R600g AMD Evergreen Geometry Shaders Proposed

[airlied]

Then what is your comment about? Airlie clearly said that the main reason why it's Evergreen+ is that he doesn't have earlier cards installed at the moment, and doesn't want to enable the functionality blindly (and for good reason, seeing that it has the potential to cause GPU hangs).

he's just bitter fglrx doesn't work anymore :-)

but really not sure why he thinks I'd delay releasing evergreen code until r600 worked, seems like he's not used to open source development.

I did plug an rv770 in and wrote some code, so far it hangs and it looks like it'll need at least one patch to the kernel CS checker for geom support.

http://cgit.freedesktop.org/~airlied/mesa/log/?h=r600-geom-shaders-r7xx

Dave.

[OneTimeShot]

I did plug an rv770 in and wrote some code

LOL - I like this approach to planning cross-company development. I assume there was some kind of plan that the AMD guys would write the SI code, and you'd have a go at the older cards.

[airlied]

LOL - I like this approach to planning cross-company development. I assume there was some kind of plan that the AMD guys would write the SI code, and you'd have a go at the older cards.

no I had no idea AMD were working on radeonsi when I started this, Red Hat didn't have a need for it either, it was just a personal side-project that I did while avoiding doing my real job.

I also didn't start the r600g code, Vadim wrote most of the initial code in his own time, I'd never have started it otherwise.

Dave.

[TemplarGR]

Then what is your comment about? Airlie clearly said that the main reason why it's Evergreen+ is that he doesn't have earlier cards installed at the moment, and doesn't want to enable the functionality blindly (and for good reason, seeing that it has the potential to cause GPU hangs).

The preference for newer gpu series in the radeon driver is a generic phenomenon. It is not simply about these patches from Airlie.

In general, the radeon devs, AMD and non AMD alike, focus on newer gpus, and forfeit the older ones, even though the older are without catalyst support.

It is this focus that i don't like.

For example, Kernel 3.14 will have dpm on by default for even newly released hardware. Even though my 3870 still has flaky dpm support and not catalyst support... This is wrong, and it is not the best way to get customers...

[TemplarGR]

@TemplarGR
Buy NV or Intel and stop trolling in each topic related to Radeon cards. I have Radeon in my main PC and I can say that it was really good choice. Open source drivers are really great, but Catalyst are also not that bad as someone suggests, I use Catalyst most of time and it works without problems.

I am sorry, did i hurt your feelings?

It is not "trolling", it is a legitimate complaint.

And rest assured that if this attitude continues, me and quite a lot of people i influence for hardware buying decisions will be choosing Intel+ Nvidia for now on... Now this will help AMD, i am sure...

[Calinou]

For example, Kernel 3.14 will have dpm on by default for even newly released hardware. Even though my 3870 still has flaky dpm support and not catalyst support... This is wrong, and it is not the best way to get customers...

They will ultimately get customers from not caring about their old hardware. Planned obsolescence.

[airlied]

The preference for newer gpu series in the radeon driver is a generic phenomenon. It is not simply about these patches from Airlie.

In general, the radeon devs, AMD and non AMD alike, focus on newer gpus, and forfeit the older ones, even though the older are without catalyst support.

It is this focus that i don't like.

For example, Kernel 3.14 will have dpm on by default for even newly released hardware. Even though my 3870 still has flaky dpm support and not catalyst support... This is wrong, and it is not the best way to get customers...

focusing on older hardware takes a lot of time for not as much return, its just economics 101.

The effort to fix DPM on r6xx is a lot harder than the effort to fix DPM support on r700 and so on down through time, as the only people who understand how dpm worked on those chips have long since moved onto something else and there are no good documents on it. Yes it might be possible to figure it out from old fglrx source code but it probably won't be that simple.

The same goes for these things, its a law of diminishing returns, I'd rather spend time fixing something for 1000s than 100s than 10s etc. and r6xx is on the long tail of things I care about anymore.

Dave.

[GreatEmerald]

In general, the radeon devs, AMD and non AMD alike, focus on newer gpus, and forfeit the older ones, even though the older are without catalyst support.

This is wrong, and it is not the best way to get customers...

So focusing on old hardware and ignoring the hardware that is currently for sale is the best way to get customers?..

[curaga]

The preference for newer gpu series in the radeon driver is a generic phenomenon. It is not simply about these patches from Airlie.

In general, the radeon devs, AMD and non AMD alike, focus on newer gpus, and forfeit the older ones, even though the older are without catalyst support.

You can't really expect anything from the volunteers. For example, I was asked about r300g, but since I have no such cards, what could I do?

Surely you don't expect volunteers to go purchase card X to please you.

[TemplarGR]

So focusing on old hardware and ignoring the hardware that is currently for sale is the best way to get customers?..

No one will currently buy a newly released radeon product to use with the open source driver. Only a small niche. Most will use Windows and Linux with fglrx.

Actually, far more people currently use r600g than radeonSI.