Slashdot Mirror

The superproprietary Shitvidia GPU present in MacBook Pros is so buggy it dies in a few months: http://macmegasite.com/node/4435

Glass, fuck NVIDIA in the ass.

The CUDA SDK includes dozens (maybe more than 100) of sample programs. Use these.

I agree there is a subset of game programmers who can take on the roll of scripting. Lua itself and Lua-like clones are facilitating this. But the industry really wants lower level programmers. Guys who can think not in terms of object-think wait()-do()-wait(), but guys who can implement a smartptr system to create a better garbage collection system or guys who plan thread synchronisation rather than just wait and see.

Matt

Soft/volumetric particles have been around for a long time now. The DirectX 10 sample implementing this technique is from June 2006. NVidia's whitepaper on it [1] is dated January 2007. Both of them offer a much better explanation than TFA with its crude ascii diagrams and wall of text (and having to click "next" a bunch of times, ugh).

[1] http://developer.download.nvidia.com/whitepapers/2007/SDK10/SoftParticles_hi.pdf

There are tons of other folks getting just as phenomenal results using NVIDIA's GPUs:
http://www.nvidia.com/object/cuda_showcase.html

Others are beating large supercomputers with just two GPUs.

The key to why they were able to do this is the CUDA C programming environment:
http://www.nvidia.com/cuda

to expand on that a bit if you're looking to get your hands dirty with cuda & linux you can start doing so here: http://www.nvidia.com/object/cuda_get.html#linux

Nvidia offers an external GPU solution specifically for "deskside supercomputing", the Tesla D870. It has only 2 cores with 1.35GHz each, apart from it being a bit more expensive, I wonder how it compares (you can connect several to a PC).

They are useful for applications that can be massively parallelized

You don't need a super-duper CPU for text editing, that's for sure. For most of the tasks people do on computers, we have had CPU enough for the last 15 years or more. But where we still need more CPU happens to be mostly in tasks that ARE massively parallel, for instance, physics simulations, of which you will find several examples in the nVidia site.

I'm following this technology with much interest, and I think I will have a major upgrade in my home computer soon. My old FX-5200 card has been more than enough for my gaming needs, but now I have a new reason for upgrading.

For information on their current HPC platform checkout http://www.nvidia.com/object/tesla_computing_solutions.html FWIW I don't think there would be that big of performance advantage of putting the GPUs on the motherboard, infact you'd probably actually get a performance decrease if you UMA'd the memory. With discrete boards each GPU has it's own framebuffer resulting in higher memory bandwidth.

Looks like I'll be trying this tomorrow:

Appendix E. OpenGL Environment Variable Settings, which was referred to by this thread, which happened to be the first hit for "compiz anti-aliasing" in Google. Go figure. :)

For me, fullscreen TV output. There's none in 8xxx series and it's broken in older card drivers. See as an example.

But you loose control when using OpenGL. It is fine for games and real-time graphics where speed is most important (and blender already uses OpenGL for its interface of course). But different cards can render things significantly different and so it isn't particularly useful for final rendering, which is what I think the parent was talking about. However because it is more deterministic, CUDA/CTM could be useful for rendering, especially now that cards are getting 32-bit floating point. Gelato is an example of a hybrid CPU/GPU render that uses CUDA (or whatever NVIDA used internally before that) to make a fast, but high-quality renderer.

Have you ever shopped around for laptops with removable cards

[...] before, let alone performed repair work for a company that manufactures cards with swappable graphics cards?

Traditional CAD packages in the 1990's were built on top of Phigs and PEX, Phigs Extension to X.

Official specification

Other animation packages were built on top of proprietary API's like SGI GL and others. It was cheaper building an emulation layer mapping Phigs and SGI GL commands to OpenGL that to rewrite the applications altogether.

At present, the latest API is CUDA

http://www.nvidia.com/object/quadro_geforce.html
The whitepaper says that Quadros have got support for window clipping, hardware accelerated clip planes, antialiased wireframe rendering, more memory, etc. Although it doesn't say if the hardware accelerated features do exist in the GeForce family but are disabled by software.

If you are using the pro drivers with a consumer card, then you are using the drivers unlicensed, which is no different from using any other piece of software unlicensed.

That's because the Quadro drivers are optimised for accuracy, since you are using them to do real calculations you will rely on, rather than small-ish floating point which is all the regular gforce allow.

There are some other things, optimised anti-aliasing for lines, interface layering over the top of render windows, etc.

For a quick and dirty explanation, see NV docs here (warning, pdf file), page 2 onward is where it gets interesting.

Apple systems already share some properties with gaming consoles, namely the harware homogeny of Apple systems.

While to me an annoyance, this standardization might actually work in Apple's favor when trying to woo game makers, as it could act to simplify development.

Apple controlled does NOT equal consistency...

Look at the various models you can buy now, there are several classes of Video Cards, CPUs and that alone is enough to make it a mess.

When you add in the messy nature of the Apple Driver model in OS X, it becames even scarier for games makers. (Apple's desing of working around the monolithic nature of the BSD interface when creating the hybrid OS X driver model has a lot of trade offs that are troublesome, and require kernel level reworking to get the performance up to the level to compete with NT in gaming.)

Mac users expect to buy a Mac and have their Mac software work well. Games running on a low end Video card Mac will PO users faster than anything, as they don't all get the whole 'specifications' of the hardware.

Just with shoving WDDM into Vista, Microsoft has moved forward in creating a consistent gaming environment, and they already own the gaming world because of the past consistency, even though it is has been considerably more inconsistent than a console.

DX10 and especially 10.1 with the WDDM model ensure than games even render consistently with consistent performance levels, even though games on the shelves don't yet reflect the Vista only and forward crowd. (DX10 won't even have performance benefits until games are running on a full DX10 engine, instead of extending textures past DX9.0c, adding in new shadow modeling and calling it DX10, as the engine is still set up for the DX9.0 optimizations and design model. When you see a game on the shelf that says DX10 and Vista Only the quality and performance of the game will reflect what DX10 and 10.1 is bringing when it 'truly' gets here.)

Apple already tried to push into gaming with Leopard and it fell flat on developer's ears with the updated OpenGL support, that was already showing some age in the OpenGL/Graphics world, and the Apple implementation is less than stellar.

Besides, you can do a simple search and find 1,000s of posts like this already, and Developers are not too happy about these types of responses. (Even the initial Vista backlash on gaming pushed developers to abandon DX10 only titles for an additional year until Vista's drivers caught up to XP. And getting OS X drivers to the same level of gaming performance would be a nightmare with NVidia/ATI/Apple all fighting.)

http://forums.nvidia.com/lofiversion/index.php?t41615.html

As an article earlier this month pointed out they are in fact in the process of porting the CUDA system to CPUs.

The advantages would be (assuming this is the wonderful solution it claims) you run your task in the CUDA environment, if your client only has a pile of 1U racks then he can at least run it, if he replaces a few of them with some Tesla racks, things will speed up a lot.

I did some programming at college, I do not claim to know anything about the workings of Tesla or CUDA, but it sure sounds rosy if this stuff would work.

those of you that use the nvidia driver from http://www.nvidia.com/object/unix.html will find the current release NVIDIA-Linux-x86-169.12.pkg1.run will not work with linux-2.6.25

i built 2.6.25 this morning and used menuconfig to load my .config from 2.6.24.4 and once i had my kernel built and rebooted and when i went to build the nvidia driver it failed so i reinstalled 2.6.24.4 which works run the nvidia driver without problems(waits patiently for a new nvidia driver)...

Thank you for your predictions, oh wise seer. Though I'm not sure if any of them will come true. However, I appreciate your unique interpretation of free software philosophy.

Here is a post from the Nvidia/CUDA forums from Mike Houston, one of the Folding at Home people: http://forums.nvidia.com/index.php?showtopic=28868&view=findpost&p=224490

Um you can, it's CUDA, you can read (yeah right) more about it http://www.nvidia.com/object/cuda_home.html and http://en.wikipedia.org/wiki/CUDA

http://www.nvidia.com/object/cuda_get.html offers: "NVIDIA Driver for Linux with CUDA Support (169.09)"

So, physics should work on Linux, having been ported to CUDA already, and CUDA being cross-platform, but the question is if any Linux games will actually support and/or make use of it.

Matlab? You meaned CUDA...

But here there is a plugin for matlab :)
Enjoy! http://www.nvidia.com/object/cuda_home.html

This particular model (Shuttle XPC SN27P2) has a 400W PSU, and I've googled around and read through some discussion boards where people claim, at least, that they have run the SN27P2 with an 8800GT. I'd like to find some more specs on the actual power output per rail from the PSU before buying this all together, though.

NVIDIA has a neat graphics and PSU comparison Flash webpage where you can drag sliders to specify your PSU wattage, and it'll recommend video cards accordingly. At least according to them, a 400W PSU should be sufficient to power 8800GT class cards.

You're not a gfx programmer yourself, are you? Otherwise you'd be linking to nVidia specifications, not consumer reviews. The relevent OpenGL extension is GL_ARB_color_buffer_float, which was indeed implemented for G70-class hardware as of R75 drivers (actually, GL_NV_float_buffer.txt was implemented even earlier).

Yes, you can use this for offscreen framebuffer objects and pbuffers, which is all you need when float texture blending for HDR rendering, but this is then tone-mapped to the 32bit displayable framebuffer for output. It's still not possible to get more than 8bit RGB actually out of the chip. Apart from SGI (who patented float rasterisation), I've only heard of an old Matrox card claiming to do real 10bit integer RGBA output (under quite specific conditions, apparently). Even nVidia's current high-end Quadros can't do it (well, unless you count 10bit 4:2:2 YUV from the SDI connector on some models). I'd welcome any comments showing real evidence to the contrary (preferably from someone who hasn't been repeatedly modded down as a troll), but I've never seen it done.

I can see it easily with my own eyes

The PS3 can decode TrueHD into PCM

There it is right there. Yes, the PS3 player supports TrueHD, but it does notpass it over the HDMI link - it gets decoded to good ol' HDMI 1.0-standard multichannel PCM first. Read the rest of the article - a Sony rep has confirmed this. And AFAIK the PS3 still does not yet support DTS-HD; it only passes through the DTS component. Incidentally, I found it ironic that you're accusing me of trolling :-)

You don't have to convince me that the PS3 is good hardware. It certainly has the edge in CPU power, and the Blu-Ray player is a valuable addition (though it's also the primary reason Sony released late and expensive, throwing away their lead from the PS2). Its graphics are debatable though, and most unbiased people consider PS3 and Xbox 360 GPU power to be roughly equivalent. More on topic, the PS3's HDMI port is more capable than the 360's (which can't even pass multichannel PCM) - but the HDMI 1.3 output is pure marketing, nothing more. Most TV sets (even those that accept deep colour) still can't actually display it, only use it for cleaner tweaking. Certainly no plasma or consumer LCD panel that I'm aware of is capable.

You're not a gfx programmer yourself, are you? Otherwise you'd be linking to nVidia specifications, not consumer reviews. The relevent OpenGL extension is GL_ARB_color_buffer_float, which was indeed implemented for G70-class hardware as of R75 drivers (actually, GL_NV_float_buffer.txt was implemented even earlier).

Yes, you can use this for offscreen framebuffer objects and pbuffers, which is all you need when float texture blending for HDR rendering, but this is then tone-mapped to the 32bit displayable framebuffer for output. It's still not possible to get more than 8bit RGB actually out of the chip. Apart from SGI (who patented float rasterisation), I've only heard of an old Matrox card claiming to do real 10bit integer RGBA output (under quite specific conditions, apparently). Even nVidia's current high-end Quadros can't do it (well, unless you count 10bit 4:2:2 YUV from the SDI connector on some models). I'd welcome any comments showing real evidence to the contrary (preferably from someone who hasn't been repeatedly modded down as a troll), but I've never seen it done.

I can see it easily with my own eyes

The PS3 can decode TrueHD into PCM

There it is right there. Yes, the PS3 player supports TrueHD, but it does notpass it over the HDMI link - it gets decoded to good ol' HDMI 1.0-standard multichannel PCM first. Read the rest of the article - a Sony rep has confirmed this. And AFAIK the PS3 still does not yet support DTS-HD; it only passes through the DTS component. Incidentally, I found it ironic that you're accusing me of trolling :-)

You don't have to convince me that the PS3 is good hardware. It certainly has the edge in CPU power, and the Blu-Ray player is a valuable addition (though it's also the primary reason Sony released late and expensive, throwing away their lead from the PS2). Its graphics are debatable though, and most unbiased people consider PS3 and Xbox 360 GPU power to be roughly equivalent. More on topic, the PS3's HDMI port is more capable than the 360's (which can't even pass multichannel PCM) - but the HDMI 1.3 output is pure marketing, nothing more. Most TV sets (even those that accept deep colour) still can't actually display it, only use it for cleaner tweaking. Certainly no plasma or consumer LCD panel that I'm aware of is capable.

You're not a gfx programmer yourself, are you? Otherwise you'd be linking to nVidia specifications, not consumer reviews. The relevent OpenGL extension is GL_ARB_color_buffer_float, which was indeed implemented for G70-class hardware as of R75 drivers (actually, GL_NV_float_buffer.txt was implemented even earlier).

Yes, you can use this for offscreen framebuffer objects and pbuffers, which is all you need when float texture blending for HDR rendering, but this is then tone-mapped to the 32bit displayable framebuffer for output. It's still not possible to get more than 8bit RGB actually out of the chip. Apart from SGI (who patented float rasterisation), I've only heard of an old Matrox card claiming to do real 10bit integer RGBA output (under quite specific conditions, apparently). Even nVidia's current high-end Quadros can't do it (well, unless you count 10bit 4:2:2 YUV from the SDI connector on some models). I'd welcome any comments showing real evidence to the contrary (preferably from someone who hasn't been repeatedly modded down as a troll), but I've never seen it done.

I can see it easily with my own eyes

The PS3 can decode TrueHD into PCM

There it is right there. Yes, the PS3 player supports TrueHD, but it does notpass it over the HDMI link - it gets decoded to good ol' HDMI 1.0-standard multichannel PCM first. Read the rest of the article - a Sony rep has confirmed this. And AFAIK the PS3 still does not yet support DTS-HD; it only passes through the DTS component. Incidentally, I found it ironic that you're accusing me of trolling :-)

You don't have to convince me that the PS3 is good hardware. It certainly has the edge in CPU power, and the Blu-Ray player is a valuable addition (though it's also the primary reason Sony released late and expensive, throwing away their lead from the PS2). Its graphics are debatable though, and most unbiased people consider PS3 and Xbox 360 GPU power to be roughly equivalent. More on topic, the PS3's HDMI port is more capable than the 360's (which can't even pass multichannel PCM) - but the HDMI 1.3 output is pure marketing, nothing more. Most TV sets (even those that accept deep colour) still can't actually display it, only use it for cleaner tweaking. Certainly no plasma or consumer LCD panel that I'm aware of is capable.

You're not a gfx programmer yourself, are you? Otherwise you'd be linking to nVidia specifications, not consumer reviews. The relevent OpenGL extension is GL_ARB_color_buffer_float, which was indeed implemented for G70-class hardware as of R75 drivers (actually, GL_NV_float_buffer.txt was implemented even earlier).

Yes, you can use this for offscreen framebuffer objects and pbuffers, which is all you need when float texture blending for HDR rendering, but this is then tone-mapped to the 32bit displayable framebuffer for output. It's still not possible to get more than 8bit RGB actually out of the chip. Apart from SGI (who patented float rasterisation), I've only heard of an old Matrox card claiming to do real 10bit integer RGBA output (under quite specific conditions, apparently). Even nVidia's current high-end Quadros can't do it (well, unless you count 10bit 4:2:2 YUV from the SDI connector on some models). I'd welcome any comments showing real evidence to the contrary (preferably from someone who hasn't been repeatedly modded down as a troll), but I've never seen it done.

I can see it easily with my own eyes

The PS3 can decode TrueHD into PCM

There it is right there. Yes, the PS3 player supports TrueHD, but it does notpass it over the HDMI link - it gets decoded to good ol' HDMI 1.0-standard multichannel PCM first. Read the rest of the article - a Sony rep has confirmed this. And AFAIK the PS3 still does not yet support DTS-HD; it only passes through the DTS component. Incidentally, I found it ironic that you're accusing me of trolling :-)

You don't have to convince me that the PS3 is good hardware. It certainly has the edge in CPU power, and the Blu-Ray player is a valuable addition (though it's also the primary reason Sony released late and expensive, throwing away their lead from the PS2). Its graphics are debatable though, and most unbiased people consider PS3 and Xbox 360 GPU power to be roughly equivalent. More on topic, the PS3's HDMI port is more capable than the 360's (which can't even pass multichannel PCM) - but the HDMI 1.3 output is pure marketing, nothing more. Most TV sets (even those that accept deep colour) still can't actually display it, only use it for cleaner tweaking. Certainly no plasma or consumer LCD panel that I'm aware of is capable.

If you have a GeForce 8800 GT, you already have a 112 processor parallel computer that you can program using CUDA.

The direction that everybody is looking at for next generation, both console and eventual graphics card stuff, is a "sea of processors" model, typified by Larrabee or enhanced CUDA and things like that

man woman

I had the very same question, and this article from Nvidia turned out to be very enlightening.
Quadro vs FX -- http://www.nvidia.com/object/quadro_geforce.html

According to the article, there are some major differences between the two architectures. Where features are programmed either at the hardware layer (quadro), or at the driver layer.

here ya go: http://www.nvidia.com/object/io_1197375200475.html

of course, the article doesn't explicitly say 2 cards are needed.. It's referring to the fact that onboard graphics were previously replaced by PCI/AGP/PCI-E graphics cards, and that the two systems wouldn't work together at all.

I can't say for sure (i haven't read up on it properly yet), but what AMD appear to have done is made it possible to have an SLI-style system that is capable of using onboard graphics and PCI-E graphics at the same time.

Not so fast.

Nforce 3 (the AMD equivalent to Intel Grantsdale a.k.a. 915) was also labeled "Vista Capable".

So AMD was probably an equal beneficiary of the marketing "favor".

You'll have everything you need. You may want to consider PC-BSD, a.. friendlier edition of FreeBSD. It uses KDE by default, and as you're a Fluxbox user, you'll know how to swap that out as needed.

The only gotcha is that nVidia's binary drivers are just as finicky as in Linux, and you're SOL if you want to use the amd64 version of FreeBSD, unless I'm out of touch. You can find their binary driver here.

The GeForce 8 series is perfectly capable of being used to accelerate physics calculations and pass the results back to the CPU. The PCI-Express bus is bidirectional, and the scheme for getting results back from calculations done on the GPU is essentially this: encode your inputs as Texture1, set the render target of your program to Texture2, use a shader to calculate the results and "render" them (drawing them to Texture2), and pull Texture2 back to system memory or reuse it for further calculations. I suggest you look into nVidia's CUDA and, for more general information, check out GPGPU.org.

Note that with CUDA you can avoid many of the headaches induced by "normal" GPGPU programming (ie, the OpenGL/DirectX coding involved in the scheme described above), but you limit yourself solely to the GeForce 8 series, whereas GPGPU programming using GL/DX offers access to a wider range of hardware.

NVidia already sell their Tesla GPGPU which is GPU-derived general computing hardware.

http://www.nvidia.com/object/cuda_learn_products.html CUDA can run on some pretty cheap cards now.

Guess what guys? We've run out of GHz (mainly a power/heat problem). Start writing parallel programs.

Here is what the article says:

To be fair, though, it is not Intel's hardware that is at fault here, but today's software. If a program only uses four of the eight processor cores, then the Skulltrail system is noticeably slower than a single-socket quad-core computer. Since there are practically no current games or desktop applications around that can utilize more than four cores (if that many), the Skulltrail system does not offer any benefit here.

Read The Landscape of Parallel Computing Research: A View From Berkeley which has the description of why, this time, there is no getting around parallel programming.

Also examine NVIDIA's CUDA platform, which scales from a handful of processors on your PC's NVIDIA chip to the 128 processor NVIDIA Tesla card. Scalable parallel processing is the future.

Guess what guys? We've run out of GHz (mainly a power/heat problem). Start writing parallel programs.

Here is what the article says:

To be fair, though, it is not Intel's hardware that is at fault here, but today's software. If a program only uses four of the eight processor cores, then the Skulltrail system is noticeably slower than a single-socket quad-core computer. Since there are practically no current games or desktop applications around that can utilize more than four cores (if that many), the Skulltrail system does not offer any benefit here.

Read The Landscape of Parallel Computing Research: A View From Berkeley which has the description of why, this time, there is no getting around parallel programming.

Also examine NVIDIA's CUDA platform, which scales from a handful of processors on your PC's NVIDIA chip to the 128 processor NVIDIA Tesla card. Scalable parallel processing is the future.

http://forums.nvidia.com/index.php?showtopic=41802

You may think that your 3d modeling and prototyping is professional work - and I'm sure it is.

However, you should be thinking of people using CATIA to build an entire car or even more exotic pieces of software for building entire airplanes. We're not talking the piddly few million of polies that the average Disney/Pixar movie ponders about in Maya/etc., even though those would benefit as well - we're talking a dew hundered million polies. Now we're talking 'pro'. Now we're talking the kind of people who used to buy SGI workstations at a couple $10k a piece, then switched to 'generic' workstations but fitted them with E&S (Evans & Sutherland) cards that were so big (similar in design to dual-GPU cards people are messing with now) they had to keep the casings off their machines or the things wouldn't even fit, and who are currently salivating at the nVidia QuadroPlex solutions in both desktop and rackmount form ( http://www.nvidia.com/page/quadroplex.html ) before crying as even they think that's just a might bit too pricy and go back to the suped-up PNY QuadroFX offerings ( http://www2.pny.com/category_buymulti.aspx?Category_ID=329 ).

Consumers, prosumers and small business need not apply. As you do say, it's not worth the extra money (and it -is- a good chunk of extra money) for those groups.

I'm surprised no one has mentioned CUDA yet, which is Nvidia's existing entry into the world of general purpose GPU computing. So far their target market is mostly dedicated calculations with limited interoperability with OpenGL/DirectX, but I expect we'll see future cards that can partition their compute resources between multiple tasks, like rendering and physics. Hopefully, porting over the PhysX SDK will help grow the GPGPU toolset, and make it easier to use.

(CUDA already transforms the 8800 GTX into quite an impressive array processor. With 128 floating point units and 768 MB of fast, fast memory, this card is chewing up the data-parallel compute tasks I'm throwing at it.)