RADV's ACO Compiler Back-End Now Supported For Older AMD "Sea Islands" GPUs

[phoronix]

Phoronix: RADV's ACO Compiler Back-End Now Supported For Older AMD "Sea Islands" GPUs

The Valve-backed "ACO" compiler back-end for the open-source Radeon "RADV" Vulkan driver has added support now for AMD GCN 1.1 "Sea Islands" graphics cards...

http://www.phoronix.com/scan.php?page=news_item&px=ACO-Adds-Sea-Islands

[HenryM]

Nice. I was just wondering why it was limited to later GCN.

is it just me, or is referring to AMD Graphics Core Next generations by anything other than GCN 1/2/3/4/5 rather confusing? I know I have a GCN 2 GPU, Hawaii specifically, but sea islands, southern islands, GFX8 etc feels like complicated redundancy that tends to lead to sometimes fruitless googling.
I've never even seen GCN 2 referred to as GCN 1.1 before today.

[geearf]

Nice. I was just wondering why it was limited to later GCN.

is it just me, or is referring to AMD Graphics Core Next generations by anything other than GCN 1/2/3/4/5 rather confusing? I know I have a GCN 2 GPU, Hawaii specifically, but sea islands, southern islands, GFX8 etc feels like complicated redundancy that tends to lead to sometimes fruitless googling.
I've never even seen GCN 2 referred to as GCN 1.1 before today.

I've owned a bunch of AMD cards and I absolutely agree!
There are way too many codes for me to remember: the marketing name or names when there is more than one (7980=>280X), the generation (GCN1.1 => GCN2), the chipset (GFX10), the die (Tahiti), and probably much more that I'm forgetting.
While these may be very useful internally, I don't know about externally.

[schmidtbag]

Very cool. Every little difference counts on this older hardware.

[smitty3268]

We haven't heard any AMD discussions in a number of months about plans for shifting GCN 1.1 (and GCN 1.0) to AMDGPU DRM by default.

Well, I think it's been pretty much confirmed by comments from AMD employees here that AMD won't ever do this. That hardware is too old at this point for them to care about enabling new features as anything more than experimental.

[geearf]

For example, https://www.techpowerup.com/gpu-spec...n-r9-390.c2664 vs https://www.techpowerup.com/gpu-specs/radeon-rx-570.c2939 can be used to determine basic differences between the cards from consumer viewpoint.

That's very nice thank you!
Does it include stuff like the on board video decoders as well? I did not see it, but I don't really know what I'd be looking for anymore.

[arQon]

Comparison between RX 570 and R9 390 might be interesting because the theoretical FP32 performance in both cases is about 5 TFLOPS. RX 570 has a newer GCN architecture, but has smaller memory bandwidth and smaller peak integer performance than R9 390.

It'll depend a lot on what workload you're trying to run, but in general (say, games) the 390 should still significantly outperform the 570.

That's one of the reasons why the fairly-artificial lack of support for GCN1 is irritating. It might need workarounds / slow paths for a couple of features, but if those specific features aren't used the high-end Gen1 hardware (7970) will still run rings around the midrange Gen2-4 cards. (Interestingly, AMD's Windows driver even reports the 7970 as "R9 200 series", despite the silicon supposedly being different rather than just a upspeeded rebrand).

I get that there are limits to driver team resources and that HW vendors want people to stay on the upgrade treadmill, but hampering support for older cards that are within the same core arch and still more than functional and performant just means more hardware going into landfills for no good reason.

[geearf]

There are version numbers for "Unified Video Decoder" and "Video Compression Engine" on the mentioned pages.

Oh yeah awesome, thank you!

[Guest]

Well, I think it's been pretty much confirmed by comments from AMD employees here that AMD won't ever do this. That hardware is too old at this point for them to care about enabling new features as anything more than experimental.

Um that's not true. Bridgman himself has previously said that CIK support was enabled by default in their separate kernel branch, the one they recommended for some clients. But for mainline they haven't bothered enabling it by default, and no one knows why.

Anyway, happy R9 390 user here, I thought I would have to try adding support myself. Excited to see if it improves performance.

[Guest]

That's very nice thank you!
Does it include stuff like the on board video decoders as well? I did not see it, but I don't really know what I'd be looking for anymore.

R9 390 only supports H.264 decode and encode. RX 570 probably supports H.265 decode and encode as well. That plus the fact that it's a newer generation card makes it better, even if it doesn't match in terms of raw performance.

I've compared to a 580, and the 580 is waaay better than the 390.

I've tested even on games that were released the same year as the 390, it's performance isn't quite as good as people make it out to be. The Linux support for it is crap as well.