An Open-Source Radeon HD 4670? Sort Of.

[phoronix]

Phoronix: An Open-Source Radeon HD 4670? Sort Of.

When the Radeon HD 4850 and Radeon HD 4870 were introduced earlier this year, it was wonderful. These latest high-end graphics cards from ATI had same-day Linux support through their Catalyst driver and the open-source ATI drivers had "just worked" with the RV770 series. The mode-setting support with the xf86-video-ati and xf86-video-radeonhd drivers just required adding in the PCI IDs for these new PCI Express graphics cards and then the rest of the magic was provided by AtomBIOS. However, with the introduction of the Radeon HD 4600 series, not everything is working instantly with the open-source drivers.

http://www.phoronix.com/vr.php?view=12850

[curaga]

I'm looking at the budget HD4000 card, 4350. Any hints on when it'll be out, or how it'll run?

Edit: also a typo, instead of 4670 there is 4780 mentioned ;)

[zombiepig]

Small mistake I think:

"The xf86-video-ati driver has been using AtomBIOS since last year when they decided to add support for the R500/600 series, while it wasn't until July when they finally adopted AtomBIOS under pressure from AMD."

Shouldn't that read "...while it wasn't until July when the RadeonHD driver finally adopted AtomBIOS..."?

[libv]

Small mistake I think:

"The xf86-video-ati driver has been using AtomBIOS since last year when they decided to add support for the R500/600 series, while it wasn't until July when they finally adopted AtomBIOS under pressure from AMD."

Shouldn't that read "...while it wasn't until July when the RadeonHD driver finally adopted AtomBIOS..."?

RadeonHD had been using AtomBIOS from before r6xx support was added to -ati at all. We just limited our usage of AtomBIOS to what we saw as appropriate. We now are using much more of AtomBIOS for the newer chips, but still don't use all of it as some of AtomBIOSes available functions are plain useless.

An example of that is the support for I2C. Even -ati doesn't use that functionality. And there radeonHD uses the HW i2c engines directly instead of banging io lines directly.

With the introduction of RV770, we just needed new code for the MC subsystem, and then the lesser-AtomBIOS code magically all worked. Later on we adopted a more intrusive use of AtomBIOS for rv770.

I have no idea what needs to happen for rv730, we have received no news from our partner about this yet.

[izual]

Things are slowing down aren't they?
No real updates for either video-ati or radonhd.
Same situation on the ati spec front. What's going on? The summer break should be over...

[yoshi314]

Same situation on the ati spec front. What's going on? The summer break should be over...they way i see it - they are probably waiting for dri2,gem and gallium.

most likely people are now waiting for airlied who was porting radeon driver to gem in his branch, till it gets ready. just a speculation on my side, though.

some developers are hacking and experimenting with new technologies on their private branches.

also, radeonhd devs are known to sit quiet for a while and then drop ~100 commits on their driver at once, once their accomplish something fairly significant.

[bridgman]

Actually yesterday was "first day back" for a lot of the developers. Between vacations, XDS, and a week-long SuSE offsite event it's been pretty quiet around here. Should start to pick up again soon.

On the spec front, 6xx/7xx 3D is by far the most difficult phase of the project so far, both in terms of IP review and getting the drivers working on the hardware. Normally we can "release the easy stuff" and let developers start working while we finish working on the more sensitive bits, but with the 6xx/7xx we couldn't find a useful way to subdivide the information so we're working in an NDA repository for now.

I haven't seen HD46xx boards locally yet but libv emailed this morning to say that he had seen listings for them in Germany.

I should probably mention again that we don't expect AtomBIOS to completely hide the changes (the digital outputs on HD46xx are "more different from rv770/HD48xx" than the rv770 was from previous chips) but we do expect it to reduce the time and effort required to get new GPUs running with the open drivers.

[izual]

...and gallium.


I hope you are wrong on this one!

[yoshi314]

yeah ,me too :]

[bugmenot]

I hope you are wrong on this one!
Why? What's bad about integrating Gallium?

[izual]

Nothing is bad about Gallium, but it will not be ready for a looong time I think.

[Svartalf]

Nothing is bad about Gallium, but it will not be ready for a looong time I think.

Something like another 6 months or so, from my gauge of things right now. It's a bit long, yes. But, if it's within that timeframe, it'll be worth the actual wait.

[chrisr]

Something like another 6 months or so, from my gauge of things right now.

I'm currently suffering with an AGP Radeon 9550 and the Mesa driver, but don't want to tax my power supply or ear drums much further with a HD3850. A low-power, passively-cooled AGP HD4670 card sounds like a very good upgrade path to me. But judging from the state of the current R300 OpenGL support, all the effort is going into the Gallium branch.

[Svartalf]

AGP is dying, unfortunately...

As for R300 support, Gallium3D will benefit the R300 cards as well. As for the AGP story, you might want to consider saving up for a more modern machine. Some of the titles, etc. for gaming will chew up that machine and spit the seeds out. ;)

[izual]

Something like another 6 months or so, from my gauge of things right now. It's a bit long, yes. But, if it's within that timeframe, it'll be worth the actual wait.

That IS a looooong time for me ;)

Well I need a graphic card for an older machine, so I think about putting my 2600Pro in the older machine and buying a 4670. That's the choice between two bad (or not good...) supported cards... (from an open source driver point of view)

I really hope that the Catalyst 8.10 will get decent 4670 support and hopefully less bugs! (tearing, 2D performance)

[chrisr]

AGP is dying, unfortunately...

As for the AGP story, you might want to consider saving up for a more modern machine. Some of the titles, etc. for gaming will chew up that machine and spit the seeds out. ;)

AGP may be dying, but by all accounts there will still be an AGP version of HD4670. And a dual P4 machine with 2 GB of RAM doesn't just lie down and gather dust, either ;). (The only failings I've really noticed recently are no NX support in the CPUs and the lack of PCI-Express.)

[glisse]

AGP is dying, unfortunately...

AGP dying is a good things, AGP is a painful technology with many buggy hw and it's a source of many problems we have. PCIE is more solid. Just to state my happiness of seeing AGP dying, even though the quicker the better.

[chrisr]

AGP dying is a good things, AGP is a painful technology with many buggy hw and it's a source of many problems we have.

I have an AGP Matrox G400 MAX card that can currently thrash my AGP Radeon 9550 card running simple OpenGL screensavers, for example.

[The_Monkey_King]

I am in the same position as I was two years ago. At that time I bought the gigabyte mobo which touted the strengths of the 690G (x1250) IGP yet lefdt nothing but ash in my mouth as the *nix drivers were no where near the features of it's windows driver counterparts.

Yes, AMD/ATI has made great strides in trying to obtain parity and will continue to do so. However, the world and card development has not stood still.

So back to my purchase. Why would anyone buy and IGP unless it's for an appliance, mainly an HTPC. And that is exactly why I bought it. The US is finally moving forward with digital TV, yet we are still trapped by PCI bus architecture and analog TV video cards for *nix platforms.

Which brings me to the right here and now. I see the new 780GX with full [onboard] HD video and Dolby sound. Yet, if I buy the gigabyte mobo this week, I will use even less of features than I am right now with 690G, right?

As much as I hate to do it, I will be stuck with buying an WinXP-64 OS and building an HTPC interface from either MS Multimedia console (MMC) or buying a SageTV build just to use the basic elements of this new IGP.

I know I am asking a lot from where just a few years ago I would have had nothing at all, but still, this is a company in competition for as many consumers as possible. It still boggles my mind that the company touts more and more hardware releases with just a fraction of its capabilities being available -- and exepcting the buyers to be happy with it.

You want to beat Intel? You want to bury nVidia? Then spend the money and the resources here and now and have fully compliant, full featured *nix drivers out by Christmas. If they cannot do it alone, then ask Canonical for some cash and people.

[deanjo]

If they cannot do it alone, then ask Canonical for some cash and people.

What makes you think that Canonical has the people that can get the job done? Heck they aren't even on the radar when it comes to X development. Same with the money, why should they have to flip the bill for driver development?

[RealNC]

When the Radeon HD 4850 and Radeon HD 4870 were introduced earlier this year, it was wonderful. These latest high-end graphics cards from ATI had same-day Linux support through their Catalyst driver and the open-source ATI drivers had "just worked" with the RV770 series.

ROFLMAO.

I've stopped bothering about Phoronix anyway, but this one was hilarious. There's your objectivity. I suspect someone here is getting paid for sucking up to AMD so much.

[bridgman]

So back to my purchase. Why would anyone buy and IGP unless it's for an appliance, mainly an HTPC. And that is exactly why I bought it.

I'm not sure I agree with this (other than "why you bought it" ;). It is probably true that most HTPCs use IGP, and it is probably true that most custom-built PCs with IGP are intended for use as HTPC appliances, but I doubt that more than a few percent of IGP systems are purchased as HTPC appliances. IGP is how you get an inexpensive, basic general purpose system and, rightly or wrongly, we are trying to support those general purpose users first before doing anything special for HTPC on Linux.

The US is finally moving forward with digital TV, yet we are still trapped by PCI bus architecture and analog TV video cards for *nix platforms.

I think the "analog" part is a function of the legal and technical requirements around digital TV -- OTA seems do-able but I haven't figured out how to support OpenCable on Linux. It's not our IP so we can't open it up even if we want to. AFAIK the reason capture cards use PCI is because it is cheap and it works :D

Which brings me to the right here and now. I see the new 780GX with full [onboard] HD video and Dolby sound. Yet, if I buy the gigabyte mobo this week, I will use even less of features than I am right now with 690G, right?

You are absolutely right. We should have had acceleration support for 6xx in place a few months ago. We lost a few months at the start with the hard-coding vs AtomBIOS debate, and it took a few months longer than I expected to get the 6xx/7xx 3D engine running but we have it working in house now and should be caught up on support pretty soon. All I can say there is "sorry but we're working as fast as we can".

As much as I hate to do it, I will be stuck with buying an WinXP-64 OS and building an HTPC interface from either MS Multimedia console (MMC) or buying a SageTV build just to use the basic elements of this new IGP.

Is it feasible to run with X11 output and shadowfb for the first month or so and then you could stay with Linux ? I don't think it will take long to get Xv acceleration happening -- less than 3D, anyways.

I know I am asking a lot from where just a few years ago I would have had nothing at all, but still, this is a company in competition for as many consumers as possible. It still boggles my mind that the company touts more and more hardware releases with just a fraction of its capabilities being available -- and exepcting the buyers to be happy with it.

The tricky thing here is that some of the hardware was explicitly designed for the Windows environment where there is a certain amount of OS-provided security allowing us to use the hardware without violating agreements. All the hardware vendors have the same problem here -- I don't think any of us are making use of all the hardware we do under Windows yet, although we are trying hard to make sure you have the same functionality even if it's not implemented the same way.

You want to beat Intel? You want to bury nVidia? Then spend the money and the resources here and now and have fully compliant, full featured *nix drivers out by Christmas. If they cannot do it alone, then ask Canonical for some cash and people.

Throwing more money at the problem would not have made things go noticably faster over the last few months unless we were to steal senior technical people from our own HW design projects (the HW folks provided a lot of help, but we could not take them away from the design of future GPUs for more than short periods of time). We were essentially single-threaded on the task of getting the 6xx and 7xx 3D engines running with "our own code" rather than a few million lines of closed source driver, and that was more difficult than I expected.

[The_Monkey_King]

Yes, indeed, it was by all definitions a Herculean task to get as far as both AMD/ATI and open source (OpenSUSE - Novell) have come. I really do appreciate and it really is breaking new ground on how Linux drivers are now published and even announced. Please do not take my grousing as undermining your achievements.

That said, my point is, up to now, I cannot recommend Linux as an HTPC platform for anything but antiquated hardware. I also cannot recommend and Linux solution from a software and OS standpoint to any non-technical client. The reason is simple, with Microsoft (and please I am no fanboy and I wish I never had to use the platform) I can get support for current hardware and have many avenues for phone support. Linux cannot provide that.

Now, I do believe that people are not dumb and can be taught how to use and troubleshoot Linux and open source applications. The idea of "it just works" may never happen in reality. And we must all help to dispel any notion that anything Linux or Microsoft is trouble free. We can however help people by giving the opportunities to help themselves and to empower them to actually feel comfortable about fixing or restarting should an issue arise.

But I am getting off point. The crux of my comment about Canonical was Mark's latest posting on his website that said Canonical must do more than just be the head company of Ubuntu. He says his company should and will be more involved with the whole notion of open source from OS to application to whole user experience.

I think he sees where all this heading. When Fiesty Fawn came out, it was an eye opening experience for the non-Linux computing world. Yes, there may be other distros that provide better functionality or better this or beeter that, but it was this version that had CNN and NBC and other general public forums telling people, yes, you can have something else besides Microsoft and it's FREE!

So why has it not taken the world by storm? Well, because companies that build the hardware are not up to the same level of support or functionality as their own Microsoft development. And there's where you and I cannot give a brand new laptop or desktop to grandma whithout a whole lot of disclaimers and telling her no, you cannot do this like you can with WinXP. So Canonical is looking to help and pay for, it seems, to get the rest of the open source community kicked into high gear and have the things grandma can use in the very near future.

Does he want to be the next Gates and the next Microsoft? Well that's hard to say. Even the most benevolent intentions sometimes become dictitorial mandates -- old episodes of Star Trek pretty much spell this out. But I think his point is clear, he can spend the money, hire the resources to help out the whole Linux community. All I can say is trust, but verify where and how his intentions are aimed.

Again, let me express my gratitude at the entire Linux community for giving me not only options but the opportunity for me to join. Do not let my judgement of the here and now let you get discouraged or upset.

If it means anything, I just bought the gigabyte GA-MA790GP-DS4H (790GX) and will be using it as a testing platform for OpenSUSE and Ubuntu.

Thanks and good luck.

[nightmorph]

It's been long enough that there's some support for the 4670 in Linux, apparently with the open-source drivers, from what I've heard.

There's no doubt that for gaming, the proprietary Catalyst driver is the one to use, but of the two available open-source drivers, which is better for pure desktop work, and not gaming? xf86-video-ati, or the xf86-video-radeonhd driver?

And here's another question: which driver seems to be making the fastest progress?

I'm considering purchasing a 4670, but I'm very pragmatic, so I want to get the most use out of my card now, rather than indulge in long-term wishful thinking.

[elanthis]

IGP is how you get an inexpensive, basic general purpose system and, rightly or wrongly, we are trying to support those general purpose users first before doing anything special for HTPC on Linux.

The biggest users of IGP I would imagine to be laptop users. Outside of the very small percentage of laptops sold as "gaming rigs" or "desktop replacements," they're pretty much all IGP.

Laptops are increasingly important to computer users in general, from business types to school kids to café addicts like me. :) Especially with Linux' justly-earned, horrendous place in the gaming market (which is what most consumers buy machines with powerful dedicated graphics cards for), I pretty firmly believe that a core priority for Linux needs to be 100% solid support for laptops, including their graphics components.