Hacker News Reader

Comment by rquirk | Hacker News Reader

There seems to be little worry about the way Linux is going wrt Android - and to a lesser extent the Raspberry Pi - that pretty much tie folk down to one version of the kernel since all of the drivers are binary-only.

Ubuntu phone will support the "Android kernel" (as one Canonical guy called it on a recent podcast), since only Android builds of the Linux kernel works with the binary-only drivers that the manufacturers supply. This is just another form of Tivoization. You have the source, but you can't really change anything.

It seems to me that the infamous "Linux in a binary world" doomsday scenario is becoming more of a reality every day https://lwn.net/Articles/162686/ - but instead of servers vs consumer PCs, it is being played out on phones, tablets and laptops.

View on HN

It is not Tivoization. Tivoization is when device manufacturers deliberately lock down their devices so the user cannot change the code even though it is open source.

What's the difference whether your kernel doesn't boot due to signature checks or due to lack of essential drivers?

You might be able to write the drivers without running afoul of the law. If it's a signature check, breaking it could be construed as a DMCA violation (in the US, probably some other law in friendly countries). Neither one's great, but there is a difference.

This is the Linux kernel developers fault to begin with. Microsoft does not break driver API for every minor update of the kernel. In fact with my Radeon x1900 I am running it under Windows 7 with Windows Vista drivers. Runs StarCraft2 and a bunch of newer FPS games just fine.

On the other hand, Linux is a disaster. The kernel devs are so determined to break binary compatibility, I haven't been able to run with ATI's proprietary binary drivers for years. AMD was a good open source citizen and released the specs. And yet the open source drivers for my card are useless for anything but 2D.

I am thankful that Google came along and is forcing a standard in the kernel. I welcome this "binary world".

The approach isn't without merit. Promoting open source is important. If there was a stable binary driver interface we would have far fewer open source drivers. With open source drivers we have added security and the ability to fix bugs. Not to mention the ability to maintain them ourselves if the need arises.

I've heard it argued that it's for this reason that Linux is more successful than other more liberally licensed open source kernels like the BSDs.

Regarding your Radeon card. I Hold Linux support very highly in my requirements for buying new hardware. For this reason I've always bought Nvidia cards over ATI. You can't just buy anything and expect it to work flawlessly. Saving $30 on the purchase of a graphics card has hidden costs. ATI can't even make a decent windows driver let alone a Linux one.

http://lwn.net/Articles/159313/

I'm looking for a tablet device where the graphics drivers are open source, or there are decent reverse engineered open source drivers. So far the list is very short: buy Intel HD graphics hardware.

The problem you raise is exactly why I'm so picky for graphics accelerators: they are most often what is stopping me from upgrading the kernel on a system, and I want to upgrade the kernel/distribution.

I'm not sure about the state of open-source AMD drivers, so if anyone can testify about their graphics drivers, I would appreciate it.

The radeon driver, I heard, works well on hardware that's not a few weeks old most of the time, with slightly better 2D performance and slightly worse 3D performance.

I think there's also the Mali400 chip in the "detect reverse engineered" department, if we're talking about tablets.

AMD has made serious commitments to their open-source drivers. They've released documentation for their chips and have several people working full-time on the OS drivers.

As for supported chips, everything up to "Northern Islands", the last generation, works really well. The latest "Southern Islands" chips also work well but don't have all the features implemented yet.

I use a HD7970 in my workstation. 2D works great and is very fast. 3D is pretty slow, b/c many things aren't accelerated yet. So, for example, minecraft runs about 20fps.

The next generation of AMD gpus "Sea Islands" will use the same drivers as "Southern islands" so the former will be supported as soon as the latter.

Probably the biggest issue is power management. dynpm dynamically changes the clock but doesn't work with multiple heads. The "profile" power management flickers when changing the clock speed.

The reasons for the lack of support for "Southern islands" is that support required many changes that couldn't be release incrementally. Better support is expected when the "Sea islands" chips come out. Check out this post by an AMD developer for a little more detail: http://phoronix.com/forums/showthread.php?73824-Bridgman-Is-...

Check out http://www.x.org/wiki/RadeonFeature for a table of chips and features and information about mapping engineering names to board marketing names.

If you have never experienced kernel mode setting(KMS) and in-kernel drivers I would _highly_ recommend it. It is such a better experience. It's so smooth, switching from xorg to a VT doesn't flicker. And having to upgrade proprietary drivers is a huge pain. I could never go back to UMS and proprietary drivers.

As for my recommendations, if the intel HD3000 or HD4000 meet your 3D needs, they work extremely well and have great in-kernel drivers.

If you need faster OS 3D or a discrete card, go for a "northern islands" Radeon. bridgman recommends a mid-range card for the biggest band-for-the-buck with OS drivers. The high-end cards won't be slower but the drivers aren't as optimized to get the highest performance.

If you are patient and like submitting bug reports you should get a HD7970 ;)

EDIT:

Oh and if you need OpenCL on gpu, intel doesn't support it yet. I believe its works on AMD "northern islands" and earlier.

For more info check out: http://software.intel.com/en-us/articles/opencl-sdk-frequent...

http://www.phoronix.com/scan.php?page=news_item&px=MTI3M...

http://www.phoronix.com/scan.php?page=news_item&px=MTI1N...

Dunno about new Radeons, but my experience with older series is very poor.

HD3000 - very common IGP, yet whenever I enable it there are GPU lockups every few days. This goes on for years, on two motherboards.

AGP Radeon 9600 - occasional crashes with bad kernel/X combinations.

The 9600 works fine on "good" driver versions. HD3000 boxes are unusable without some old 8-series GeForces running under nouveau.

And now, Intel. I've been using a laptop with GMA950 IGP for almost five years. Never a single crash, except for one time when I enabled KMS long before it was cool and everybody's dog was using it. And nowadays this driver runs some Wind0ze games faster than genuine XP on the same hardware.

EDIT: I've just checked that this Intel KMS bug was happening only on some -rc release.

What version of linux, mesa, xorg were you having those HD3000, Radeon 9600 problems on? The "radeon" driver is completely different from the "radeonhd" and the old reverse-engineered xorg ati driver from a few years ago.

And just to be clear, anywhere I said "HD3000" I was referring the integrated graphics chip in various Intel processors not the AMD chip.

The past couple of years AMD has directly supported driver development and the drivers have greatly improved.

9600: no idea, it sits in my parents' box, 300km away and no ssh.

3000: 760G chipset, many (most? all?) Linux/Mesa/Xorg version released in last two years. Completely random lockups, usually at few days of uptime. I no longer use one of this mobos so I'll have a look at this issue if I find some HDD to boot it.

I have been running HD3000/HD4000 for years and never had a crash or lock up. You're not talking about AMD's Radeon HD 3k/4k are you? Because I believe both dmm and I are talking about Intel's HD range.