S3 Graphics Releases Linux Driver With OpenGL 3.0, VA-API

[phoronix]

Phoronix: S3 Graphics Releases Linux Driver With OpenGL 3.0, VA-API

For months we have seen S3 Graphics advertise a magical Linux driver in their press releases that promised to offer OpenGL 3.0 support and advanced video functionality. They had reported to us the driver would be released in December, but that deadline had passed and they continued to announce Linux support when launching the Chrome 540 GTX, but still there was nothing. However, S3 Graphics has now actually delivered such a driver! They have delivered a Chrome 500 series Linux driver that not only provides OpenGL 3.0 support but also H.264, VC-1, and MPEG-2 hardware decoding on the GPU. While it may appear to be good, this driver is still far from perfect.

http://www.phoronix.com/vr.php?view=13546

[cruiseoveride]

damn, even S3 is pwn'ing ATi

[Kano]

I guess nobody really tested that card with Linux yet. The drivers are 32 bit only and contain a binary blob for the kernel module. So they have to change the licence for it to S3 or whatever - at least not GPL.

[RobbieAB]

All the same...

Where is the flying pig picture?

[nico342]

What's happen ATI ? :confused:

[DanL]

damn, even S3 is pwn'ing ATi
fglrx isn't the greatest, but if you really believe S3's drivers are better, then I've got some ocean-front property in Bolivia I'd like to sell you.

Or maybe you're just a frustrated ATI customer that wanted to take a shot at ATI...

[TechMage89]

I seriously think that any claims by s3 at this point should be taken with a grain of salt. The fact that their decidedly closed-source driver modules report themselves as GPL ought to tell you something.

If there's someone with a Chrome 500 who can test this driver, *then* maybe we can find out how well it actually works and what its actual capabilities are.

[conholster]

Who makes gfx cards based S3 chips? I wanna try.

[Thetargos]

I hope that soon enough, you'll be able to test the driver, and the hardware on Linux, Michael. I'd really be interested in seeing how well it performs, how does it compare to the earlier tests you ran on cheap hardware and nVidia GFX with VDPAU and how well does it do 3D compared to Intel, nVidia and ATi (on reasonably similar hardware... Do we even know how many stream processors the 500 series pack or if it has a vertex engine? How about memory interface, etc?)

[stan]

S3 is already violating the GPL by distributing the binary without the source. They obviously need to claim it's GPL in order to use Linux's GPL'ed symbols. So now it's up to someone who owns copyright to the Linux kernel to sue their behind.

[damentz]

I hope that soon enough, you'll be able to test the driver, and the hardware on Linux, Michael. I'd really be interested in seeing how well it performs, how does it compare to the earlier tests you ran on cheap hardware and nVidia GFX with VDPAU and how well does it do 3D compared to Intel, nVidia and ATi (on reasonably similar hardware... Do we even know how many stream processors the 500 series pack or if it has a vertex engine? How about memory interface, etc?)

Yes, this would be interesting. S3 has some interesting technologies, especially the ability to use anisotropic filtering without any framerate drop.

Though, it would be cool for them to create some real drivers for linux with up to speed Xorg and kernel support if they decide to continue the proprietary method _completely_. If not, try what AMD is doing and hire someone to work on the open source driver too.

[colo]

Non-free drivers are worthless. I would have bought a Chrome if they'd ship source under an apropriate license, but this way, I'm just not interested. At all.

[hdas]

Lol, to be honest, I am shell-shocked. The fact that they came up with something apparently even remotely working is beyond my wildest imaginations, let alone issues like license, new hardware, amd64 arch, features and quality.

All this time I took it as hoax/vaporware and thought Phoronix was having some cheap fun going after clueless chicken in S3's PR dept.

In any case, since it will head towards nowhere, I just have to say, this is what it called *epic fail*.

(Just a reminder - this is *2009* and if you can't even create a 3d driver with compositing, you shouldn't use the term linux support. Vesa can do better :D.)

[Thunderbird]

I haven't seen the drivers but judging by the GL string whether it supports 3.0 is wrong. Officially 3.0 should be reported when a opengl 3.0 capable glx context is created using GLX_ARB_create_context. That's the official way. Unfortunately Nvidia broke with that rule in a recent 180.x release likely for marketing reasons. If S3 supports GLX_ARB_create_context then they have OpenGL 3.0 support.

I just checked their libGL.so and they do have GLX_ARB_create_context / glXCreateContextAttribsARB, so they really support OpenGL 3.0. Please update the article.

[Extreme Coder]

Quite surprised by this.

I'd like to see an article testing this, comparing performance with similar price-ranged cards from ATi and nVIDIA.

[tball]

I haven't seen the drivers but judging by the GL string whether it supports 3.0 is wrong. Officially 3.0 should be reported when a opengl 3.0 capable glx context is created using GLX_ARB_create_context. That's the official way. Unfortunately Nvidia broke with that rule in a recent 180.x release likely for marketing reasons. If S3 supports GLX_ARB_create_context then they have OpenGL 3.0 support.

I just checked their libGL.so and they do have GLX_ARB_create_context / glXCreateContextAttribsARB, so they really support OpenGL 3.0. Please update the article.

@Michael, Thunderbird is right.

My fglrx 9.2 doesn't report opengl 3.0 either with glx:
OpenGL version string: 2.1.8494 Release

But opengl 3.0 is certanly there.

[curaga]

Wikipedia has the specs:

Chrome 540 GTX 65nm 800Mhz core
32 unified shaders 1700Mhz ram (850 gDDR3) 64bit bus

[[Knuckles]]

Not that I would buy a S3 graphics card, but hey, at least they have something now.

I just hope they recognize the value that is added to their hardware if the driver is opened up.

[d2kx]

They are not producting high-end 3D cards, why oh why don't they go for free drivers?

[Thunderbird]

@Michael, Thunderbird is right.

My fglrx 9.2 doesn't report opengl 3.0 either with glx:
OpenGL version string: 2.1.8494 Release

But opengl 3.0 is certanly there.

The only reason nvidia changed the strings (which is actually very bad as when you don't use a opengl3 context you aren't using the real opengl3) is for marketing reasons. A new glxinfo would be needed.

[sreyan]

What's happen ATI ? :confused:

ATI seems to be making progress. I checked out radeon hd code today from a git repo and it seemed to have some 3d support for my radeon 2600. If this extends to the r700 series too, I will absolutely buy an ati card in the price range of the chrome 540.. Hopefully one of those new rv740 based cards.

[sreyan]

for those that do want to buy:
https://s3gstore.s3graphics.com/

[kraftman]

The only reason nvidia changed the strings (which is actually very bad as when you don't use a opengl3 context you aren't using the real opengl3) is for marketing reasons. A new glxinfo would be needed.

So nvidia lied, that's it.

[deanjo]

So nvidia lied, that's it.

More like they are waiting for the foss to get their act together and update their supporting code to todays standards.

[Duo Maxwell]

ATI seems to be making progress. I checked out radeon hd code today from a git repo and it seemed to have some 3d support for my radeon 2600. If this extends to the r700 series too, I will absolutely buy an ati card in the price range of the chrome 540.. Hopefully one of those new rv740 based cards.

I just checked, within the price range of the 540 which will more then likely be seriously gimped out with only 64 bit memory, even though it comes with GDDR3 you can get either an AMD HD3850 or HD4670 or the Nvidia 9600GSO. (http://www.newegg.com/Product/Productcompare.aspx?Submit=ENE&N=2010380048%204026&bop=And&CompareItemList=N82E16814140103%2CN82E16814102794% 2CN82E16814140095%2CN82E16814130437%2CN82E16814131 096%2CN82E16814131146)

All 3 of which retail for less, come with 128 or 256 bit, GDDR3 or GDDR4 memory and all 3 will kick the crap out of the 540 up and down the block. I'm not actually sure which one is the most powerful of the 3.

But I'd wager that the 3 are fairly close in raw performance, since the HD3 series wasn't that powerful, the HD3870 usually coming in around the performance of the 8800GT and 9600GT, so the 9600GO and HD3850 should be fairly close, the HD4670 looks pretty close as well and should have a notably lower power draw then the HD3850.

You can check the GPU chats http://www.guru3d.com/category/vga_charts/ sucks that they haven't updated the charts on ET:QW though due to it's quick fall off in popularity since it's pretty much the most demanding game on Linux...


Oh and for anyone interested, here's a preview review of the new AMD HD4750/rv740 gpu http://www.guru3d.com/article/radeon-hd-4750-rv740-review-preview-test/ It looks to fall between the HD4830 and the HD4850, so if you have either its prolly not worth it to you.

I've long believed that Nvidia and VIA/S3 should merge, so what if we lose them dropping docs, it's not like anyone will actually knowingly buy and S3 hardware these days, not to mention the fact that you can't find any VIa hardware in normal shops, I've only ever sen it for sale on shops that specialize in sff and *-ITX sized mobos in the last 3+ years. They don't even seem to make AMD or Intel chipsets anymore.

Why should they merge with nVidia? Simple we could use a 3rd real competitor in the cpu and mobo market that offers a complete system, AMD already has their complete low to high end cpu+chipset+gpu platform, Intel is going to complete their final piece of the puzzle with the release of Larabee, this leaves nVidia out in the cold since they licensed SLi to Intel for the X58 chipset, but Intel wont let them make chipsets for the i7 series CPUs.

nVidia also can't just go out and make their own x86 CPU, due to Intel controlling who gets license to the designs, only having signed out those licenses due o IBM and government contract requirements of having a 2nd supplier for hardware. After years of failed companies in the tech market I think the only ones with x86 licenses are AMD and VIA.

VIA merging with nVidia would allow them to enter the cpu market and allow them to offer at least a high end chipset and gpu market to start with while they apply nVidia's knowhow to the VIA Nano to get a decent performer out of it, its already a pretty good cpu at the low end, but with a little help form nVidia's team theres no reason that they couldn't have a competitive mid to high end platform within 5 years.

[bnolsen]

VIA merging with nVidia would allow them to enter the cpu market and allow them to offer at least a high end chipset and gpu market to start with while they apply nVidia's knowhow to the VIA Nano to get a decent performer out of it, its already a pretty good cpu at the low end, but with a little help form nVidia's team theres no reason that they couldn't have a competitive mid to high end platform within 5 years.

That should have happened about the time AMD acquired ATI. 5 years is an eternity in the cpu/gpu market, it's probably too late at this time.

[Thetargos]

nVidia could also flex their immense performance muscle and move away from x86 to produce really high performance CPUs if they wanted, there are other licensable CPU specs (MIPS/POWER) which can be made much more efficient and powerful than the aging x86. Problem will be how to market it (and get [commercial] support for such platform[1]), not to mention that even though nVidia's big, Intel is MUCH bigger (and not even they were able to shift consumer PCs architecture from x86).

Back to VIA's S3 500 series, I welcome their efforts for competing in this rather tough market, however, only 64-bit bus width is a bit too low (nowadays low-to-mid range cards use 128-bit bus width. I really hope they'll produce improved versions and price them more in tune to AMD/nVidia, however they can price their products much lower due to the volume they sell, so VIA would have to make a heavy investment in this regard to be able to price these products competitively.

[list=1]
We know Linux is most likely to support it, provided there is enough information released.

[Duo Maxwell]

Thats the problem though, due o the windows monoculture all other cpu archs have gone under, even the last holdout in the consumer market, Apple with the IBM Power4 derived G5 gave up and went to x86. The only place you're seeing other archs these days in the consumer market is in stand alone devices like game consoles like the Cell in the PS3.

Nviida already has the ARM based Tegra platform http://en.wikipedia.org/wiki/Nvidia_Tegra but you won be seeing that powring anything more powerful then your next cellphone.

As far as my stated 5 year mark, yes, that is an eternity in tech years, but with curent trends the CPU is becomming less and less important, while the GPU is becoming more and more essential. I do't see why the couldn't though have a version of the Nano die shrunk between 55-45nm and at least dual core running 2.5Ghz within a year, which would at least get them in the door at the low end with the Pentium dual core a few of the low end C2Ds and the AMD X2 line. The would be able to hold onto their gpu position for a while as well to remain profitable while they claw their way up. The real challenge though is if they can get it into the OEMs, nvidia has had little trouble getting their GPUs and chipsets into the OEMs for the last few years, I've seen more of them recently there then I've seen Intel GPUs and AMD gpus when poking around the laptops at the local worst buy.

But seriously you don' think t'd take close to 5 years for them to get a cpu that would be within a few flops oft the best that AMD and Intel have to offer at the time? By then we'll be up to 12-16 core CPUs pushing something close to 4.5Ghz a core. It's not something anyone could build over night.

Maybe someday soon we'll get multicore multicore cpus that will have several different archs running oon the same die ith some level of interpreter that decides what arch to execute what code fore some truly amazing cpu power. But alas, I think the difficulty of making such a cpu will leave it an infeasible pipedream...

No matter what it's an uphll battle for both nVidia and VIA since its now much harder to pull out an antitrust on Intel for monopolistic practices since AMD bought ATI. The only thing we can hope is that AMD has something up their sleeve to retake the cpu market and maybe a new chipset that can give us at least 4x16xPCIe slots, or even better, a 7x16xPCIe chipset with 2x CPU sockets, anything to get them more market space ahead of Intel. Just think if you could get a 14 way crossfire mobo with 2x 4-4.5Ghz OCd quads and 12Gb of tri channel DDR3. Sure theres no real reason for it, but it would get AMD more much neded advertising all over the web on various tch forums and it'd allow them to just brute force games like Crysis Warhead into actually running lol.

[kraftman]

More like they are waiting for the foss to get their act together and update their supporting code to todays standards.

So they lied. Nvidia drivers replace much of X, so I don't think they were/are waiting for FOSS.

[Redeeman]

They are not producting high-end 3D cards, why oh why don't they go for free drivers?
lol.. as if highend has anything to do with it..

[deanjo]

So they lied. Nvidia drivers replace much of X, so I don't think they were/are waiting for FOSS.

No because they do support OGL3.

[deanjo]

Thats the problem though, due o the windows monoculture all other cpu archs have gone under, even the last holdout in the consumer market, Apple with the IBM Power4 derived G5 gave up and went to x86. The only place you're seeing other archs these days in the consumer market is in stand alone devices like game consoles like the Cell in the PS3.

Nviida already has the ARM based Tegra platform http://en.wikipedia.org/wiki/Nvidia_Tegra but you won be seeing that powring anything more powerful then your next cellphone.

As far as my stated 5 year mark, yes, that is an eternity in tech years, but with curent trends the CPU is becomming less and less important, while the GPU is becoming more and more essential. I do't see why the couldn't though have a version of the Nano die shrunk between 55-45nm and at least dual core running 2.5Ghz within a year, which would at least get them in the door at the low end with the Pentium dual core a few of the low end C2Ds and the AMD X2 line. The would be able to hold onto their gpu position for a while as well to remain profitable while they claw their way up. The real challenge though is if they can get it into the OEMs, nvidia has had little trouble getting their GPUs and chipsets into the OEMs for the last few years, I've seen more of them recently there then I've seen Intel GPUs and AMD gpus when poking around the laptops at the local worst buy.

But seriously you don' think t'd take close to 5 years for them to get a cpu that would be within a few flops oft the best that AMD and Intel have to offer at the time? By then we'll be up to 12-16 core CPUs pushing something close to 4.5Ghz a core. It's not something anyone could build over night.

Maybe someday soon we'll get multicore multicore cpus that will have several different archs running oon the same die ith some level of interpreter that decides what arch to execute what code fore some truly amazing cpu power. But alas, I think the difficulty of making such a cpu will leave it an infeasible pipedream...

No matter what it's an uphll battle for both nVidia and VIA since its now much harder to pull out an antitrust on Intel for monopolistic practices since AMD bought ATI. The only thing we can hope is that AMD has something up their sleeve to retake the cpu market and maybe a new chipset that can give us at least 4x16xPCIe slots, or even better, a 7x16xPCIe chipset with 2x CPU sockets, anything to get them more market space ahead of Intel. Just think if you could get a 14 way crossfire mobo with 2x 4-4.5Ghz OCd quads and 12Gb of tri channel DDR3. Sure theres no real reason for it, but it would get AMD more much neded advertising all over the web on various tch forums and it'd allow them to just brute force games like Crysis Warhead into actually running lol.

If via has a history of anything it's making crappy chipsets and buying once competitive companies for their tech and then totally fscking them up. *cough*S3*cough* *cough*IC Ensemble*Cough*, *cough* Cyrix *cough*

[kraftman]

No because they do support OGL3.

As someone mentioned before they don't fully support OGL3.

[Zhick]

lol.. as if highend has anything to do with it..
Actually it quite clearly has. Their GPUs are obviously not competing with AMD/nVidia performance-wise. So they don't have to be afraid to lose some kind of super-duper-special-secret-technology which gives them maybe 2-3 FPS in current games by opening up their driver.
What they are competing in (or would like to be competing in) is features, like video-acceleration, 2d and OGL3. Things that would actualy be easier to implement by integrating into the X Framework and opening up the driver (not to mention x86_64 and current Xorg and Kernel support).
Still there is the issue of DRM which seems to make it impossible to simply open up the closed-drivers (same for AMD/ATi).

[Duo Maxwell]

Actually they do have allot to lose, they still have technologies that they are licensing to their competitors, as well as have several technologies that they are licensing themselves. It would cost them allot in legal-fu to do what AMD are doing to have devs and lawyers go over all of their driver code to release it. Not everyone does things with the intention of ever releasing the docs, let alone have most of their code easily readable by anyone other then the devs that wrote said code. Remember this is VIA/S3 they aren't very profitable as it is...

It'd be interesting to see if bridgeman could get us some number on how much it's actually cost AMD to release all of the docs they've done so far., including dev time and pay, and lawyer time and pay.

As for VIA's history of failing at technology, who says vIA stays the one in control? Nvidia already had the GPU and chipset down already, all they need VIA for is VIA's x86 license, call the new company nVIAdia if you like, they could easily then work on what they already do best while trying to hammer out a CPU to go with their chipset and GPU. You also forget that the massive amount of patents that nVidia and VIA have accumulated over the years, they could, possibly with a little help from AMD, start to lean on Intel till they can get their CPU tech on it's feet.

Remember, the only real competitions thats going to matter for all of them is Intel, they're the 800Lb gorilla in the room that is within a stones throw of being the only game in town. When your only choice for new hardware is one company you can bet that the prices are going to be far and away more for what you're getting then if there was some competition. Theres no way that the Chinese knockoff cpu would be allowed in the US or most EU nations, I forget the name of it, it was a few years ago, but I do remember that someone got one and benched it and it was slower then a P2 or something like that...

[Redeeman]

Actually it quite clearly has. Their GPUs are obviously not competing with AMD/nVidia performance-wise. So they don't have to be afraid to lose some kind of super-duper-special-secret-technology which gives them maybe 2-3 FPS in current games by opening up their driver.
What they are competing in (or would like to be competing in) is features, like video-acceleration, 2d and OGL3. Things that would actualy be easier to implement by integrating into the X Framework and opening up the driver (not to mention x86_64 and current Xorg and Kernel support).
Still there is the issue of DRM which seems to make it impossible to simply open up the closed-drivers (same for AMD/ATi).
those excuses are null and void.. the only reason not to make open drivers is if you either violate others patents, or have inferior stuff.. The simple fact is, they dont want to do opensource.

[Veerappan]

those excuses are null and void.. the only reason not to make open drivers is if you either violate others patents, or have inferior stuff.. The simple fact is, they dont want to do opensource.

Not quite. If you happen to have licensed IP from another company, opening up your drivers with the licensed code intact will probably end up violating the patent licensing agreement you have in place, making you the target of a shiny new lawsuit.

I'm not saying that S3 couldn't be doing an open source driver, or that they shouldn't, but the amount of work necessary for them to create a closed driver (and not have to remove all the proprietary/licensed IP) was probably much less than creating a new open source driver from scratch with no licensed/proprietary IP.

Of course if they're claiming that their driver is licensed under the GPL, I'd hope to see source code real soon now, or someone else will probably be filing suit against them.

[Chrome-Center]

I've just tested the driver with Chrome 430 GT: It's faster than Windows with Nexuiz 2.4.2!!

http://www.chrome-center.net/index.php/en/articles/211-driver-comparison-linux-vs-vista-chrome-400500-series