The Initial Performance Of NVIDIA's R515 Open-Source Linux GPU Kernel Driver

[pWe00Iri3e7Z9lHOX2Qx]

Even if it was backported, you'd be lacking userspace support. Maybe nouveau could pull that off, but I wouldn't be holding my breath.

Yup. I wouldn't be surprised if Nouveau eventually gets access to the Maxwell / Pascal firmware and ends up being a good solution. I also wouldn't be surprised if that is a few years away and past the point I own any of those cards.

[Danny3]

Now if only we could have HDR on Linux...

At least for movies, like we can have even on Windows 7 with MPC-HC+MadVR that can send HDR metadata to and HDR capable display!

People say that this is not possible on Linux because there's no stable API for that yet, but I think Windows 7 didn't have an HDR API either and the MadVr developer found a way to send all that information on HDMI along with the movie data.

[nordkamp]

Correct. Kepler ended with the 470 branch which is a LTSB with bug fix support into 2024. Maxwell 1.0 and newer are supported by the current production branch of the closed source binary driver. Only Turing / Ampere and beyond will be supported by the new open source driver.

Is there an official statement on this? I would hope that they eventually release all the source for their current proprietary branch, given they most likely won't want to maintain two different kernel drivers once the open-source drivers gain real traction (which they will). AMD does that anyway though...

If they really want to make an effort to gain community trust and good-favour after the decade of being an asshole I see no reason not to release open kernel drivers for all currently supported parts in future.

[theriddick]

Pretty good start honestly. Does it work with 20series? I know it doesn't support 10 series which is a damn shame as many people still have those (and cards like the 960/80). Maybe in time that will happen also.

[Michael]

Pretty good start honestly. Does it work with 20series? I know it doesn't support 10 series which is a damn shame as many people still have those (and cards like the 960/80). Maybe in time that will happen also.

Don't expect it to be backported to 10 series since it lacks the GSP and this kernel driver is designed explicitly to use the GSP.

[jeoshua]

I'm really surprised there were no gaming benchmarks done, here.

That said, I'm happy that Nvidia is finally opening up their driver, but I'm actually pretty surprised that it's performing at ALL worse than the closed source driver they just pulled it out of. Why is it not at parity? What did they change that hurt performance, like this?

[Michael]

I'm really surprised there were no gaming benchmarks done, here.

Huh, did you miss the two pages of gaming benchmarks?

[pWe00Iri3e7Z9lHOX2Qx]

Is there an official statement on this? I would hope that they eventually release all the source for their current proprietary branch, given they most likely won't want to maintain two different kernel drivers once the open-source drivers gain real traction (which they will). AMD does that anyway though...

If they really want to make an effort to gain community trust and good-favour after the decade of being an asshole I see no reason not to release open kernel drivers for all currently supported parts in future.

This new driver is architected around the GSP which cards prior to Turing lack. Other than high fives from the Linux community which earn them zero dollars, the incentives for them to go back to 6+ year old Pascal and earlier architectures don't seem great. The more realistic approach would be for them to allow Nouveau to use the firmwares for the Maxwell and Pascal cards and lots of reference code from this new driver and any other documentation they need. Basically just give the community what they need to support it themselves.

[pWe00Iri3e7Z9lHOX2Qx]

I'm really surprised there were no gaming benchmarks done, here.

That said, I'm happy that Nvidia is finally opening up their driver, but I'm actually pretty surprised that it's performing at ALL worse than the closed source driver they just pulled it out of. Why is it not at parity? What did they change that hurt performance, like this?

Your surprised that a new, self-proclaimed "alpha" GPU driver that isn't a line for line copy of the original driver is typically 2%-3% slower than the original driver with years of optimizations and bug fixes?!?

[nordkamp]

This new driver is architected around the GSP which cards prior to Turing lack. Other than high fives from the Linux community which earn them zero dollars, the incentives for them to go back to 6+ year old Pascal and earlier architectures don't seem great. The more realistic approach would be for them to allow Nouveau to use the firmwares for the Maxwell and Pascal cards and lots of reference code from this new driver and any other documentation they need. Basically just give the community what they need to support it themselves.

Ah makes sense, cheers. A little disappointing but at least there's hope for Nouveau.