> on PPC it's not that easy... If you have a look at Linux drivers you will realize Polaris
> & Navi only run on Little Endian mode if at all. And I think for Navi you still need
> out-of-tree kernel patches to get them going at all. Also they are far from stable.

Unlike those wannabe coders on Linux, the mighty MorphOS developers surely have the skills (and will) to write endian-agnostic drivers :-) Also, they can write stable code, and Linux kernel patches are irrelevant for MorphOS ;-)

> IMHO 'sweet spot' still would be the Radeon HD2400 - HD7670 cards
> (r600 up to Northern Islands)

Already supported in MorphOS in 2D (except HD6930 to HD6990 based on Terascale 3), even overlay for R700 and below.

Quote:
True. How could I miss that?

Quote:
Also true. 3D support is the key here.

Since work is being done on MorphOS x86-64 perhaps it would make more sense to focus on available modern Radeons and skip the models in-between.

Quote:

For MorphOS x86-64, that would make sense.

For the majority of MorphOS systems in use today (which are primarily AGP based not PCIe) it would not.

I have an (officially) unsupported G5 Quad up and running, and I've had a few different cards installed in it. Its kind of cool that we can run a vastly larger selection of gpus under MorphOS (on this hardware) than can be run under OSX. And Linux usually supports these cards as well. So unless you're married to the idea of using and older, no longer supported OS...

Back to "modern Radeons", what would you recommend the developers focus on?

If the future focus will be MorphOS x86-64, I'd suggest we consider something that is compatible with AMD's APUs.

That brings up a good question (and one that won't get answered).
What X64 platform does the development team intend to target?

Intel based hardware would require a discrete video card to maintain Radeon compatibility, while an AMD based system could offer "modern" AMD GPU capability via an APU (without the need for a seperate video card).

I apologize for taking this thread off course.

>> perhaps it would make more sense to focus on available modern Radeons
>> and skip the models in-between.

> For the majority of MorphOS systems in use today (which are primarily AGP
> based not PCIe) it would not.

Then why develop new graphics drivers for MorphOS/PPC at all? The last Radeon AGP GPUs are in the R4xx series, which has long been 3D-supported. The last AGP-bridged Radeon PCIe GPUs are in the R7xx series. So let the MorphOS team add 3D support for R6xx/R7xx and be done with graphics driver development for MorphOS/PPC, right?

> what would you recommend the developers focus on?

Assuming significant work hasn't already been carried out for GCN1/2/3 driver development, and based on my list there, I'd recommend focussing on GCN4 drivers. On the other hand, taking into account the effort and time needed for graphics driver development carried out by a tiny team, as well as the prospect of MorphOS/x64 (there are no AMD APUs with GCN4, except on gaming consoles), focussing now on GCN5 or even RDNA may make sense.

> If the future focus will be MorphOS x86-64, I'd suggest we consider
> something that is compatible with AMD's APUs.

Ignoring old TeraScale here, that would be GCN2, GCN3 or GCN5, then. Or RDNA for upcoming APUs.

> Intel based hardware would require a discrete video card to maintain Radeon compatibility

Yes, it seems the announced Intel CPU with on-chip AMD GPU went nowhere ;-)

>Yes, it seems the announced Intel CPU with on-chip AMD GPU went nowhere

That was no surprise. I can't imagine AMD would want to give up detailed info on their graphics hardware (and Intel would probably just use such an opportunity to gather info anyway).

>focussing now on GCN5 or even RDNA may make sense.

Or even RDNA2.

I'd have to agree with you. But Mark doesn't seem to like to make jumps (unless you count jumping over the Radeon HD69XX series).

In fact, moving straight to RDNA support seems to be the most practical.


[ Edited by Jim 24.06.2020 - 10:19 ]

> I can't imagine AMD would want to give up detailed info on their graphics hardware

That wouldn't have been the case. The chip was supposed to be two HBM2-linked dies put together into one processor package. No GPU internals required to be revealed to Intel for this.

>> focussing now on GCN5 or even RDNA may make sense.

> Or even RDNA2.

Are there any RDNA2-based APUs or graphics cards out yet that MorphOS developers could now put their focus on?

> jumping over the Radeon HD69XX series

It's good that TeraScale 3 support will be omitted, if true. I'm curious what Bigfoot will jump to from TeraScale 2.

> moving straight to RDNA support seems to be the most practical.

I still would like to see GCN4 support for those potential dual booters. But let's see what Hans will go/jump to next, GCN5 or RDNA :-)

>It's good that TeraScale 3 support will be omitted, if true. I'm curious what Bigfoot will jump to from TeraScale 2.

That would be anyone's guess, I don't ask. But I'd agree that GCN4 makes sense.
And RDNA may be way more power than we need.

> RDNA may be way more power than we need.

Of course it is, as is GCN and most of TeraScale ;-)

Quote:

Agreed, I moved back to R500 in order to maintain OSX compatibility.
That is more than powerful enough for me, and I'm not sure the R300s is all that well optimized.

As to more than powerful enough,on a system basis we're already there, X64 will only take us further into that.

[ Edited by Jim 25.06.2020 - 17:43 ]

Gets interesting when your low end R7-240 and R7-250 become hard to find.
What would determine a shift is if A-Eon can no longer source these for the complete basic AmigaOne X5000 systems. Polaris cards are still available though .

Generally available.

Quote:



[ Edited by Spectre660 28.06.2020 - 17:50 ]

>> Pega2-2GB-RadeonHD.png
>> [...] the card [...] works in a Pegasos II, or Efika.

> Wow, that's seriously cool :-) Thank you for posting this.

This is now being tested with OS4 as well, but instead of a PCI Radeon HD card with onboard PEX8112 PCIe-to-PCI bridge chip, a PCIe Radeon HD card and a separate PEX8112-based PCIe-to-PCI bridge card are used:

https://www.amigans.net/modules/newbb/viewtopic.php?topic_id=9276

Let's see how this works out in comparison.

Update:

> Let's see how this works out in comparison.

Seems I missed out on the success of this project for the past 3 months. Pegasos II with OS4 and Radeon RX card running GPU-decoded videos and 3D software:

https://www.amigans.net/modules/newbb/viewtopic.php?post_id=149892#forumpost149892