Hardware Based XRender Slower than Software Rendering?

Neon Spiral Injector writes "Rasterman of Enlightenment fame has finally updated the news page of his personal site. It seems that the behind the scenes work for E is coming along. He is investigating rendering backends for Evas. The default backend is a software renderer written by Raster. Trying to gain a little more speed he ported it to the XRender extension, only to find that it became 20-50 times slower on his NVidia card. He has placed some sample code on this same news page for people to try, and see if this is also experienced on other setups."

Comments

2D acceleration using OpenGL?

[gloth]

He didn't really get too far into that, but it would be interesting to see how feasible it is to do all the 2D rendering using OpenGL, encapsulated by some layer, like his Evas.

Has anyone done that? Any interesting results? One would think that there's a lot of potential here...

Re:2D acceleration using OpenGL?

[Animats]

That's technically viable, and I've worked with some widget toolkits for Windows that render everything through OpenGL. On modern graphics hardware, this has good performance. After all, the hardware can draw complex scenes at the full refresh rate; drawing some flat bitmaps through the 3D engine isn't too tough.

One problem is that multi-window OpenGL doesn't work that well. Game-oriented graphics boards don't have good support for per-window unsynchronized buffer swapping, so you tend to get one window redraw per frame time under Windows. (How well does Linux do with this?) Try running a few OpenGL apps that don't stress the graphics hardware at the same time. Do they slow down?

One of the neater ways to do graphics is to use Flash for 2D and OpenGL for 3D. Quite a number of games work that way internally. The Flash rendering engine typically isn't Macromedia's, but Macromedia authoring tools are used. This gives the user interface designers great power without having to program.

Re:2D acceleration using OpenGL?

[Rabid Penguin]

Yes, and yes. :-)

The current version of Evas is actually the second iteration. The first version had a backend written for OpenGL, which performed quite well for large drawing areas, but was sluggish with many small areas (bad for window managers). The software engine easily outperformed in those cases, and will be used for the resulting window manager's border drawing.

For now, there is not an OpenGL engine in Evas, because of time constraints. E has a relatively small active development team atm, so it's difficult to say when someone will get around to adding the OpenGL engine. There should be one eventually, all nicely encapsulated except for a couple setup functions.

Re:2D acceleration using OpenGL?

[mvdwege]

On using OpenGL in multiple windows....

How well does Linux do with this?) Try running a few OpenGL apps that don't stress the graphics hardware at the same time. Do they slow down?

While my graphics hardware is not quite representative (the Matrox G450 is not known for great 3D performance), I ran two instances of glxgears.

Short conclusion: MesaGL on Linux has the same problem. Long conclusion: the windows showed noticable slowdowns, up to the point where animation was suspended in one window while the other ran, with the system switching the running window at seemingly random intervals.

System specs:

• Athlon 1600XP
• MSI K7TPro2 Motherboard
• Matrox G450 AGP Graphics Card
• Linux kernel 2.6.0-test3
• XFree86 4.2.1 (Debian patchlevel 9)

Hope this helps,

Mart

Re:2D acceleration using OpenGL?

[Rufus211]

Must be your hardware. I have an Ath 2700 XP with a ATI 9800 running Debian with X 4.3

Single glxgears: 3600
3 glxgears: 1200
5 glxgears: 700

(All aprox numbers). So basically it scales almost perfectly with the number of open windows.

Re:2D acceleration using OpenGL?

[Animats]

So basically it scales almost perfectly with the number of open windows.

Which means it's broken. All the windows should run at full speed until the graphics pipeline saturates.

There are several problems. First, make sure that you're not running with "wait for VBLANK" off. There's a stupid overclocker mania for running the graphics system faster than the display can refresh. This leads to high, meaningless frame rates, and to lower system performance because the useless redraws are using up all the CPU time.

Once you're past that, the issues are more fundamental.

The real problem is that OpenGL is double-buffered, but most windowing systems don't understand double-buffering or frame-synchronous drawing very well. Even OpenGL has no notion of time. But this could be fixed.

Usually, each app draws into the back buffer, then makes the OpenGL call to swap the buffers. This blocks the app (older NVidia drivers for Windows spin-locked, but I got them to fix that), but worse, it typically locks up the OpenGL subsystem until the frame ends and the buffer swap occurs. Implementations like that can only draw one window per frame time, obviously.

What ought to happen is that a request for a buffer swap should schedule a buffer swap for the next frame cycle, block the app, then let other apps get in their draw time. At the end of the frame, when everybody is done drawing, all windows get buffer swapped, and all the apps stuck in the OpenGL buffer swap call then unblock simultaneously. That way, multiple OpenGL apps running in different windows all run at full frame rate, until the scene complexity hits the limits of the graphics hardware.

Part of the problem is that X and OpenGL are such drastically different architectures that making them play well together is tough. X assumes a network-centric model; OpenGL assumes you're local. X expects a weak terminal; OpenGL needs good graphics acceleration. X is built around a windowing concept; OpenGL doesn't know about windows. X and OpenGL are defined by different organizations.

Microsoft is pulling this together in the Windows world, but it's all done with Microsoft APIs, and, recently, undocumented hardware that favors those APIs.

The damndest thing.

[Raven42rac]

I have used both ATI and NVIDIA,(and 3dfx, and matrox, but staying relevant). Generally the NVIDIA cards I have owned have been vastly outperformed by the ATI cards right off the bat, without tweakage. (This is under Linux, mind you) Even with tweakage, in my experience, you rarely get the full potential from your card.

One word:

[i_am_nitrogen]

Irix.

IrisGL or OpenGL (I think OpenGL is based on IrisGL, so Irix probably now uses OpenGL) is used extensively in Irix, for both 2D and 3D.

Re:One word:

[Talez]

Also, Microsoft are getting in on the act. The new Desktop Composition Engine for Longhorn is based on the same type of compositing but using DirectX instead of OpenGL.

It's great for using 3D effects on 2D windows for what has normally been wasted horsepower. Finally, eye candy that won't slow down your system!

Not enough details

[bobtodd]

Raster doesn't say whther he had 'Option "RenderAccel" "True"' enabled, which you must do on Nvidia cards if you want XRender acceleration.

Here is the entry from the driver README:

Option "RenderAccel" "boolean" Enable or disable hardware acceleration of the RENDER extension. THIS OPTION IS EXPERIMENTAL. ENABLE IT AT YOUR OWN RISK. There is no correctness test suite for the RENDER extension so NVIDIA can not verify that RENDER acceleration works correctly. Default: hardware acceleration of the RENDER extension is disabled.

Following that option, this one is noted:

Option "NoRenderExtension" "boolean" Disable the RENDER extension. Other than recompiling the X-server, XFree86 doesn't seem to have another way of disabling this. Fortunatly, we can control this from the driver so we export this option. This is useful in depth 8 where RENDER would normally steal most of the default colormap. Default: RENDER is offered when possible.

An important truth about X

[frovingslosh]

It may be big and bloated, but at least it's slow.

Graphics cards and computation

[Amit J. Patel]

There has been some work on using graphics cards for computation. The tough part is figuring out how to rephrase your algorithm in terms of what the GPU can handle. You'd expect matrix math to work out but people have tried to implement more interesting algorithms too. :-)

- Amit

Re: Graphics cards and computation

[Black Parrot]

> There has been some work on using graphics cards for computation. The tough part is figuring out how to rephrase your algorithm in terms of what the GPU can handle.

Isn't there a lot of sloth involved in reading your results back as well?

Meanwhile, users of GCC can exploit whatever multimedia SIMD instructions their processor supports by telling the processor you want to use them. For x86 see this and this; for other architectures start here. (Notice the GCC version in the URL; the supported options sometimes change between versions, so you should look in a version of the GCC Manual that matches what you're actually using.)

I confess I haven't benchmarked these options, but in theory they should boost the performance of some kinds of number-crunching algorithms.

BTW, Linuxers can find what multimedia extensions their CPU supports with cat /proc/cpuinfo, even from a user account. Look for multimedia support in the list at the end of the cpuinfo. Lots of those extensions only support integers or low-resolution fp numbers, but IIRC SSE2 should be good for high-precision FP operations. Use google to find out what your extensions are good for.

And post us back if you do some benchmarking, or find some good ones on the Web.

Raster's on holiday

[Rabid Penguin]

Normally, he would answer some questions or comments posted about something he has written, but he will be out of town for at least a few days.

I highly doubt he meant for this to get wide-spread exposure beyond developers of Enlightenment or X. Since it has, this is a good opportunity. I'll make this clear for anyone that didn't catch it, raster WANTS XRENDER TO BE FASTER! If there is a way to alter configuration or to recode the benchmark to do so, he wants to know about it.

Rather than posting questions about his configuration (which he can't answer right now), grab the benchmarks that he put up and get better results.

Now back to your regularly scheduled trolling...

Lessons from the ancient

[Empiric]

There's an example from back in the 80's that still probably serves as a good engineering reference for people working on hardware/software driver issues.

In those days of yore (only in the computer industry can one refer to something 20 years ago as "yore"...) there was the Commodore 64. It retains it's place as a pioneering home computer in that it offered very good (for the time) graphics and sound capability, and an amazing 64K of RAM, in an inexpensive unit. But then came its bastard son...

The 1541 floppy disk drive. It became the storage option for a home user once they became infuriated enough with the capabilites of cassette-tape backup to pony up for storage on a real medium. Unfortunately, the 1541 was slow. Unbelievably slow. Slow enough to think, just maybe, there were little dwarven people in your serial interface cable running your bits back and forth by hand.

Now, a very unique attribute of the 1541 drive was that it had its own 6502 processor and firmware. Plausibly, having in effect a "disk-drive-coprocessor" would accelerate your data transfer. It did not. Not remotely. Running through a disassembly of the 6502 firmware revealed endless, meandering code to provide what would appear, on the surface, to be a pretty straightforward piece of functionality: send data bits over the data pin and handshake it over the handshake signal pin.

As the market forces of installed base and demand for faster speed imposed themselves, solutions to the 1541 speed problem were found by third party companies. Software was released which performed such functions as loading from disk and backing up floppies as speeds that were many, many times faster than the 1541's base hardware and firmware could offer.

The top of this particular speed-enhancement heap was a nice strategy involving utilizing both the Commodore 64's and the 1541's processors, and the serial connection, optimally. Literally optimally. Assembly routines were written to run on the both 64 and the 1541 side to exactly synchronize the sending and receiving of bits on a clock-cycle by clock-cycle basis. Taking advantage of the fact both 6502's were running at 1 Mhz, the 1541's code would start blasting the data across the serial line to the corresponding 64 code, which would pull it off the serial bus within a 3-clock-cycle window (you could not write the two routines to be any more in sync than a couple 6502 instructions). This method used no handshaking whatsoever for large blocks of data being sent from the drive to the computer, and so, in an added speed coup, the handshaking line was also used for data, doubling the effective speed.

The 1541 still seems pertinent as an example of a computer function that one would probably think would best be done primarily on a software level (running on the Commodore 64), but was engineered instead to utilize a more-hardware approach (on the 1541), only to be rescued by better software to utilize the hardware (on both).

There's probably still a few design lessons from the "ancient" 1541, for both the hardware and the software guys.

Re:Lessons from the ancient

[The Vulture]

The 1541 drive itself was actually quite fast, reading an entire sector in much less than a second (if you set the job code directly in the drive). It was the serial transfer that was super slow (as you stated).

Unfortunately, the fast loaders assuming that the CPU and the drive both ran at exactly the same speed was a cause for problems. The PAL version of the C64 ran at a different speed (a bit slower, I believe), thus making fast loaders either NTSC or PAL specific (although there may have been one or two that could actually take the clock speed into consideration). The same fault meant that fast loaders sometimes didn't work with some variants of the drives (different CPU's, all supposedly 6502 compatible, but not necessarily so).

Additionally, because these fast loaders required exact timing, something had to be done with the VIC-II (interrupts from it would cause the 6510 in the C64 to lose it's timing) - usually the screen was blanked (basically turning off the VIC-II), or at the least, turning off sprites (sprites by the way, while nice, were a PITA becuase they disrupted everything, including raster timing).

Commodore did screw things up... They had four (or was it six?) connectors on each end of the cable, they could have made it at least quai-parallel, rather than the serial with handshaking. Unfortunately, they only hooked up two, CLK (handshaking clock) and DATA (for the data bit). However, seeing as the 1541 was the same hardware mechanism as the 1540 (it's predecessor for the VIC-20) and contained most of the same software (you could use a "user" command to change the speed for the VIC-20), they couldn't just go out and change the design. I almost get the feeling that they took the serial bus from the VIC-20, put it in the C64, figuring that they'd be able to use the 1540 drive. Then at the last minute, they realized that it wouldn't work and they made the 1541, as well as a ROM upgrade for the 1540 to work with the C64.

While getting rid of the handshaking and transferring an extra bit over that line made sense then, with modern computers, I wouldn't trust it. There's too many components from too many manufacturers, and I really like my MP3 and pr0n collections too much to lose them to one bit being corrupted.

-- Joe

nVidia Linux woes

[bleachboy]

I have an nVidia GeForce2 Ultra, and recently upgraded my kernel to 2.5.75. It caused my X graphics to become unbelievably slow -- like 2400 baud modem slow when doing a directory listing or anything where text was scrolling. Downgrading to 2.4.21-ac4 (ac4 needed for some Adaptec drivers) and it was back to fast again. Further, my favorite 3D shooter was about 60 fps faster with the 2.4 kernel. The kernels were compiled identically, or at least as identically as you can get with 2.4 vs 2.5. Here's a few tips I can offer to the nVidia users out there:

• In case you don't know, nVidia provides official (but woefully non-GPL) drivers. They also have a message board which I found to be quite informative at times.
• Compile your kernel with MTRR support. It will speed things up a great deal.
• Compile your kernel without AGPGART support. The nVidia driver(s) are faster.
• If you want to try the nVidia driver with a 2.5 kernel, you'll need a patch.
• If you have an nForce chipset, make sure to add "mem=nopentium" to your kernel boot parameters, or else your system will be incredibly unstable. Better yet, ditch your nForce chipset (I did) since the Linux support totally blows, at least for now. Give your old nForce chipset to your wife, girlfriend, mother, Windows box, or whatever.

The results are not obviously broken

[asnare]

A lot of people are questioning the results claimed by Rasterman; however try downloading the thing and running it for yourself. I see the same trend that Rasterman claims when I do it.

My system: Athlon 800, nVidia 2-GTS.
Drivers: nVidia driver, 1.0.4363 (Gentoo)
Kernel: 2.4.20-r6 (Gentoo)
X11: XFree86 4.3.0

I've checked and:

1. agpgart is being used;
2. XF86 option "RenderAccel" is on.

The benchmark consists of rendering an alphablended bitmap to the screen repeatedly using Render extension (on- and off-screen) and imlib2. Various scaling modes are also tried.

When there's no scaling involved, the hardware Render extension wins; it's over twice as fast. That's only the first round of tests though. The rest of the rounds all involve scaling (half- and double-size, various antialiasing modes). For these, imlib2 walks all over the Render extension; we're talking three and a half minutes versus 6 seconds in one of the rounds; the rest are similar.

I'm not posting the exact figures since the benchmark isn't scientific and worrying about exact numbers isn't the point; the trend is undeniable. Things like agpgart versus nVidia's internal AGP driver should not account for the wide gap.

Given that at least one of the rounds in the benchmark shows the Render extension winning, I'm going to take a stab at explaining the results by suggesting that the hardware is probably performing the scaling operations each and every time, while imlib2 caches the results (or something). The results seem to suggest that scaling the thing once and then reverting to non-scaling blitting would improve at least some of the rounds; this is too easy, however, since while it helps the application that knows it's going to repeatedly blit the same scaled bitmap, not all applications know this a priori.

- Andrew

Render Bench

[AstroDrabb]

I just ran the render bench from the link. The results are pretty amazing.

Available XRENDER filters:
nearest
bilinear
fast
good
best
Set up...
--ROUND 1
--
Test: Test Xrender doing non-scaled Over blends
Time: 22.842 sec.
--
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.501 sec.

--ROUND 2
--
Test: Test Xrender doing 1/2 scaled Over blends
Time: 11.438 sec.
--
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.188 sec.

--ROUND 3
--
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 225.476 sec.
--
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 3.963 sec.

I've experienced this myself.

[Anonymous Coward]

The problem is in *sending* the graphics commands to the hardware. If you're manually sending quads one at a time, I found that for 16x16 squares on screen, it's faster to do it in software than on a GEForce 2 (that was what I had at the time - this was a few years back). Think about it:

== Hardware ==

Vertex coordinates, texture coordinates and primative types are DMA'd to the video card. The video card finds the texture and loads all the information into it's registers. It the executes triangle setup, then the triangle fill operation - twice (because it's drawing a quad).

== Software ==

Source texture is copied by the CPU to hardware memory, line by line.

Actual peak fill rate in software will be lower than hardware - but if your code is structured correctly (textures in the right format, etc) - there's no setup. The hardware latency looses out to the speed of your CPU's cache - the software copy has the same complexity as making the calls to the graphics card. :)

The trick is to *batch* your commands. Sending several hundred primatives to the hardware at the same time will blow software away - especially as the area to be filled increases. Well.. most of the time, but it really depends on what you're doing.

the usual superficial analyses of X11

[penguin7of9]

XRender is a new extension with only a reference implementation in XFree86. The point is to experiment with an API prior to freezing it. I know this may come as news to people who have grown up on Microsoft software, but real software developers first try out various ideas and then later start hacking it for speed. It would be quite surprising, actually, if it were faster than a hand-tuned client-side software implementation.

It will be a while until XRender beats client-side software implementations. Furthermore, you can't just take a client-side renderer and hack in XRender calls and expect it to run fast--code that works efficiently with a client-server window system like X11 needs to be written differently than something that moves around pixels locally.

Re:accelerated?

[Spy Hunter]

Well I ran Renderman's benchmark on my Radeon 9100/Athlon XP 2800 system, and here are the results (edited for lameness filter):

*** ROUND 1 ***
Test: Test Xrender doing non-scaled Over blends
Time: 15.925 sec.
---
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 15.927 sec.
---
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.321 sec.
*** ROUND 2 ***
Test: Test Xrender doing 1/2 scaled Over blends
Time: 7.125 sec.
---
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 7.134 sec.
---
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.133 sec.
*** ROUND 3 ***
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 131.495 sec.
---
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 131.703 sec.
---
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 2.487 sec.
*** ROUND 4 ***
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 113.890 sec.
---
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 113.945 sec.
---
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 1.778 sec.
*** ROUND 6 ***
Test: Test Xrender doing general nearest scaled Over blends
Time: 197.817 sec.
---
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 197.800 sec.
---
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 5.171 sec.
*** ROUND 7 ***
Test: Test Xrender doing general smooth scaled Over blends
Time: 268.509 sec.
---
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 268.656 sec.
---
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 7.507 sec.

Obviously XRender is getting crushed here by Imlib2. There are a million reasons this might be happening, it's definitely worth looking into. In the best Slashdot tradition, here's some wild speculation about what might be causing the slowdown:

• Renderman's code might be giving an unfair advantage to Imlib2. The Imlib2 results are never shown on the screen. However, XRender is tested both with display and without, so it seems like it should be fair.
• Renderman's code might be using XRender in an inefficient way. I'm no X programming expert so I have no idea if what he's doing is the best way to do it, but Rasterman is supposed to be some sort of expert in producing nice fast graphics on X so I'd say this is unlikely.
• XRender might not be hardware accelerated for some reason, probably having to do with driver configuration or something. But geez, does the software rendering have to be that slow? Maybe they could learn something from Imlib2.
• The hotly debated "X protocol overhead" might be causing this slowdown. But given the magnitude of the slowdown, I think this is unlikely.

Hopefully someone knowledgeable like Keith Packard himself will come and enlighten us with the true cause.

Works nice and fast for me

[Trogre]

After installing imlib2, and running render_bench's 'make', it gives me the following:

cc -g -I/usr/X11R6/include `imlib2-config --cflags` -c main.c -o main.o
main.c: In function `xrender_surf_new':
main.c:67: `PictStandardARGB32' undeclared (first use in this function)
main.c:67: (Each undeclared identifier is reported only once
main.c:67: for each function it appears in.)
main.c:67: warning: assignment makes pointer from integer without a cast
main.c:69: `PictStandardRGB24' undeclared (first use in this function)
main.c:69: warning: assignment makes pointer from integer without a cast
main.c: In function `xrender_surf_blend':
main.c:153: `XFilters' undeclared (first use in this function)
main.c:153: `flt' undeclared (first use in this function)
main.c:154: `XTransform' undeclared (first use in this function)
main.c:154: parse error before `xf'
main.c:156: `xf' undeclared (first use in this function)
main.c: In function `main_loop':
main.c:439: `XFilters' undeclared (first use in this function)
main.c:439: `flt' undeclared (first use in this function)
make: *** [main.o] Error 1

It seems to do this at the same speed, whether or not I have render acceleration enabled.

I ran the benchmark with RenderAccel true

[Sits]

And the results were pretty much the same. Using render was several magnitudes slower on tests 2 - 7. I have a GeForce1 with 1.0.4349 nvidia driver and haven't had the same trouble others have with this option on so I run with this extension on all the time.

Here are the results for the interested:

Available XRENDER filters:
nearest
bilinear
fast
good
best
Set up...
*** ROUND 1 ***

Test: Test Xrender doing non-scaled Over blends Time: 0.190 sec.

Test: Test Xrender (offscreen) doing non-scaled Over blends Time: 0.303 sec.

Test: Test Imlib2 doing non-scaled Over blends Time: 0.697 sec.

*** ROUND 2 ***

Test: Test Xrender doing 1/2 scaled Over blends Time: 10.347 sec.

Test: Test Xrender (offscreen) doing 1/2 scaled Over blends Time: 10.231 sec.

Test: Test Imlib2 doing 1/2 scaled Over blends Time: 0.315 sec.

*** ROUND 3 ***

Test: Test Xrender doing 2* smooth scaled Over blends Time: 207.028 sec.

Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends Time: 205.275 sec.

Test: Test Imlib2 doing 2* smooth scaled Over blends Time: 5.695 sec.

*** ROUND 4 ***

Test: Test Xrender doing 2* nearest scaled Over blends Time: 164.460 sec.

Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends Time: 166.281 sec.

Test: Test Imlib2 doing 2* nearest scaled Over blends Time: 4.119 sec.

*** ROUND 6 ***

Test: Test Xrender doing general nearest scaled Over blends Time: 313.187 sec.

Test: Test Xrender (offscreen) doing general nearest scaled Over blends Time: 310.261 sec.

Test: Test Imlib2 doing general nearest scaled Over blends Time: 11.444 sec.

*** ROUND 7 ***

Test: Test Xrender doing general smooth scaled Over blends Time: 477.511 sec.

Test: Test Xrender (offscreen) doing general smooth scaled Over blends Time: 474.695 sec.

Test: Test Imlib2 doing general smooth scaled Over blends Time: 17.290 sec.

(reformatted to get past the lameness filter)

Re:It's RIGHT to use 3D functionality for 2D graph

[rmlane]

On vaugley modern hardware the 3D path is so much faster than the 2D path that it ends up being significantly faster to use the 3D path to render your desktop if your desktop is at all complicated (not a dozen mono xterms).

This ends up being even more true if you do any sort of complex compositing (eg: alpha blending, hardware accelerated mpeg / video, openGL windows, etc, etc). Enlightenment uses alpha channels, it would be fater to composite in hardware than software. These sorts of operations are not accelerated at all on the 2d path, and have to be done in software.

Go check out Quartz Extreme at http://www.apple.com/macosx/jaguar/quartzextreme.h tml (excuse the space in html).

Having used Xfree86 and Quartz extreme on the same graphics hardware, I can tell you there's no comparison. Quartz is much faster and much more capable.

Quartz Extreme in a few words

[The Ego]

Apple's OSX does all rendering through Quartz, (as PDFs) which is accelerated by OpenGL, and called QuartzExtreme.

That's not accurate. Quartz is really made of two parts: Quartz 2D and the Quartz Compositor.

The Quartz Compositor is reponsible for compositing all the layers (desktop, windows, layers inside windows) on-screen. It offers Porter-Duff compositing, which was developped at Pixar more than 15 years ago. See this post from Mike Paquette for details. Mr Paquette is one of the main developpers of Quartz. Quartz Extreme is "simply" an OpenGL implementation of Porter-Duff compositing and modern graphic cards offer the primitives needed to do that very efficiently.

The Quartz 2D layer is what offers drawing primitives following the Postscript drawing model. The same drawing model is used with PDF (no surprise), Java2D and SVG (and Microsoft's GDI+ ?). This part is not HW accelerated. I am sure Apple is working on it, but it wouldn't surprise me if new HW will be required to make this possible. There is a strong incentive for card manufacturers to offer acceleration, since Longhorn is supposed to use GDI+ extensively. I doubt that such acceleration will fit in the traditionnal OpenGL/Direct3D rendering pipeline.

The Apple JVM team implemented HW accelerated Java2D drawing in their 1.3.1 JVM. Their 1.4 JVM doesn't offer it (1.4.1 was a massive rewrite for them, 1.3.1 was more of a quick port to OS-X using some of their "old" carbon code). There were quite a few problems when HW acceleration was used. I hope they can and will wait for a system-wide Quartz-2D HW acceleration, it seems ludicrous to have the JVM team spend resources on an effort that will be wasted once Quartz2D is accelerated.

See Apple Marketing page, another post from Mike Paquette, and the presentation from Apple at SIGgraph about Quartz Extreme and OpenGL.

If that post doesn't end-up rated +5 informative, I don't know what will ! :-)