Slashdot Mirror
Xi Announces Hardware Accelerated 3D X Server
Thanks to Jeremy Chatfield for poking me about Xig [?] 's recent press release regarding them being the first to market a commercial hardware-accelerated 3D X
server. They will be showing at Comdex, next week, and shipping in December. Support for cards looks nice as well, with over 30 cards supported,from companies like 3Dlabs, ATI, Evans & Sutherland, Diamond, S3 and Matrox.
For example: 3dfx cards are limited to 256x256 textures, 16bpp rendering (don't give me any of that 22bit crap, all that is is a lowpass filter on the RAMDAC's lower bits to 'compensate' for the dithering), have a really crappy memory architecture, no stencil or accumulator buffer... there's no real reason for a workstation graphics developer to support what amounts to a kludgy legacy gaming card. A TNT1 is much cheaper and much more powerful, in terms of OpenGL features if not fillrate, than 3dfx's highest-end card.
That said, there's much less of a need for 3dfx support in the X server, Mesa already supports it through Daryll Strauss's Glide port. It works about as well as can be expected. (The buffer-copying thing you mentioned has been in fxmesa for quite some time, though it's quite slow.)
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Regardless, what functionality could pros need which isn't in the entertainment edition? Don't they realize OpenGL is OpenGL, and that to restrict the hardware based on what amounts to not paying as much for drivers is somewhat asinine? If the professional edition means having better/faster emulation for what the card doesn't support in hardware, that's fine, but if it means crippling the hardware, that's outrageous. Then again, $100 is a bit outrageous for a video card driver, too. :P
I'm just going to wait for XFree 4.0. It'll be out soon enough; in the meantime, Mesa's software renderer is enough for me. (Yes, it's slow, but that just helps me optimize my code even better. :)
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
3Dfx cards can do in-window rendering, just it's not supported in any X server as yet. Hopefully XFree 4 will have a good driver that can do this, tho.
Sam: "That was needlessly cryptic."
Max: "I'd be peeing my pants if I wore any!"
Yes, they're the first to release a commercial hardware OpenGL-capable X server for Linux. nVidia released an alpha-quality free OpenGL-capable X server quite some time ago, Matrox released the specs for one... I can't help but wonder if Xi has taken the various design documents put on the web by the various parties developing the free servers and used them to try to undercut them, beating them to market and trying to make a first impression.
As I've said elsewhere in this thread, I'm waiting for XFree 4.0. I know they can deliver, and will put out something which will hopefully work much better, and will certainly be freer. I'd be willing to accept a slightly slower OpenGL performance if it meant not having to spend $100 for a video card driver. So far the various clued-in vendors seem to be supporting the XFree efforts, in the meantime.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Basically, this amounts to "only professionals need multiple GL contexts, and so you need to pay 2.5x as much to get multiple GL contexts." I find this somewhat outrageous, myself. They also make a big deal about features which are standard parts of OpenGL and which, as I already stated, are basically just parts of a graphics card driver which are only being enabled for people willing to pay more.
Also, 64 MB RAM for an X server? That's just beyond ridiculous. (That's for both the entertainment and pro versions.)
None of the feature lists say anything about GLX (network-transparent OpenGL), either. XFree can do GLX now.
Oh, another thing: they are a bit incomplete, at best, when saying what cards they support. The only nVidia chipset they seem to support is the TNT2 - they don't have any listed support for the TNT, or the Riva 128 series. They also don't have any listed supported Matrox cards. For their sake I hope they just forgot to complete their supported cards list.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
My advice: Don't buy this GL server. Wait for XFree 4. Just because they're first to market doesn't mean they're the best; personally, I can't stand the thought of supporting any company which resorts to Microsoftian tactics to try to put themselves above the rest of the Linux world.
Speaking of Microsoftian tactics, they even have separate versions of their CDE for executives and developers each on desktop and laptop computers! Four different versions "optimized" to the supposedly different needs of different users, rather than having a single product which can be tuned to the needs of the individual! Pathetic.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Why do they bother? Because they have a good product. They support more cards than XFree86 does overall (last I checked atleast).
Not everything in your happy little linux world needs to be free for it to be good. A Free X server can't support as much hardware as a commercial one, because the commercial one will pay companies to write drivers or for specs, XFree86 can't afford to do this.
Don't forget that we have an open source project that works right now. Accelerated GLX for XFree86 3.3.5 supports Matrox G200 and G400 as well as Riva TNT2 cards. Support for Rage Pro chipsets is on the way, too. So you might want to consider contributing some code to the project rather than shelling out for the new Xi server...
F0 07 C7 C8
"the first to market a commercial hardware-accelerated 3D X server." Pardon my ignorance but haven't SGI been doing that for a decade?
-- SIGFPE
people shouldn't have to tailor make their PC's to use an OS...if I bought a new gfx card, I don't want to downgrade to something less powerful just to use XFree86.
That's a nice bit of rhetoric and perhaps FUD. However, the fact remains that you are still spending a non-trivial amount of money to ensure compatibility. In most instances you can ensure that compatibility with the same cost just by replacing hardware.
This also eliminates any future need for you to deal with the extra complexity of dealing with XiG again for any future version of your Linux distribution.
Furthermore, there is little likelihood that you will have to downgrade to something less powerful just to use Xfree. The vendors that are supporting Linux now are the top tier. If anything, you would be bending over backwards with XiG just to use something inferior like an ATI product.
AccelX doesn't even support the TNT correctly... the *FASTEST* refresh rate I can get with my TNT and my Optiquest V95 (19") is not what I asked for in the config. I asked for 1600x1200 @ 75, witch my hardware is capable of with XFree86. Now, I get 1600x1200 @ 66.7 even if my /etc/Xaccel.ini look like this...
[SCREEN]
Board = "nvidia/tnt-16.xqa";
Monitor = "viewsonic/v95.vda";
Depth = 24;
EnergyStar = YES;
SoftwareCursor = YES;
Desktop = 1600x1200;
[RESOLUTIONS]
1600x1200@75;
Thats pretty bad for a 100$ piece of software!
The reply I got from their techsupport was : Yes.... we know, your'e not the only one to complain. We will check that out for future releases.
WOW what a serious compagnie!
So guess what... XFree is what I use! =))
Unix is user friendly... it just chooses it's friends selectively!!
"something inferior like an ATI product."
Funny, really it is. Actually, everyone I know who has an ATI board is more than happy with it. Never has problems, etc. And I'm not talking about costs, yes the X server costs more...I have no reason nor need to buy AccelX, but the fact remains, that it's somethign you shoudln't have to do.
Commercial software gives people someone to rightfully bitch to if it doesn't work.
As a rather unsatisfied Xig customer, let me address this comment. Yes, you can "rightfully bitch" to Xig if you aren't happy with their product. I hope it makes you happy to do so, because it won't do much else--the only satisfaction I have ever gotten from Xig about some serious failures of their servers is "Hmm. We may have a fix for that coming up. Watch our ftp site for the next patch, which should be out sometime in the future."
And that's when they're being pleasant. The normal tone of their tech support is rather more caustic ... I'd have to rate it as "grudging". Like they were doing me a favor by fixing their own mistakes. Jeremy Chatfield especially seems to have a chip on his shoulder.
I am now using AccelX 4.1 with my Matrox Millenium II. It has all the latest patches, and it still biffs one or two characters on every page. It's a transient failure--if I repaint the page, different letters are blotted out. Probably why they haven't fixed it yet. So why don't I use XFree? Because the latest version I've tried (3.3.1--old now, admittedly) had even more serious problems. I'd drop AccelX in a heartbeat if those were fixed. Hmm ... maybe I should download the latest and try again!
I've been using AccelX on and off for years now, since before Caldera bundled it with their first product. And here's my analysis of the company, to add a data point to help folks make an informed decision.
Pros:
Cons:
On the other hand, I've never actually gotten a MetroX server to even work, so I cannot even compare their product to the others.
I hope this helps folks in their quest for knowledge.
Let me tell you my experience with 4front's OSS. It was decent, and nearly worth the $30 I paid for it ($20 base license, $10 additional for AWE64 functionality). Then I upgraded my computer (from a K6-233 to a Celeron 300A), and then the troubles began. First off, I changed my kernel's scheduler to operate at 1000Hz instead of the default 100Hz, and OSS was horribly confused by this. So I tried explaining this to the 4front guys. Their response was that I was obviously an evil overclocker, that soundcards are proven not to work on overclocked PCs, and that their product was only for newbies. This angered me, and I responded my views, stating that I wasn't an overclocker, but even if I was, a Celeron 300A overclocked to 450 looks to the bus just like a normally-clocked P2-450, and that it would make no difference. This was all the 'proof' that the 4front representative needed to claim that I was a dirty overclocker. It took many messages to explain that I wasn't overclocking, and the problem was with the kernel scheduler, and that I had bought OSS for the purpose of having a fully-functioning AWE64, and if there wasn't any way for OSS to deal with a simple change in kernel scheduling latency, then there is a problem with OSS, and not with me, whether I was a "dirty overclocker" or not.
Eventually, I got my money back (I didn't even care about it, but they sent me a refund anyway even though it was over a year after I'd purchased the license), but it was still frustrating to have assumptions made about me and to be lambasted for something which was irrelevant to the discussion.
To sum up, they (4front) evidently only care about their largest market segment, because that's where their profit is, and don't care about any sort of improved quality of Linux multimedia, which is their business. I think 4front can go collectively screw an active toaster for all I care.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Actually, his exact words (I was at that talk too) were, "Currently, it's sometime around December the 48th."
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
I've noticed a huge influx of ACs lately who don't seem to understand the concept of a default score. I've gotten into quite a few flamewars recently because people were wondering why my crap posts are "moderated up" when really they start at 2, due to my high karma.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Using OpenGL to do rendering is more common than you'd think. Also, there's more to rendering than doing the final render - it's nice to have a preview of the image too, and even if it's at lower quality, it's better to wait 10 seconds for a high-quality OpenGL render than 10 minutes for a low-quality raytrace.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
For the most part, it's automatic. Quake3 will be pretty much automatic, for example, since it's linked against OpenGL correctly (i.e. dynamically against libGL). Quake2 is kinda borked in how it's linked to OpenGL, and so can only work with fxmesa unless you do a lot of serious tweaking. I'm not sure how Quake1 is. Since most Linux programs are distributed in source form, on average all you have to do is recompile, and at worst you'll just have to fire off an email to the maintainers of the program to properly link their binary dynamically against libGL (and ones improperly dynamically linked to libMesaGL, which has been deprecated, can be coerced into working with libGL anyway by just symlinking libMesaGL to libGL).
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
My opinion (and this is, of course, just opinion) is somewhat the opposite... I find KDE/Qt to be much better, in many respects, than CDE/Motif. Qt is much more streamlined, efficient, and modern, and KDE is, as a result much faster. KDE is also much more flexible and configurable, and easier to use (IMO). It also looks much nicer, and if you don't like the look of it, you can change it easily.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
Of course Mesa's slower. With the exception of fxmesa, it's a software-only renderer right now. DRI and GLX are separate things which work with Mesa, but don't replace it and aren't part of it. Comparing Mesa to the OpenGL ICD is like comparing a Honda Civic with a racecar.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?