Open Graphics Project Status Update (KernelTrap) [LWN.net]

Open Graphics Project Status Update (KernelTrap)

Posted Oct 4, 2005 23:17 UTC (Tue) by sbergman27 (subscriber, #10767)
Parent article: Open Graphics Project Status Update (KernelTrap)

I can't say that I really see a reason for this to take off commercially. They *think* that by the end of 2006 they'll have something that has medium OK performance by today's standards, and that it should come in around $120 per card. And they see no reason for that dollar figure to increase, despite the fact that their previous expectations didn't pan out due to unexpected (though predictable) problems.

There is certainly an "I believe in freedom and I am willing to vote with my wallet" aspect to this.

I might even buy one myself, though I suppose it will be mail order only.

But I can't help but put in my prediction that this thing is gonna be a commercial flop.

In the end, I can't help but think that in the broader community, it will be known, to the extent that it *is* known, as "The Open Source Hardware Project That Didn't Work".

And I would be delighted to be proven wrong on this one. :-/

(Log in to post comments)

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 0:25 UTC (Wed) by tjw.org (guest, #20716) [Link]

There is certainly an "I believe in freedom and I am willing to vote with my wallet" aspect to this.

There is also the potential slide towards 3D-only graphics cards. If it becomes the trend to stop offering 2D acceleration on newer cards after Vista use becomes the norm, we will be in an unpleasent situation We will be forced to either use proprietary OpenGL drivers for X acceleration, or not use newer cards.

Right now it's just the game players (myself included) that rely on proprietary NVidia/ATI drivers, but if it becomes everyone who uses X, the market for a card like this will surely be there.

It's also important to note that the Radeon 9200 is the end of an era. Unless this OGC card is made, it will be the end of open 3D drivers. Not to mention three companies will have essentially leveraged themselves into control of the Linux desktop.

Personally, I would buy this card for the tinkering value alone. The potential path to salvation this card may bring makes it even more appealing.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 0:39 UTC (Wed) by sbergman27 (subscriber, #10767) [Link]

Good point.

Thanks for helping to prove me wrong. :-)

I'm not sure I believe that the 3D only era is upon us. But it is best to be prepared.

Just as Samba is good, not because SMB is good, but because we need it to compete.

And just as Mono is good, not necessarily because .NET is good, but because we might need it to compete.

So might we need Open-Source graphics hardware for OSS software to compete.

I hope it doesn't come to that. But if it does, we would have it.

Like I say, good point.

-Steve

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 2:27 UTC (Wed) by drag (subscriber, #31333) [Link]

The 2D era is pretty much dead right now anyways.

Check out previous coverage on LWN.net...
http://lwn.net/Articles/149716/
:-)

Specificly look at this high-level diagram of a modern video card:
http://dri.freedesktop.org/%7Ejonsmirl/architecture.gif

See all the subsystems and different acceleration and logical bits and peices? That's all for 3d acceleration. (which is perfectly capable of displaying 2d stuff itself)

See the little gray box upper-left that is labled '2d engine'. That's for 2d acceleration. Most of that would be a unchanged copy of the 2d core used in cards many many generations ago. It's more of a legacy add-on similar to vbios or basic vga support.

In Linux generally your only able to use less then 10% of the card's capabilities for all your desktop display needs. And even then it's usually kinda piss-poor. (which is what EXA is designed to fix)

Thank goodness for work on XGL, and OpenGL-accelerated stuff like the optional Glitz backend for Cairo and efforts with QT and vector based graphics. Even Mozilla browser is working on having it's rendering engine with Cairo with Glitz support. (and as a side effect it would solve 90% of the Linux desktop printing woes)

Right now the only card left that would be considured 'modern' and have full open source DRI drivers is Intel's 'Graphics Media Accelerator' used on it's 915 and newer motherboard chipsets. It's a very capable replacement of their fantasticly inadiquate 'Extreme Blaster' style embedded chips. Used commonly in 'Sonoma' generation of the centrino platform. It's between a ATI 9200 and 9800 in terms of performance... The nicest version on the 945g motherboards sport a 400mhz cpu with shared memory over the 16x PCIe bus.

Definately nothing wonderfull, but good enough for general opengl support and fast desktop compositing.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 14:06 UTC (Wed) by sbergman27 (subscriber, #10767) [Link]

That's the article I was thinking of when I posted that I'm unconvinced.

Some of the reasons I remain unconvinced:

1. The current 2D capabilities of cards are more than enough for most any desktop.

2. With all the legacy cards out there, and a continuing market for cheap video cards or built-in video chipsets for servers, 2D support can't really go away.

3. The massive 3D section is there for doing 3D stuff, not 2D. (i.e. games)

4. Whenever I try to get a glimse of this massive power (for 2d use) hidden away in the 3D chipset, I get a disappointment. E.g. (and I'll use my NVidia GForce 6800 GT AGP 8x card as an example, since its drivers are the most feature complete available under Linux), mplayer, in benchmark mode, shows the xv output driver absolutely blowing away the gl and gl2 drivers. I've posted about this before, and have been told that the 3D rendering just needs to be "optimized". OK, so now we've gone from "the 3D section is so much more powerful than the 2D that the 2D section could never keep up" to "err, ummm, the 3D usage needs to be optimized". And this on one of the more powerful 3D cards on the market, but whose 2D capabilities are not considered to be all that hot.

5. I have not seen one single feature made possible by 3D that is really useful for post people. Mainly what I hear is that once this power is put into the desktop designers' hands, they will find really useful things to do with them. (Yeah, like menu transparency and drop shadows?) Wiggly windows aren't useful. I've seen the ability to rotate the window along the Y axis to temporarily take it "out of the way" presented as a "useful" feature, but really, isn't minimizing it, or window shading it ultimately better?

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 16:53 UTC (Wed) by sbergman27 (subscriber, #10767) [Link]

Oops.

s/glimse/glimpse/

s/post people/most people/

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 17:54 UTC (Wed) by oak (subscriber, #2786) [Link]

> 5. I have not seen one single feature made possible by 3D that is really
useful for most people.

For engineers the profit is in modeling architecture, lightning, houses,
bridges, molecules, engines etc. In the shops 3D is useful for
visualizing things to customers: what their new kitchens will look etc.

But I guess at home for "most" people 3D is useful only for gaming...

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 21:43 UTC (Wed) by man_ls (subscriber, #15091) [Link]

Apple does some nice things with OpenGL in Quartz Extreme, available since Mac OS X 1.3 (Panther I think). E.g. the login screen performs smooth rotations, and I hear that presentations with Exposé look nice too. Now, when I remember how cheesily those Powerpoints zoom and slide by... There was also a cool trick some guy did with the Powerbook's internal gyroscope, where every window was set to oscillate according to its age, activity and the laptop's orientation.

Ok, ok, nothing earth-shattering, but at least it shows somewhere! With Quartz 2D Extreme, it seems that processing is offset almost completely to the graphics card. So yes, the all-3D era is here long before Vista appears.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 21:47 UTC (Wed) by man_ls (subscriber, #15091) [Link]

Ooopsie.

If I had only read the linked article to the end, it says that Quartz Extreme was available since Mac OS X 1.2.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 23:24 UTC (Wed) by sbergman27 (subscriber, #10767) [Link]

Wow, a login screen that performs smooth rotations? Windows that oscillate according to their age, activity, and the laptop's orientation?

I was so wrong to say that 3D had little use on the desktop. Right...

Could someone please come up with a single freaking convincing example of why 3D is *useful* for a 2D desktop?

Open Graphics Project Status Update (KernelTrap)

Posted Oct 6, 2005 18:31 UTC (Thu) by man_ls (subscriber, #15091) [Link]

Could someone please come up with a single freaking convincing example of why 3D is *useful* for a 2D desktop?

I think there is none. After all, it was thought as a 2D desktop; by definition, 3D has no place here. Scaling and displacing should be much faster when done in specific 2D circuitry; other effects like rotations and the OSX-minimize thing are just eye candy.

On the other hand, in Quartz 2D Extreme you use a powerful and expensive DSP for drawing primitives and compositing images. The fact that it happens to be a 3D chip is just an accident. But is it really useful? Hardly.

Don't know much about Vista. Judging by past history it will probably be a cheap copy.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 6, 2005 19:59 UTC (Thu) by oak (subscriber, #2786) [Link]

Some vector stuff that is useful:

- Enlarging the item being focused (menu row, toolbar buttons etc)
would be a nice accessability feature for people with bad
eyesight. They could still use the desktop without having font size
where you can fit only a few lines of text on whole screen

- Mac OSX feature to see all windows at the same time
(all scaled down to fit to screen at the same time),
and selecting the one to focus (after which normal window
sizes are restored) is quite nice

- Dimming other windows than the focused one (requires just alpha
blending support and memory, not vectors)

- Desktop items, such as icons, could scale e.g. according to display
DPI or number of pixels. This will be nice with display migrations
(ability to move applications from one sized screen to another, e.g.
from laptop to 32" TV, to mobile phone etc.)

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 22:44 UTC (Wed) by sbergman27 (subscriber, #10767) [Link]

I should have been more specific in my wording, I suppose. I see little advantage to a 3D *desktop*. Obviously, 3D has been useful for a long time in specific applications.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 12:40 UTC (Wed) by job (subscriber, #670) [Link]

It's also important to note that the Radeon 9200 is the end of an era

Not true. There is the R300 DRI project, which is actively working on support for the Radeon 9500-9800 in DRI/MESA. You can't download anything from this particular project, but from dri-devel it looks like they've started to include this support in the CVS head of DRI. I found some success stories on the mailing list.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 15:49 UTC (Wed) by AJWM (subscriber, #15888) [Link]

I think by "end of an era" the original poster meant that the Radeon 9200 was the last chip for which specs were made available by the vendor. ATI stopped doing that after that model, and nVidia has never (or if so, not for years) made them available. (As in, made available not under an NDA that would prohibit open source code development.)

Drivers for anything after that have to rely on reverse engineering.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 1:07 UTC (Wed) by ncm (subscriber, #165) [Link]

Somewhat like Ogg, OG will sink or swim on whether it's taken up by self-interested embedded-system manufacturers. If they do, the rest of us benefit from the prospect of volume manufacturing. So, the best thing to do is make sure their engineers can get a working system up quickly, without per-unit cost, and then just make sure their fixes/improvements go into the common pool. At some point there will be cheap aftermarket chips to drop onto a PCIe, PCIx, PC-Card, or what-have-you board to sell as add-ons, or it will be integrated into (probably PC104) motherboards' soldered-on chipsets.

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 3:04 UTC (Wed) by drag (subscriber, #31333) [Link]

Yep. That's what the ASIC version will be targetted for.

But don't forget the FPGA version. As a video card it won't be "uber", it'll be slightly slower, run hotter, and be more expensive then the ASIC version will be.. however it does have some potential in a very different way.

It would basicly be a low-cost PCI card with on-board memory that can be used by hardware hobbyists to mess around with programming their FPGA chips. As far as I know there isn't anything like that in existance.

I am no hardware person and have limited experiance with electronics.. I don't know what sort of limitations there will exist with this card in terms of programmablity. However I know of one company that produced a network'd camera with a built-in hardware Theora encoder. It's capable of realtime HD-quality video over regular ethernet networks, and it uses a much smaller/slower chip.

http://www.xilinx.com/publications/xcellonline/xcell_53/x...

So I figure a hobby person could adopt that code that man developed and have develop a design for hardware accelerated encoder/decoder for uses in Mythtv or video editors.

Or maybe develop a way to accelerate raytracing renders like Povray or Yafray for 3d graphics? A hardware based encryption device for high speed VPNs? Hardware random number generator?

Or something like that. Who knows what people could figure out if they are able to have a PCI/PCIe card with 128megs of ram and reprogrammable proccessor aviable for cheap?

Open Graphics Project Status Update (KernelTrap)

Posted Oct 5, 2005 7:05 UTC (Wed) by jwb (subscriber, #15467) [Link]

More importantly, widespread use of the CPD version will allow the design to be well and truly shaken out before committing to the ASIC version. There have been many video card ASICs, past and present, with totally ridiculous hardware bugs. But this one should ship without major problems, owing to its relative simplicity, and the unprecendented luxury of a public beta for hardware.