Intel Posts New Linux Driver For "Versatile Processing Unit" Coming With 14th Gen CPUs

[phoronix]

Phoronix: Intel Posts New Linux Driver For "Versatile Processing Unit" Coming With 14th Gen CPUs

Intel has posted a new open-source Linux VPU driver today... Not Video Processing Unit, but it's for a Versatile Processing Unit coming with 14th Gen Core processors...

https://www.phoronix.com/news/Intel-VPU-Driver-Linux

[tildearrow]

And here I thought "Versatile" meant something like an FPGA inside the CPU.....

[timofonic]

I wonder if there will be a proper subsystem for FPGA and AI accelerators someday.

[Developer12]

Shit sounds a bit like a habanna labs accelerator that was moved on-die. Not quite the same though.

[qarium]

"The firmware for the Intel VPU will be closed-source."

no thank you intel. i would had consider switch to intel with open-source firmware but no intel goes the shit road.

[timofonic]

"The firmware for the Intel VPU will be closed-source."

no thank you intel. i would had consider switch to intel with open-source firmware but no intel goes the shit road.

I agree. Massive closed source firmwares are progressively infecting everything, even more than in the past and now. This has to end someday, it sucks.

[kiffmet]

It seems like it's time to rename DRM in order to reflect that it's not limited to GPUs anymore. Maybe to something like AIF (Accelerator Interface Framework), GAS (General Accelerator Subsystem) or ACM (Accelerator Control Manager)?

[stormcrow]

I agree. Massive closed source firmwares are progressively infecting everything, even more than in the past and now. This has to end someday, it sucks.

Uh. Don't know what rosy past you're thinking of, but firmware has almost always been closed source. FOSS firmware is an aberration to the norm. It's only been taken for granted that's the case because it was also nearly always on board and static rather than loaded on board bring-up like it often is now. It's more noticeable now because drivers require firmware to load on board bring-up (to save a nickle on an NVRAM chip - gotta love unbridled greed) and security researchers are now publishing their results into their reverse engineering efforts and not liking what they find.

Even the early commercial computers where nearly everything else was relatively well documented, the 'firmware' that operated the basic system was closed source (not libre) even if you could purchase manuals that described the interfaces.

[timofonic]

Uh. Don't know what rosy past you're thinking of, but firmware has almost always been closed source. FOSS firmware is an aberration to the norm. It's only been taken for granted that's the case because it was also nearly always on board and static rather than loaded on board bring-up like it often is now. It's more noticeable now because drivers require firmware to load on board bring-up (to save a nickle on an NVRAM chip - gotta love unbridled greed) and security researchers are now publishing their results into their reverse engineering efforts and not liking what they find.

Even the early commercial computers where nearly everything else was relatively well documented, the 'firmware' that operated the basic system was closed source (not libre) even if you could purchase manuals that described the interfaces.

I know it. The issue is that firmwares are powered by more powerful SoC-microcontrollers and even more is hidden to the hardware drivers. The obfuscation level is exponentially increasing.