AMD To Drop Radeon HD 2000/3000/4000 Catalyst Support

[allquixotic]

Something that works.

And another expansion of Bridgman's team's budget so they can hire 4 more people

[Kano]

@chithanh

Really nice find that AMD even renamed 880G to 980G as well. But i never saw a board using it.

I do not care that much for the desktop users, this time the boards are all PCI-E and you can put in a 20$ card in case of real trouble (at the expense of a bit more power usage). But as always: laptop users are left behind. Basically it is really hard to suggest any laptop that is not Intel only (SNB or newer) to a Linux user. The muxless designs do not make it easy to use binary drivers for AMD/NVIDIA, you have to play with bumblebee or similar if you want. Even disabling the dedicated chip is not possible in every BIOS. But when you got one of those fully stripped down AMD laptops with only onboard it was not that hard to install fglrx, well most of those will need to use oss drivers in the future. I heard that there should be a xvba update coming with full support of h264 l5.1, maybe this at least happens after they dropped support for dx10 hardware. The "funny" thing however would be that some igp chips are using a weird combination of older dx10 gfx core + newer uvd...

[Kivada]

I see this as a good thing across the board, it'll reduce their costs for supporting older hardware on newer Mesa releases as well as having to support pre OpenGL4/DirectX11 class hardware on non Windows 8 installs.

This will also help Gallium3D driver development, hopefully AMD will pick up a few more driver devs.

If you want to move to Intel, fine, they only have an OSS driver and allot more devs working on it then AMD does and yet they also drop support for recentish hardware that is usually still on the market as they tend to still push their very outdated parts onto the market for years.

Case in point, the current crop of Intel Atom based Chromebooks all using the Pineview GMA3150 which is only OpenGL2, 2 Pixel Pipelines@200Mhz and only MPEG2 hardware acceleration, it was a terrible chipset when it was new let alone all these years later. Even the current Intel HD Graphics 3000 GPUs are no competition for what you get from an AMD APU in the same price bracket.

Lastly, who gives a flying fuck about Nvidia's blob support? Nvidia is in a far worse position then AMD is as they only have their GPU business and ARM SoC business, and if they can't keep companies buying Tesla servers their high end GPU business is cooked.

[del_diablo]

I think I agree with some of the critic earlier in the thread: A part of the problem is too on the FLOSS side, there is no part of the API that will remain stable for several years of updates, while on Windows we know for a fact that a service pack won't break it, the only thing that annoys legacy hardware is whenever there is a new OS out. As somebody said, Debian Stable and Ubuntus "longer LTS" will last quite a few years anyhow, so from a standpoint where security updates is a concern its not a problem. Its only a problem due how the Linux infrastructure is built, its built on the idea of rapid development of the entire ecosystem, which can lead to backporting issues.
While, its also correct to blame the blob droppers for their inflexibility, especially because of the way its hooked into the rest of the system, which do mean that something that breaks on every single component update is just a symptom over how exactly its hooked into the rest of the kernel and major "infrastructure". But a more elegant hook would likely break each half a year-year instead of on each update, which also means we have a blob hostile environment in the long run. Which is a good thing, if we only have had the markeding leverage to get manifacturers to assimilate.
And for those of you who say "Nvidia will treat us better", you know what, does your precious randr work yet? The real problem is that as consumers, we don't really have a choice at the moment, its between broken and not working. For those of us who have such hardware, we will be stuck with Ubuntu or Debian, and we can't run Gentoo or Archlinux on the fresh and bleeding edge. There is not really any "premium linux manufacturer" either, it would not surprise me if even the Raspberry Pi will suffer extreme problems with this at a certain point in the future.

"The solution" we are still looking for is that a random manufacturer copies Apple, but use a popular distro instead, and once it get enough marked leverage it can negotiate itself with one of the lesser known graphic manufacturers and get a deal that enables us to get a stable FLOSS brand for what is today a quite proprietary piece of hardware. This will at the least enable us to "fall back" to that brands products whenever something like this happens, presenting us with hardware we actually can acquire for our own greed and future planning.

Something that works.

HEAR HEAR!

[Kivada]

4890 was out in April 2009, so I'm not sure where you're looking for this info...

As for performance - see, the thing is, at least on the Windows side, the 4890 is so powerful that it doesn't need any upgrades to play the latest games. I only have very occasional slowdown in the absolute latest games, and that is because of my CPU, not the GPU! As for gaming on Linux, well, in this case it all comes down to the drivers, and not the card itself. Upgrading to the absolute latest would probably even decrease the performance at this point...

You may have missed the memo that by 2013 the market is supposed to be flooded by 15~21" LCDs at 3840x2160. You're HD4890 can't even produce a 2D desktop at that resolution. Intel Ivy Bridge will support an output of 4096x4096 and the AMD HD6000 series already supports 16000x16000. They won't run Unigine Heaven at frames per second though...

[Kano]

@Kivada

You are right, the raw speed of Intel HD 3000 is not optimal, but Intel HD 4000 is a nice step in the right direction. Some boards allow even to oc the Intel IGP, so you can get about 10% more speed, that's not much, but AMD APUs are not so much faster anymore. Intel will use the HD core for Atom chips next year, so those chips will have got nice Linux support out of the box. I agree with you that this does not help much for older Atoms, also GMA 500/600 are pretty bad on Linux. But Intel definitely has much more developers working on OSS Linux drivers than AMD so you can expect much better support in the future. Initial IVB support was done basically several month before launch, maybe a coincidence that those chips have been delayed at least 3 month because of the buggy SNB chipsets last year (and the dual cores are still delayed most likely due to production issues). But did you ever see prelaunch AMD OSS drivers? AMD is so stupid with their PCI ID whitelist approach that even binary drivers which can be used for new cards as well get a watermark that the hardware is not supported. Well when the hardware would not be supported nobody would see that watermark, absolutely no logic behind AMD driver decisions...

[Alejandro Nova]

Remember neither KWin nor Gnome Shell works well under Catalyst, on *ANY* card. So the supreme irony of this: in KDE 4.9 you'll see how KDE drops support for the Radeon HD 4000 series, while the MUCH LESSER Intel i945 IGP works like a charm.

[smitty3268]

Remember neither KWin nor Gnome Shell works well under Catalyst, on *ANY* card. So the supreme irony of this: in KDE 4.9 you'll see how KDE drops support for the Radeon HD 4000 series, while the MUCH LESSER Intel i945 IGP works like a charm.

Huh? KDE works quite well with the r600g driver.

[Kivada]

AMD still sells the 980G chipset which was released last year (with 4250 graphics): http://www.amd.com/us/products/deskt...g-chipset.aspx
The situation with the 740G (Radeon 2100 graphics) chipset from three years ago is now repeating; no fully-featured drivers for modern Linux systems.

And due to AMD's rebranding-happy marketroids, they drop not only support for 2000/3000/4000 chipsets, but also 5100 and 500V series which are R700 based.

Nobody ever buys from "AMD's store", they are usually by far the most expensive option. In the brick and mortar stores here in the US I haven't seen anything AMD that wasn't based on the APUs in at least 8 months. Online shops still had a few old bits of old backstock hardware.

740G? March 4, 2008 (780G, 740G). 890G? March 2, 2010 (890GX) HD4290. 900 series? It has no IGP chipset. If you want an IGP on current AMD hardware you're looking at getting an APU(CPU/GPU/Memory Controller/Northbridge) Parts using it are the Z, C, E, E2, A4, A6 and A8 series parts makeup the current consumer grade lineup from tablet to midrange desktop. Slowest is the C-30 1.2Ghz 64 bit single core, HD6250 80 shaders@280Mhz. Fastest is the A8-3870K 3Ghz 64 bit quad core, HD6550D 400 shaders@600Mhz

[Kivada]

@Kivada
*snip* But did you ever see prelaunch AMD OSS drivers? AMD is so stupid with their PCI ID whitelist approach that even binary drivers which can be used for new cards as well get a watermark that the hardware is not supported. Well when the hardware would not be supported nobody would see that watermark, absolutely no logic behind AMD driver decisions...

First off, Holy text block Batman!

Yeah, well like I said, Intel has allot more OSS driver devs as well as they have like a decade of lead time(I have no idea when Intel started making their OSS driver) on AMD in producing OSS drivers.

Now though, Intel's HD Graphics 4000 will have to contend with the HD7660D on the A10 series that will be out in a few months, not the current HD6000 series GPUs. The bottom end will probably finally be beaten by the updated Atoms as AMD axed the die shrink to Brazos as it was only going to be on the market about 6 months before the Tamesh series was to go into production, So we're getting a speed bump called Brazos 2.0, Fastest being the AMD E2-1800 which is rumored to be 1.7Ghz dual core 64 bit HD7340 80 shaders@680Mhz.