Meriman]

An interesting thought, but it would run like a dog, assuming everything fell into place.

Compiling for the ARM (Hardware Floating Point) likely isn't so bad - although a lot of work to compile all the libraries and dependancies, and you would never get Vivox (ever) nor the proprietory LibKakadu (using Opensource J2K though instead).

Could a TPV Dev or a sufficiently aware layman do it? I'd say - yes!

The main and deciding issue though will be speed and likely memory, with the killer being OpenGL in my eyes.

The ARM CPU isn't exactly speedy, you can see this just running a standard destop environment. Memory is a big issue, 512MB of RAM won't cut it.

The CUDA cores mean little, the viewer is not using any optimisation there, and honestly, I'd rather use the video cores as shaders than for compute.

OpenGL could be the biggest issue though, the drivers for most low power chipsets are notoriously buggy and feature incomplete in terms of OpenGL.

I'd say forget it. The ARM SOCs are just not up to task to natively run this behemoth application that is the viewer. As mentioned above SLGo (maybe) or if on Android; Lumiya.

---

Jenni Darkwatch wrote:

No. Linux is on the way out for SL. You'll be lucky to even get a Linux viewer in the near future.

---

I don't know where you get that ridiculous idea from. :smileyindifferent: Seriously.

Linux viewers will exist until the day the Lab shutter SL - and then long beyond that.

[Yakov Oh]

"The ARM CPU isn't exactly speedy, you can see this just running a standard destop environment. Memory is a big issue, 512MB of RAM won't cut it."

I have 2GB memory. Unless you mean vram.

[Jean Horten]

Fully featured SL viewers require  some gpu power.

Can your arm machine provide that?

J.

[Yakov Oh]

It should provide enough power. Are 192 CUDA cores enough?

http://www.nvidia.com/object/jetson-tk1-embedded-dev-kit.html

[Jean Horten]

No, cuda does not necessarily mean gpu power.

J.