Mesa "Terakan" Driver Aims To Provide Vulkan Support For Old Radeon HD 6000 Series

[aviallon]

Modern GPU standards also cost modern GPU prices. GPU prices have gone down, but not enough for even a GTX 1060 owner to finally upgrade. Also, a lot of laptops still use VLIW5 and VLIW4 that work perfectly fine for even modern games.... on Windows. On Linux if you want to play DX11 or DX12 games then you need Vulkan. Nobody has a Gallium 11 or DX12 to OpenGL abilities on Linux.

Well, there is WineD3D, which does work reasonably well on an old iMac 12,1 featuring a venerable AMD Radeon HD 6670M.
I was able to play Starcraft 2 at 60 fps using it!

[SteamPunker]

Well, there is WineD3D, which does work reasonably well on an old iMac 12,1 featuring a venerable AMD Radeon HD 6670M.
I was able to play Starcraft 2 at 60 fps using it!

I agree, this is a really cool project! I have an old Mac Mini lying around with the same GPU in it. Apple has stopped supporting it since quite a few macOS releases ago, but It still runs Linux like charm.

I've been hoping for some skilled volunteer to take a stab at writing a Vulkan driver for this generation of GPUs. I'm curious to see what performance could still be squeezed out of it, at least in non-demanding Vulkan games. I'm also curious to see how well Zink could be made to perform on top of Terakan, and if such a stack would compete at least reasonably well with the R600 Gallium3D driver. That would be beneficial in the longer term, having only one driver to maintain for both Vulkan and OpenGL.

At any rate, it will extend the usefulness of such older hardware, which is good from a sustainability perspective as well. Even if the resulting Vulkan driver ends up being useful only for basic UI acceleration, that alone would keep it useful for many users.

Thanks for working on this and sharing it with the world, Triang3l! Good luck with it, and have fun.

[gerddie]

A few month back I did a stab at Vulkan on r600 by using lavapipe with r600 as backend didn't display anything though (lavapipe/winsys is not prepared for a HW backend). I hope Triang3l! gets further with a real Vulkan driver

[chithanh]

Is that so?

$ sensors
...
amdgpu-pci-2d00
Adapter: PCI adapter
vddgfx: 700.00 mV
fan1: 0 RPM (min = 0 RPM, max = 3200 RPM)
edge: +49.0°C (crit = +110.0°C, hyst = -273.1°C)
(emerg = +115.0°C)
junction: +49.0°C (crit = +105.0°C, hyst = -273.1°C)
(emerg = +110.0°C)
mem: +0.0°C (crit = +105.0°C, hyst = -273.1°C)
(emerg = +110.0°C)
PPT: 4.00 W (cap = 135.00 W)

I think that is just the GPU power and not the card power? Anyway if you take a high end card today it will consume the same ~20 W idle power that they have consumed 10 years ago. With a midrange card you can maybe get to 8-10 W, that hasn't changed either.

[caligula]

I think that is just the GPU power and not the card power? Anyway if you take a high end card today it will consume the same ~20 W idle power that they have consumed 10 years ago. With a midrange card you can maybe get to 8-10 W, that hasn't changed either.

The output I pasted shows a low midrange GPU. Since the cap basically lists the max TDP for the card, I'd assume this is the actual consumption for the whole card. PPT stands for powerplay tables / package power tracking ?

[chithanh]

Since the cap basically lists the max TDP for the card, I'd assume this is the actual consumption for the whole card.

On AMD cards it is not possible to read accurate values for board power via software. That distinguishes them from NVidia cards, where this is possible. Igor's Lab has an extensive article on this:

https://www.igorslab.de/en/graphics-cards-and-their-consumption-read-out-rather-than-measured-why-this-is-easy-with-nvidia-and-nearly-impossible-with-amd/4/