Slashdot Mirror

Err? All the highest end cards still come in AGP. You don't -have- to upgrade everything to upgrade your video card. If I were making the switch to PCI express and a new video card, I'd just buy a new computer honestly

Yes, and still carry the same hefty price tag..

However, the industry seems to be pushing their PCI-X cards, so that they can :

a) Pump the industry up, sell more chipsets

b) Sell more cards, specifically with their SLI implementation you can find here.

Which basically means, more $$$ out of your pocket to keep up. The SLI looks promising, and delivers numbers to what you'd expect using 2 PCI express video cards at once. And I understand you don't -have- to upgrade to the PCI-X, but when we're talking about games, every little bit counts. Most of these brand new spanking cards see PCI-X first, and are trickling down to AGP now. I believe the industry will shift from AGP to PCI-X just like it did when the AGP slots were first introduced. It sure as hell is pushing for the change.

I've noticed that everything on my nForce2 board was supported after turning APCI to "off" on bootup.

NVidia offers all of the drivers needed for linux here. In fact, they were just updated today (Oct. 20th).

NVidia has done a good job at keeping the drivers up to date, in fact... as much as they do their Windows drivers (if not more). I applaud them for this.

It's not an AC97 soundcard. It's nVidia's own sound doohicky. It was originally developed for the Xbox I believe and since been refinied several times by nVidia for their motherboard chipsets.

There is a unified driver package for Linux, but I'm not sure how well supported the various features are.
http://www.nvidia.com/object/linux_nforce_1.0-0283

I only had time to read Anandtech's preview this morning for the nForce4 chipsets, and I wasn't sure that the Ultra and SLI chipsets would be made available for Socket 754 A64 CPUs.

I checked Nvidia's website for information on this, and I found tech specs for each chipset:

nForce4 - http://www.nvidia.com/page/pg_20041014863476.html
nForce4 Ultra - http://www.nvidia.com/page/pg_20041015990644.html
nForce4 SLI - http://www.nvidia.com/page/pg_20041015917263.html

As you can see -- no specifics on the socket support. I'm wondering if this will be at the discretion of the motherboard manufacturers. My hope is that Nvidia will encourage both Socket 754 and Socket 939 variants of the motherboards with these chipsets.

I'm an owner of a Socket 754 CPU, and I know that a lot of friends invested money as early adopters of the A64 CPU in these Socket 754 platforms. I unloaded nearly $375 for my Socket 754 A64 before AMD started cutting prices and introducing the early, and very expensive, Socket 939 CPUs.

That's an investment that I can't just shirk off in order to take advantage of a much less expensive chipset/motherboard upgrade for, say, $125 for a top tier nForce4 motherboard (just guessing at the pricing here -- don't take it literally).

IronChefMorimoto

I only had time to read Anandtech's preview this morning for the nForce4 chipsets, and I wasn't sure that the Ultra and SLI chipsets would be made available for Socket 754 A64 CPUs.

I checked Nvidia's website for information on this, and I found tech specs for each chipset:

nForce4 - http://www.nvidia.com/page/pg_20041014863476.html
nForce4 Ultra - http://www.nvidia.com/page/pg_20041015990644.html
nForce4 SLI - http://www.nvidia.com/page/pg_20041015917263.html

As you can see -- no specifics on the socket support. I'm wondering if this will be at the discretion of the motherboard manufacturers. My hope is that Nvidia will encourage both Socket 754 and Socket 939 variants of the motherboards with these chipsets.

I'm an owner of a Socket 754 CPU, and I know that a lot of friends invested money as early adopters of the A64 CPU in these Socket 754 platforms. I unloaded nearly $375 for my Socket 754 A64 before AMD started cutting prices and introducing the early, and very expensive, Socket 939 CPUs.

That's an investment that I can't just shirk off in order to take advantage of a much less expensive chipset/motherboard upgrade for, say, $125 for a top tier nForce4 motherboard (just guessing at the pricing here -- don't take it literally).

IronChefMorimoto

I only had time to read Anandtech's preview this morning for the nForce4 chipsets, and I wasn't sure that the Ultra and SLI chipsets would be made available for Socket 754 A64 CPUs.

I checked Nvidia's website for information on this, and I found tech specs for each chipset:

nForce4 - http://www.nvidia.com/page/pg_20041014863476.html
nForce4 Ultra - http://www.nvidia.com/page/pg_20041015990644.html
nForce4 SLI - http://www.nvidia.com/page/pg_20041015917263.html

As you can see -- no specifics on the socket support. I'm wondering if this will be at the discretion of the motherboard manufacturers. My hope is that Nvidia will encourage both Socket 754 and Socket 939 variants of the motherboards with these chipsets.

I'm an owner of a Socket 754 CPU, and I know that a lot of friends invested money as early adopters of the A64 CPU in these Socket 754 platforms. I unloaded nearly $375 for my Socket 754 A64 before AMD started cutting prices and introducing the early, and very expensive, Socket 939 CPUs.

That's an investment that I can't just shirk off in order to take advantage of a much less expensive chipset/motherboard upgrade for, say, $125 for a top tier nForce4 motherboard (just guessing at the pricing here -- don't take it literally).

IronChefMorimoto

The coolness factor of having dual monitors is usually achieved easier than setting it up a dual monitor system for productivity. I think a virtual setup that you can tab to like NVidia's cards are capable of is a solution to be more productive than a dual monitor setup for a windows enviroment. Of course in a Unix enviroment it is much easier to have virtual monitors or screens and in my opinion much easier to control.

Just so you know, nvidia doesn't really sell mobile chips to "custom builders" as they see their market as large OEMs only.

Your best bet might be to find a custom builder who activly supports linux on some models with an ati or intel chipset, or suck up the price and buy something proprietary with windows on it, and convert it to linux yourself.

Best.

I've used this before.. almost a year ago...

Or wait, maybe it's that Keyhole thingy from Nvidia.

NV's versions may have less info and only Nvidia video cards are compatible with Keyhole, but this HAS been done before FYI.

Last time I checked Keyhole had a 14 day trial and a fairly low subscription fee per month.

does anyone know if the nVidia 6600 will be DDL, thus letting people use the 30" Cinema Display?

From the Quadro FX Features and Benefits page:

Single Dual-Link Digital Display Connector
Dual-link TMDS transmitters support ultra-high-resolution panels (up to 3840 x 2400 @24Hz)--which result in amazing image quality producing detailed photorealistic images.

does anyone know if the nVidia 6600 will be DDL, thus letting people use the 30" Cinema Display?

From the Quadro FX Features and Benefits page:

Single Dual-Link Digital Display Connector
Dual-link TMDS transmitters support ultra-high-resolution panels (up to 3840 x 2400 @24Hz)--which result in amazing image quality producing detailed photorealistic images.

The ADC interface made sense in that power, video and USB were brought up to the monitor in a single cable but I'd rather have a larger selection of video cards for my Mac.

Seem's it still experimental:

"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."

ftp://download.nvidia.com/XFree86/Linux-x86/1.0-61 11/README.txt

Only two years behind the times!

Just different enough from existing GPU programming languages to be annoying, without any added functionality or ease of use!

No standard intermediate representation, requiring OpenGL drivers to contain full-blown compilers! Hello, latency!

OpenGL -- the best API and shading language a politics-laden commitee could design!

Seriously, if it weren't for Mr. Carmack, the dinosaur that is OpenGL would be deader than the dodo bird. Sad, as I spend half my day developing OpenGL apps, but true.

Added support for Linux Installer

Added support for SoundStorm (Hardware Mixing supported)

Added support AC3 pass-through

Added support for Ethernet driver statistics and configuration information through procfs

Added support for 2.6 series kernels

Added a new audio application, NVIDIA NVMixer, to do Volume control per channel, input selection, Speaker selection, Speaker cloning, Swap mic to Center/Lfe & LineIn to Surround L/R

NVMixer is nice, both under Windows (XP) and linux. Under XP it allows you to go through their speaker wizard. Makes for easy downmixing, making my 4.1 speaker system sound like 5.1

The computer is an *ducks* eMachines T3025. Works great in the sound area. Just plug the "back" speakers into the line-in and the center/sub into the microphone (even without the Nvmixer application, check these option in Windows mixer's advanced section under master)

Like I said, I've got a Cambridge Soundworks (now creative) brand 4.1 speaker option that performs nicely. They are old but work like anything bought today. In fact with the NVMixer application you really don't need 5 speakers and a subwoofer. Just tell it you've only got 4 (or two) and it will figure out where to put the sound. Remember it all depends on the timing. When you think something is coming from a certain point it's because when it hits your ears. The mixer just delays certain sounds and works flawlessly.

.... I find amusing about this article is the NVIDIA add.

On the nVIDIA web site, following the add's link:

"A test performed by NVIDIA computer security experts found that a PC was detected and hacked within 15 minutes of being connected to a standard high-speed DSL broadband Internet connection. Within 24 hours, hackers loaded over 1,000 songs files using the PC's hard disk space as a storage device. This test illustrates how quickly hackers can steal your passwords and gain access to sensitive information such as your bank account."

I bet'ya all the hackers in the world are running after DSL connected PCs to use their storage space and system as digital jukeboxes...

I should have thought of it before!! A great way to get MP3s and not risk being sued by the RIAA.

As WIAKywbfatw points out, GPUs became more powerful than CPUs (on a FLOP basis) a decade or more ago. This was the whole reason Intel created the AGP port - to prevent the GPU from becoming the center of the the computer (it was a huge threat to their business).

nVidea used the term GPU to refer to its fixed function T&L capable NV10 chip, which was released on August 31, 1999 as the GeForce 256.

The AGP 1.0 standard dates to 1996, and it was intended to provide fast bandwidth for textures and video. The video cards it fed were generally incapable of running shaders, geometry transformations, or lighting calculations.

Programable floating point units are a relatively recent development-- the functionality was introduced to the mass market when the Radeon 9700 and GeForceFX 5700 were released.

The GPU is *great* at them. Anything that can be expressed as a massively parallel simulation or effect is a cake-walk for the GPU.

As the author mentions, the bottleneck right now is getting data back from the GPU. However, the more successive steps you can do in a row on the GPU, the better off you are. Even with the new PCI express, this will remain true. The more successive steps you can do without having to retrieve the data, the faster your processing will be.

There are lots of cool demos over at NVIDIA's website that exploit the cool things that can be done with the parallelism of the GPU.

I've been toying with writing a 'Go' simulator on the GPU. Just haven't gotten around to it yet.

It's not possible with AGP, but have no fear. PCI Express to the rescue!

As mentioned, the entire General Purpose GPU (GPGPU or GP2U) community has been on this a while. They recently put on a SIGGRAPH 2004 Course and there is the GPU Gems book which as some GPGPU in it.

HDTV output, not hdtv-ready.

The 6600 "Provides world-class HDTV-out functionality up to and including 1920x1080i resolutions." The 6800 does not.

This doesn't mean it's not the better choice for a media center, but it looks like it has the HDTV stuff.

No such thing as a free lunch, baby.

What's certain, according to her, is that even though Genoa's technology increases the range of colors, it's not recovering the full original color information of a movie on film, lost in the conversion to other formats, like DVD. "It's kind of arbitrarily making images look better," she says, though people will in fact prefer the resulting colors, which will typically be more saturated and brighter.

Various video media may not have the necessary color resolution to drive these displays, but (given quality art assets;) newer video cards do.

I wonder how these types of displays compare to Iridigm's upcoming products on color fidelity. Those look quite interesting, especially at effective 200 DPI.

"NVIDIA's latest generation of chipsets gives exceptional performance and feature gains across the entire line, from the consumer cards to the specialty cards. I use a GeForce 6800 class card in my primary workstation, which is the best endorsement I can give."

If you are doing this through a computer, you may be interested in a feature Nvidia has in their drivers. "NVKeystone" display correction they call it, and works on pretty much all nvidia gpus after the Geforce 256, so you could just purchase cheap Gf2 MX or such. http://www.nvidia.com/object/feature_nvkeystone.ht ml

One product that striked me as really slick was an image manager being demonstrated @ the nVidia booth called Taos Image Finder.

Apparently, it uses OpenGL shaders or GLSL, which is now part of the recently announced OpenGL 2.0, to do real time image filtering. It's pretty neat. The user interface is kinda space age and nothing like what you would expect to see in an image manager. It can search images based on color and shape and the results were very accurate. Definitely worth a look.

The stonybrook monitor was another stunning demo and it was hard to believe that such fantastic imagery could be displayed on a monitor. Apparently it costs somewhere around $90K :) , a tad expensive compared to my $150 19" :)

Long ago John Carmack said in Jan 2003 Doom3 was going to have a ARB1, nv1x, nv2x, R2x0, nv3x, and ARB2 path. We recently heard the nv3x path was dropped now that nvidia's driver compiler does a good enough job on optimizing a ARB_fragment_programs (Pixel Shader 2.0 in D3D terms) that the nv3x isn't needed. But is the ARB1 path still around?

The article says the min spec is:
*Supported 3D Graphics chipsets:
ATI: Radeon 8500, 9000, 9200, 9500, 9600, 9700, 9800
NVIDIA: GeForce 3, GeForce 4MX, GeForce 4 Titanium, GeForce FX, GeForce 6
It doesn't list other manufacturers, but I don't know if this really implies other manufacturers currently can't run it.

What about the PowerVR Kyro (no cube map support), Radeon 7x00, Intel Extreme Integrated, SiS Xabre (useless drivers), Matrox Parhelia, 3DLabs P10/P9? Without the ARB1 path they wouldn't be able to run the game. Well 3Dlabs used to support nvidia's register combiner extensions so could use the nv1x path if their drivers are up to snuff. It is one thing to be able to run the game quickly (the reson the GeForce4MX is supported when it is technically just a quicker version of the older GeForce/GeForce2/GeForce2MX) to ensure the game is enjoyable, but it is another thing to not be able to run it at all due to not supporting modern OpenGL extensions. It would be nice if older cards could run Doom3, slowly and without specular, instead of not at all.

Also what about GLSL? Even longer ago Mr. Carmack said in June 2002 (wow, they've been working a long time!) "I am now committed to supporting an OpenGL 2.0 renderer for Doom through all the spec evolutions" in refernce to the GL shading language. Will there be a Doom3 renderer which uses the high level GLSL extension instead of the fragment program extension? I get the impression there won't be, and it would be pretty pointless with the fragment program support, but 3Dlabs currently only supports the high level fragment shaders, not the low level fragment programs...

Or, I'll go out and buy a 5500 for $76.00 [newegg.com], and play it in medium resolution. Or the best solution of all. I will wait until Christmas, when NVIDIA and ATI bring our their next card, and the 6800 256MB is old news, and can be purchased for $99-$20MIR=$79.

As it turns out, the 5500 is actually a weaker card than the GeForce 4 Ti line. Compare and contrast:

Geforce FX 5500 (the page is titled "5200" but features the 5500 specs)

--Memory bandwidth: 6.4 GB/s
--Fill rate: 1.1 billion texels per sec.
--Vertices per sec: 68 million

GeForce4 Ti:

(8x AGP GeForce4 Ti4200):

--Memory bandwidth: 8GB/s
--Fill rate: 4 billion AA samples/sec.
--Vertices per sec: 113 million

And come X-mas, nVidia and ATi will likely come out with 512MB versions of their cards, instead of higher model numbers. It was about a five-month gap between the 128MB and 256MB 9800 Pro.

You'll also likely never see the 6800 GT or Ultra anywhere near $100--not until they're certifiably ancient. 256MB of GDDR3 is expensive, no matter how you slice it. DDR2 currently costs about twice that of DDR1, per megabyte. The price of the card just won't come down that fast. I expect more than a few jaws to drop when the MSRPs for the 512MB cards are revealed.

Or, I'll go out and buy a 5500 for $76.00 [newegg.com], and play it in medium resolution. Or the best solution of all. I will wait until Christmas, when NVIDIA and ATI bring our their next card, and the 6800 256MB is old news, and can be purchased for $99-$20MIR=$79.

As it turns out, the 5500 is actually a weaker card than the GeForce 4 Ti line. Compare and contrast:

Geforce FX 5500 (the page is titled "5200" but features the 5500 specs)

--Memory bandwidth: 6.4 GB/s
--Fill rate: 1.1 billion texels per sec.
--Vertices per sec: 68 million

GeForce4 Ti:

(8x AGP GeForce4 Ti4200):

--Memory bandwidth: 8GB/s
--Fill rate: 4 billion AA samples/sec.
--Vertices per sec: 113 million

And come X-mas, nVidia and ATi will likely come out with 512MB versions of their cards, instead of higher model numbers. It was about a five-month gap between the 128MB and 256MB 9800 Pro.

You'll also likely never see the 6800 GT or Ultra anywhere near $100--not until they're certifiably ancient. 256MB of GDDR3 is expensive, no matter how you slice it. DDR2 currently costs about twice that of DDR1, per megabyte. The price of the card just won't come down that fast. I expect more than a few jaws to drop when the MSRPs for the 512MB cards are revealed.

Chicken of the Sea Girl

I heard she got a job modelling for Nvidia.

* How does the NVIDIA GeForce 6800 Ultra DDL card for the Mac differ from the PC version (i.e. Does the PC version have dual link DVI?)

The GeForce 6800 Ultra DDL card was designed specifically for the Mac to provide two dual-link outputs to support Apple's displays.

* Does the Apple version of the GeForce 6800 Ultra GPU run at the same core/memory clock as the PC version?

The Apple cards run at 400/550, just like the GeForce 6800 Ultra GPU on the PC.
(Note: Some vendor's 6800 cards are clocked higher than the standard/reference design.)

* The GeForce 6800 Ultra for the PC has two Molex power connectors - does the Mac version source all the power from the G5's AGP pro slot? (or does it have a aux power connector?)

There is an on-board power connector on the graphics card and the motherboard to provide power, so there is no need for an external power connector from the power supply.
(although the only Mac 6800 photos I've seen are tiny, it appears there's a stub connector on the card that (I suspect) uses the ADC (28V or 24V usually) DC power connector on the motherboard that's normally used for ADC display power to provide additional power (regulated down) for the 6800 card. That eliminates the need for Aux. (Molex) P.S. connector(s) like the PC/standard 6800 card versions have.)

* Does the GeForce 6800 Ultra DDL have a low-noise fan?

Yes, the GeForce 6800 Ultra DDL runs very quiet.

* Will there ever be a control panel with 3D/GL/FSAA controls for the NVIDIA cards on the Mac platform? (ATI's retail Radeon cards (and OEM models with the 3rd party patch) have a '3D/GL overrides' feature - which is seen as a big plus by many end users.)

Apple provides all the drivers for NVIDIA-based add-in cards. We supply them with the source code and they provide the final driver.

* Regarding the previous question - if there's no chance of an Apple supplied NVIDIA card control panel (for advanced features/FSAA, etc.) - if a 3rd party wanted to do this, can NVIDIA provide some assistance?

Apple is our customer, so if this is something that they requested, then we would support it.

* There's been talk of previous NVIDIA cards taking a bigger than expected performance hit from using some types of shaders (on the Mac) - is this a concern with the GeForce 6800 Ultra DDL?

GeForce 6 Series GPUs have the most advanced shading engines on the market. Compared to previous generation parts, the vertex and pixel shader engines on GeForce 6 Series GPUs have been completely redesigned from the ground-up. Pixel Shader performance is 4 to 8 times faster. Vertex shader performance is twice as fast. Performance numbers have been outstanding.

* Will there updated/new drivers for the GeForce 6800 Ultra?

Yes. Apple provides all the drivers for NVIDIA-based add-in cards. We supply them with the source code and they provide the final driver. Apple will control the release schedules for drivers that provide even more performance, features and image quality enhancements.

* Do you have any idea how performance compares on the Mac between the GeForce 6800 Ultra and the ATI 9800 Pro/XT card?

GeForce 6800 Ultra represents the largest leap forward in graphics performance in our company's history. As

You have geforce 4 cards in your work desktops? What are folk doing in your office that they need 3D accellaration?

http://www.nvidia.com/object/feature_nview.html Putting dual-desktops in front of our powerusers helps ease the stress levels during those busy multiple document edit/copy/paste sessions. Besides, they think it's cool and the cards are pretty cheap these days.

Cg/HLSL are definitely older and hence have picked up developer mind share, but being the first few high-level realtime shading languages, they have design flaws. GLSL is based on an excellent design and if your API of preference is OpenGL, then GLSL is what you will want to use.

Probably the most eagerly awaited feature for OpenGL developers, everything in the OpenGL state is directly accessible using built-in variables in the shader and need not passed explicitly from the program before executing the shaders.

'varying' variables are another cool design feature, where you can pass data from the vertex shader to the fragment shader directly.

GLSL is based on excellent design and it is only a matter of time before OpenGL developers switch to it. I, for one, have switched to GLSL. nVidia's latest SDK (7.0) already has many GLSL demos and their 0.60 drivers for FX and 6800 already support GLSL.

--

http://www.nvidia.com/page/geforce4ti.html

It doesn't?

Trevor

Except for that one NVidia made.

It might worth noting that the nVidia drivers used in the Anandtech benchmark are from January. nVidia released new 6xxx drivers for both IA32 and AMD64 on June 30.

It might worth noting that the nVidia drivers used in the Anandtech benchmark are from January. nVidia released new 6xxx drivers for both IA32 and AMD64 on June 30.

Yes. The specific extension is known as ARB_texture_non_power_of_two. This was supported by the Radeon and the GeForce cards, but not by the Rage128. This last chipset is used by my iBook. The TNT2 was not used in any Macintosh.

By the way, Quartz Extreme can be turned off by using 8 bit colour.

What pain and suffering? Nvidia's drivers are the most pain-free to install of any third-party driver. The useful module options are well documented, and there is a support community. Hell, Nvidia just added a new configuration utility. Unlike other drivers that are included with the kernel, Nvidia's modules are consistently functional, and aren't mystified by outdated HOW-TOs or the requirement to poke around inside the kernel source tree for a text file with module options; A file that may or may not exist.

On the contrary, my issue with hardware installation and Nvidia had to do with the open source components not produced by Nvidia. The Mesa OGL library changed their build scripts (which was the source of much confusion) because of issues with automake. Installing Mesa at the wrong time breaks Nvidia's OGL interface. The agpgart module was also the source of much frustration because it wouldn't support faster transfer modes on certain chipsets. So, when I couldn't get my GF3 Ti to run at 4x, I discovered that - in true OSS developer fashion - the option appears only inside a source code file, with no explanation. Thank heavens for the 'modinfo' utility and experience, whereas a less technically-inclined user wouldn't have a clue what an "int" is.

Nvidia shows a lot enthusiasm for GNU/Linux that other vendors do not. Their Linux drivers are always current, and well documented. They host a lot of complete or demo games for free and without registration hassles. Nvidia is fighting tooth-and-nail with ATI in a technology war that has resulted in ultra-high performance at affordable prices, yet even with ATI nipping at their heels, they've still managed to find the time to cater to the OSS community. Since there is a wide range of acceleration features that modern applications may or may not support, Nvidia's peformance secrets still remain inside their drivers. Putting these trade secrets out in the open would guarantee Nvidia's end. As much as I'd like to see the drivers become part of the kernel's source tree, I wouldn't want to see a friend of open source operating systems put out of business. Continued criticism of Nvidia only reinforces the Linux community's reputation as the enfant terrible of the computing world.

What pain and suffering? Nvidia's drivers are the most pain-free to install of any third-party driver. The useful module options are well documented, and there is a support community. Hell, Nvidia just added a new configuration utility. Unlike other drivers that are included with the kernel, Nvidia's modules are consistently functional, and aren't mystified by outdated HOW-TOs or the requirement to poke around inside the kernel source tree for a text file with module options; A file that may or may not exist.

On the contrary, my issue with hardware installation and Nvidia had to do with the open source components not produced by Nvidia. The Mesa OGL library changed their build scripts (which was the source of much confusion) because of issues with automake. Installing Mesa at the wrong time breaks Nvidia's OGL interface. The agpgart module was also the source of much frustration because it wouldn't support faster transfer modes on certain chipsets. So, when I couldn't get my GF3 Ti to run at 4x, I discovered that - in true OSS developer fashion - the option appears only inside a source code file, with no explanation. Thank heavens for the 'modinfo' utility and experience, whereas a less technically-inclined user wouldn't have a clue what an "int" is.

Nvidia shows a lot enthusiasm for GNU/Linux that other vendors do not. Their Linux drivers are always current, and well documented. They host a lot of complete or demo games for free and without registration hassles. Nvidia is fighting tooth-and-nail with ATI in a technology war that has resulted in ultra-high performance at affordable prices, yet even with ATI nipping at their heels, they've still managed to find the time to cater to the OSS community. Since there is a wide range of acceleration features that modern applications may or may not support, Nvidia's peformance secrets still remain inside their drivers. Putting these trade secrets out in the open would guarantee Nvidia's end. As much as I'd like to see the drivers become part of the kernel's source tree, I wouldn't want to see a friend of open source operating systems put out of business. Continued criticism of Nvidia only reinforces the Linux community's reputation as the enfant terrible of the computing world.

Well, the card is more likely spec'd for the G5, but as you can see, it's very similar to the newly announced NVIDIA cards for the rest of the people. GeFroce 6800. Similar price too...

What are you talking about? I'm running the Nvidia drivers from their site on my Core 2 workstation and it works fine.
Also they release the drivers for "Linux" rather than any particular distribution. The drivers from the link above will work with any of the 2.6.x kernels, just download and run the installer.

Which begs the question, who is this aimed at?

Something that I'm really interested in knowing is how it compares to Alienware's Video Array technology. Video array suffers from the problem of being totally useless, because it only increases fillrate but it doesn't improve shader performance. Almost no game these days is fillrate-limited, but many games are limited by shader performances. I hope that NVIDIA's SLI solution is more interesting.

Obviously, the other big thing here is just to have another one-up over ATI. :)

This is not at all true for most games unless the game requires the same under Windows.

I have played a number of games that actually had improved framerates under WineX 3. It seems all that Windows backend stuff puts a higher tax on the system than most people realize, since the dev team for Windows focuses on making things feel responsive at the cost of overall performance. When it comes time to do purely heavy computations (such as in games), this approach costs CPU time that would otherwise have gone to the computation, ultimately resulting in lower framerates.

OpenGL games require very little emulation from the WineX engine since most of those calls are 1-to-1 Windows OpenGL to Linux OpenGL calls, and the performance cost is a fraction of a percent for these games. Almost all OpenGL games would perform better under WineX Linux than they do in their native Windows environment.

Even DirectX games which require a fair amount more emulation since DirectX and OpenGL do not line up on a call-to-call basis suffer very little if any performance loss for most games.

There are notable exceptions. American McGee's Alice was talked about elsewhere, and this relies heavily on kernel synchronization calls. Because of an architectural difference between Windows and Linux, there's no way to accomodate the same thing directly under Linux (there are other methods to accomplish the same tasks) with out a kernel patch, which is of course very unportable. Early reports state that Alice has a significant performance increase under the new WineX, "Cedega."

Also, some people are not aware that you must have an accelerated driver for your video card in order to do OpenGL with any level of performance. Basically for most systems, the driver you get by default is a non-accelerated driver, optimized for 2-D performance. The reason for this is that most card manufacturers (such as nVidia) have proprietary drivers which you need to download from the vendor themselves. Nvidia's are available here for 32 bit Intel processors. If you do not run an accelerated driver, you'll see terrible performance as all of the 3d and graphic computations are done by the CPU with out being able to use the video card for any of these tasks. It's akin to trying to run a game in Windows using the "Generic VGA Driver," but the difference is that Windows wouldn't let you do this at all since they have no software based GL emulation layer like Linux does. Linux will at least let you try to play the game.

So, the Linux drivers exist, and I know they have hardware-accelerated OpenGL support built in... Does anyone know if this machine+display combo will work "out of the box" without any tweaks to applications?

An option or two applied to XF86Config is OK by me, but per-application mods would be a little excessive... Anyone tried UT2K4 on one yet?

One word: Gelato

This already exists. Do a Google Search for "Geltato renderer."

NVIDIA wrote a great renderer, with all the features of the best software renderers out there (yes, including raytracing), that runs on their GPUs instead of the computer's CPU. Tremendously fast.

You might want to look into nVidia Gelato. It's a 3D renderer that uses Quadro FX cards as secondary FPUs, supposedly doubling or more the speed of rendering. They claim it's two to six times faster than the leading renderer. There's a demo, so you can verify those claims for your uses.

It runs under Linux, and "will function with whatever [render farm] management system you currently use.".

To reiterate, it's a SOFTWARE renderer, that is hardware accelerated by using the video card as a co-processor.

I think you'll find out rather quickly that it isn't very nice to work with. There's a very good reason for accelerating on the card. A quick list of what the graphics card can do faster than the CPU:

• bitmap block transfers - The video card has much faster access to framebuffer because it doesn't have to go through the bus to get to it. It also means less interferance with other devices that may be stuck on the same bus and/or bridge as the video card.
• Hardware cursors - The video card can move a hardware cursor around on the screen without the need to transfer the data the cursor is going to obscure back to the PC's main memory. It's a small cost, but it adds up quickly. It can be the difference between a silky smooth mouse movement, and one that is a bit jerky.
• Video playback - Even at 2.xGHz, a PC CPU is still slow at resizing and colorspace converting a video stream. To do a video playback at 1600x1200, you need at least 220MB/s of bandwidth, and that doesn't even include double or triple buffering, trying to sync to vertical refresh, the actual video decompression or even sending audio data to the sound card. All modern video cards support a video overlay capability where you paint the target area with one color, program the card to display the overlay over the painted area, and then start uploading your unscaled video data to part of the video card's memory. The card takes care of colourspace conversion (if needed), scaling and sometimes even filtering.
• Anything where memory would need to be transfered twice - Any operation where you'd need to transfer the memory out of the video card, work on it, and then transfer it back is worth accelerating in the video card itself. AGP 4x might be able to do about 1 GB/s to/from the video card, but the video card can do much more to/from it's own memory (the GeForce FX 5700+ cards can do a whopping 30GB/s, but even value models can do between 4-6GB/s).

Let's link the latest FreeBSD nvidia drivers at least. These are still a little dicey on the 5.x branch, though. Something about using static ldt allocations. I recommend turning off GL hacks in your screensaver and you can probably get by with your GL games. More informed comments, as always, are welcome.

K.C.