News for hackers. Stuff that matters.

A comment on Nvidia releases open-source GPU kernel modules. Parent comment

Re:Nvidia releases open-source GPU kernel modules by prash_ant on Wednesday May 11, 2022 @10:34PM (#31348615)
There have been reports earlier [1] that processing power of some GPUs is suppressed by software rather than the hardware capability itself. It is frequently easier to mass produce similar chips than to have different chips for different priced devices. I had come across comments in other online forums where the users alleged that some software flags restrict the capability. I hope someone else will link those webpages if they come across them

[1] https://www.tomshardware.com/news/nvidia-gpu-system-processo...

[ Parent | on HN ]
Re:Nvidia releases open-source GPU kernel modules by azeirah on Friday May 13, 2022 @06:07AM (#31365372)
I suppose this is also used for yield quality?

For instance, AMD is known to sell the same actually-8-core chip as a 6-core if the transistor yield was poorer.

I believe this is a standard in the industry, right? You can theoretically unlock these cores yourself, but there's a decent chance it will break or cause other significant problems.

[ Parent | on HN ]
Re:Nvidia releases open-source GPU kernel modules by imtringued on Thursday May 12, 2022 @08:16AM (#31352872)
They can move those restrictions into firmware otherwise someone is going to develop a tool that unlocks the driver like people did for the Intel compiler.
[ Parent | on HN ]
Re:Nvidia releases open-source GPU kernel modules by hyperionplays on Wednesday May 11, 2022 @10:51PM (#31348745)
this was the case with RTX voice, a simple regedit made it work on non-RTX cards fine.
[ Parent | on HN ]
Re:Nvidia releases open-source GPU kernel modules by cvuls on Thursday May 12, 2022 @12:17AM (#31349221)
From my understanding, it used CPU acceleration instead
[ Parent | on HN ]
Re:Nvidia releases open-source GPU kernel modules by IntelMiner on Friday May 13, 2022 @02:14AM (#31363966)
It would use RT cores by default, but if they weren't available or in use (EG: running a game with Ray Tracing) it could fall back to the older CUDA

That's also why they split it out into two products now

"Nvidia Broadcast" is their "you bought an RTX GPU so here's a shiny toy"

And "RTX Voice" for people with GPU's dating back to the Geforce 400 series that only does it via CUDA

[ Parent | on HN ]