AMD's Hiring Open-Source Graphics Developers Still

[crazycheese]

You can continue telling me how it doesn't work, I'll keep using them.
Good luck changing resolution or hotplugging monitors using your nvidia card. As long as something is not needed for Windows, you don't get it in your "linux" driver either.

I was not telling you they don't work, I was telling you they don't project even 20% of features of the card. If I had IGP only, I would use them. But at current development rate, they will be useable for me in 10 years, yet I won't buy last-gen card just for opensource driver to work no more nor believe anything amd marketing says. Lesson learned.

[monraaf]

But they're not documenting anything, and AMD is documenting 99% of the stuff. Surely, that is a huge difference.

Not really. They may have opened up the stuff that's interesting or relevant to the way you use your GPU. But what about the guy who doesn't care about gaming but just want to use his brazos powered HTPC or netbook to watch some HD content at home or when commuting or traveling?

These APUs or simply not powerful enough to decode HD content without using the UVD block. So they are forced to use the blob with all its deficiencies.

If it was the other way around and AMD documented UVD but not the 3D engine would that be okay with you? I think not. I think we'd so a lot of people bitching why AMD has not opened up the 3D engine.

Different people different needs.

[bridgman]

These APUs or simply not powerful enough to decode HD content without using the UVD block.

Just curious, is this just a guess or do you have anecdotal evidence ? I haven't had a chance to run a multithreaded H.264 decoder on the most common APU configuration today (a dual-core Bobcat) and don't know anyone else who has tried.

On the GPU side, AFAIK the code to offload decode work to shaders is just being implemented now so I don't think we know the answer yet.

[monraaf]

Just curious, is this just a guess or do you have anecdotal evidence ? I haven't had a chance to run a multithreaded H.264 decoder on the most common APU configuration today (a dual-core Bobcat) and don't know anyone else who has tried.

Well that's what I read from other tech sites focusing on Windows. If they are powerful enough and don't really need UVD then why bother wasting silicone die in the first place

[V!NCENT]

Your logic has some fails. First, the real bottleneck is hard disk. The NANDs are very low quality and very high price/performance now, so there is currently no ways to cut costs. My 10 year old athlon 3200 with 1gb of ram was sufficient over whole period, quad cores inclusive.

Hah! I have an AMD 2800, 1GB Corsair RAM with perfect timings and a 9800 pro. You are simply lying -.- It sucks. But now with the open source drivers it could accelerate full HD Mpeg4. Take that! Even the blob doesn't work there anymore.

But it wasn't the point of discussion.

Sadly moronic comments from you are, which are in desperate need to be shot down

The point is, if we project the current AMD opensource driver state to RAM chips(thanks to von Neumann we don't need a driver for them)

Your OS doesn't need to interface with your x86 CPU and registers directly? Seriously, what are you running? No wonder you don't get it with all this closed source crap. Hmpf...

you would have 2GiBs free out of 8GiBs installed. You invest 60 dollars and have 15 dollar effect. Consuming whole as-if "60 dollar" electricity. Producing as-if "60-dollar" heat. Delivering as-if "60 dollar" features. Thanks only to driver. Which AMD writes happily for windows for free.

Oh jesus nVidia produces cards that suck the suns energy directly. It is a wonder that you can't cook an egg on it! BTW the Radeon card runs at 100% load. It is just that what it is running at 100% load is not as efficient as the blob.

Not always the case. Low-end cards exist even now, and they cannot beat 10 year old hi-end cards. Nothing out of the ordinary though.

No shit. Why do you think I have not bought a low end card? It is not as if it isn't a great investment to have a decent card with enough power. And the blob isn't creating any advantage in speed with vsync in KDE.

You lease your card, if its something new to you. The hardware is not covered by GPL, but belongs to AMD/Nvidia. You bought the ability to use your card only.

What on earth. There is no law possibly enforcable that can make AMD take back my card, especialy because it isn't even manufactured and sold by AMD. I have no contract. I paid for ownership, and if they were to take it back I would call the cops and they will be arrested for theft.

I don't need your screencast, I used git versions of opensource driver (et al components, libdrm, kernel, mesa etc) with 4770 for more than a year and I had kernel panics, low performance and low features. Hence I use nvidia now.

Ah so you fucked up and went to nVidia for their blob.

[V!NCENT]

I don't need your screencast, I used git versions of opensource driver (et al components, libdrm, kernel, mesa etc) with 4770 for more than a year and I had kernel panics, low performance and low features. Hence I use nvidia now.

PS: Why didn't you use Fedora if you appearantly can't make a propper Gentoo instal? I know why I'm running Fedora....

[Qaridarium]

I haven't had a chance to run a multithreaded H.264 decoder on the most common APU configuration today (a dual-core Bobcat) and don't know anyone else who has tried.

the answer is simple to get use 1 single core of a modern phenomII or corei7 overclock it to 3.2ghz or a little more.

and the answer is no..*G*

2 points because no: 1) the multicore scale is only 90% at best 2) the bobcad cpu do have less cache overall and need the ram for the gpu to.

On the GPU side, AFAIK the code to offload decode work to shaders is just being implemented now so I don't think we know the answer yet.

with only 80 shader-cores.. and less memory bandwidth than a hd2900 the answer is no
but yes maybe much faster than the cpu solution.

i think the reverence implementation of an shader based solution on the R600 gpu architecture is the hd2900...

and i don't know how many shaders are in use in that solution.

[glisse]

I was not telling you they don't work, I was telling you they don't project even 20% of features of the card. If I had IGP only, I would use them. But at current development rate, they will be useable for me in 10 years, yet I won't buy last-gen card just for opensource driver to work no more nor believe anything amd marketing says. Lesson learned.

You want to have top 3D performance for games & GPU video playback working right now under linux, fine NVidia binary is effectively a very good choice for this. But consider that not all user have same need or wish as you. Many are happy enough with working desktop and CPU video playback if they have a fast enough CPU.

But please stop pretending you know anything about the open source driver, you just clueless. Less than one year ago r600g wasn't doing much. As of today it capable of running many games with decent performances. So in less than one years we have made huge progress on support you might think that's slow, well maybe it's but consider the number of people working on open source vs the number of people working on closed source.

And no, neither AMD or NVidia sees linux as true market beside workstation and GPGPU. So don't complain to them that your GPU is not supported under linux, it's just not a huge market to them. If you really want more linux support complain to ASUS,ABIT,DIAMOND,... ie the company to who you actually buy your GPU from.

And we target way more than 20%, our aim is to support all we can.

[agd5f]

with only 80 shader-cores.. and less memory bandwidth than a hd2900 the answer is no
but yes maybe much faster than the cpu solution.

i think the reverence implementation of an shader based solution on the R600 gpu architecture is the hd2900...

and i don't know how many shaders are in use in that solution.

What reference implementation are you talking about? An HD2900 should have ample bandwidth and shader power for decode. The gallium decode work is just being implemented now and Christian is working on an RV710 which is the same level, but a generation behind the GPU in the ontario APUs and he is getting good results.

[agd5f]

Also, note that whether you are using UVD or 3D, they should have similar bandwidth requirements for decode.