Slashdot Mirror

OK after reading these pages...
http://www.nvidia.com/object/gpus_supporting_adobeflash.html
http://www.adobe.com/products/flashplayer/systemreqs/

it seems Flash supports PureVideo VP2 onwards, and UVD2 onwards for ATI. Isn't as good as the desktop is but MUCH better than 4 lousy GPUs.

I still have my reservations about Adobe's hardware decoder though (it doesn't compare up to the other h/w decoders).

How is its performance with Flash videos?

Most of the on-demand video services use it, and for Flash, CPU power is very important.

There is GPU hardware acceleration for the NVidia ION in the latest 10.1 version of Flash. You get flawless low-CPU-usage decodes of 1080p Flash video. Also, I recently saw a video of a dual core Atom with NVidia ION that was decoding two simultaneous 1080p streams and only using about 60% CPU. The Apple TV is limited to a single 720P stream for output. The little box I bought has a couple orders of magnitude more CPU and GPU / Video Decode Power than the Apple TV.

The ION is fast enough for all 3D casual gaming and even gets some decent frame rates on hard-core games if you don't have quality set on ultra-high and you are running at 720P rather than 1080P resolution. It's a decent 3D chip that kicks the crap out of all the other "integrated" GPU's out there right now.

Plus I put in a cheap 80 GB Intel SSD I picked up at NewEgg for $125 so it has lots of storage and very fast boot times (under 15 seconds including BIOS POST) -- although it would work fine with any cheap notebook drive as well.

So you prefer playing games at 30 fps[1] in 1280x720 resolution[1], without stereo vision, which you won't even be able to emulate (for nostalgy reasons) 10 years from now?

Congratulations.

[1] - Theoretically better rates are possible, but hardly any game developer targets higher figures (sometimes - on PS3 particularly - they are even Ok with upscaling sub-HD resolutions). It's unreasonable to push hardware to its limits because optimizing content creation pipeline pays a lot more than optimizing the performance. On PC it's the same developer-wise, but you can always add more power to your gaming machine.

DISCLAIMER: I'm a console game developer. But I don't think the above will hurt the business I'm in because there's a lot of people who don't even understand what does it mean "maximum 30 frames per second".

800 cores.... that's like 134 CPU's

Or two GPUs.

If it can run Crysis it can simulate galaxies.

There was also Tegra-based Toshiba AC100 on display, with a lot more nerd-appeal IMHO.

I might buy one of these, if you can install a proper distro instead of this silly little smartphone OS. Tegra as such is relatively open, and there are instructions on getting Ubuntu and Gentoo on the devkit, but the netbook may be a different matter.

Here "relatively open" means that you need to run Nvidia's own kernel branch based on 2.6.29, and the video drivers seem to be closed blobs. I think I'll get a GNU/RMS-approved Lemote Yeeloong instead.

http://tegradeveloper.nvidia.com/tegra/forums/tegra-forums/linux-development

Really? I've had the capability to use 3D in many games since the late 1990's with the Elsa Revelator brand of Riva TNT cards, that supported hard-wired LCD Shutter glasses, meant to be used with CRT displays and refresh rates of 100Hz+. NVidia has had 3D support for a long, LONG time now (Check out the "Supported games" list). That they're now posting guidelines for it, and helping developers out if they request it (Their TWIMTBP program) doesn't negate that.

Fooey, I'll need a new card for vdpau.

Sample of one? http://hplies.com/ http://h10025.www1.hp.com/ewfrf/wc/document?lc=en&dlc=en&cc=us&docname=c01087277 http://support.apple.com/kb/ts2377 http://www.nvidia.com/object/io_1215037160521.html http://en.community.dell.com/dell-blogs/Direct2Dell/b/direct2dell/archive/2008/09/12/nvidia-gpu-update-limited-warranty-enhancement-details.aspx

I recommend reading the programmer's guide to a modern graphics architecture; Caching is essential to them.

Modern GPU architectures face the same clock speed/bus speed disparity and memory latency problems as CPUs and have taken their response much farther. They have several thousand registers per core and an L1 size & speed cache per processor group. Cache misses carry a typical penalty of several hundred cycles.

do you really need to boot the machine with a livecd? by intel/realtek wireless, buy HP/Epson printers/scanners, by just about any nvidia/intel/amd videocard you want(GMA500 is to be avoided). Other than that, i'm not aware of a lot of main stream hardware that doesn't work. Have any examples that i didn't already cover?

Printer/scanner compatibility can be found at http://www.linuxprinting.org/
AMD compatability can be found at https://a248.e.akamai.net/f/674/9206/0/www2.ati.com/drivers/linux/catalyst_107_linux.pdf
nvidia compatibility found at http://www.nvidia.com/object/linux-display-amd64-256.44-driver.html
Intel graphics compatibility found at http://intellinuxgraphics.org/documentation.html
NDISWrapper compatibility found at http://sourceforge.net/apps/mediawiki/ndiswrapper/index.php?title=Main_Page

You should probably mention that to the nVidia, ATI, and Linux folks so they can stop working on it.

http://www.linuxquestions.org/questions/programming-9/opencl-on-linux-693926/

http://developer.nvidia.com/object/opencl-download.html

http://www.theregister.co.uk/2010/05/03/ati_stream_sdk_2_dot_1/

The official site: http://www.nvidia.com/object/tesla_computing_solutions.html And some price info (unofficial of course): http://nextbigfuture.com/2007/06/nvidia-tesla-supercomputer-for-1500-to.html

And it was not stupid, just old and bad technology before. He is probably thinking the old red-blue glasses. 3D viewing has improved a lot in the recent years, and for example gives a totally new feeling with games. If you have tried Left 4 Dead with NVidia's 3D Vision Kit and a 120hz 3D capable monitor, you know what I'm talking about (it is a lot scarier too).

It really gives a completely new feeling, when done right, and remember this is all along the path for technology that can render the environment completely and realistically around us, with a complete feeling of "being there".

Nvidia tegra already has all kernel component open sourced. Qualcom does not want to loose open source fans. http://nv-tegra.nvidia.com/gitweb/

GPUs have extremely fast RAM connected to them, much faster than even system RAM

I'd like to see a citation for that little bit of trivia

Ok, so my Geforce GTX480 has GDDR5 ( http://www.nvidia.com/object/product_geforce_gtx_480_us.html ) which is based on DDR3 ( http://en.wikipedia.org/wiki/GDDR5 )

My memory bandwidth on the GTX480 is 177 GB/sec. The fastest DDR3 module is PC3-17000 ( http://en.wikipedia.org/wiki/DDR3_SDRAM ) which gives approx 17000 MB/s which is approx 17GB/sec. So my graphics ram is basically 10x faster than system ram as it should be.

They've been around for a while. LINK

you mean the tegra 250?

http://www.nvidia.com/object/tegra_250.html

> It would mean that development cycles slow down, algorithmics finally win over brute force and that software quality would have a chance to improve (after going downhill for a long time). Um, nope. Companies will simply sell bigger boxes to run their bloated code. > GPUs as CPUs? Ridiculous! Practically nobody can program them http://www.nvidia.com/object/cuda_apps_flash_new.html > and very few problems benefit from them. Media encoding/transcoding. Scientific code, minimum spanning trees can also be done a a GPU. If you mean by a 'few problems' that it doesn't run Word/Office/Java etc, then yes. Otherwise if it's a case that the algorithmics (sic) can be done in a data parallel fashion, then the problem might be able to done on a GPU.

Frankly, even equally worrisome is that Matlab doesn't appear to take advantage of GPGPU yet. The concept has been around for over half a decade, and I'd have expected the MAtrix LABoratory to jump on the bandwagon quicker than most. It's a game changer in their core competency, after all.

I guess it depends on the exact question you're asking. A google search for "matlab gpgpu" shows that there are lots of ways to take advantage of GPGPU (NVidia's CUDA specifically) from within Matlab.
MATLAB plug-in for CUDA
AccelerEyes
GPUmat

However AFAIK there's no plan for native support of GPGPU within Matlab. It's kind of ridiculous that there's not considering the 10x speedup frequently reported by using the above tools.

I was thinking Soekris Engineering's vpn accelerator card would help, but it appears to only be able to do 250 Mbps. (You wanted 1 gigabit/s, right?)

That card is really old too. I first read about it probably 10 years ago. I don't think it has changed in that time... I wonder if someone makes a faster accelerator? Then again, what about the GPU? Has anyone tried encryption with GPUs before? They've done other supercomputing tasks. A quick search says they have.

Well the GPGPU landscape is still forming in significant ways, and the very architectures of the software/hardware are still very much evolving. What this means is that in most cases you can certainly write code that "in effect" accomplishes the same overall computational goals with an older card like many of the ones in the GT2xx series, but you'll be restricted architecturally in the ways in which you have to design the algorithms themselves since some constructs / capabilities that are present on the newer cards are either absent or are so inefficient that they're useless on the older cards. NVIDIA calls their architectural divisions "compute capability" classes and they revision them like 1.0, 1.1, ... 2.0, etc. depending on
the generation of card. The newer cards have things like unified memory spaces to better support pointer use, the ability to run multiple different compute kernels at the same time, better more efficient ability to share data across compute units / threads / kernels on the GPU, better double precision, and so on.
You should read the following documents to get an idea of the capabilities Fermi cards have that GT2xx cards do not.
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_FermiTuningGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/docs/NVIDIA_FermiCompatibilityGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_CUDA_ProgrammingGuide.pdf

In many cases the advantages are compelling, and can even mean the difference between algorithms that will run well on a GPU and ones that just won't, i.e. better caches, data sharing, kernel parallelism, better debugging, etc.

So if you want to have the most modern architecture available to program on, I'd say wait 3 months for a lower cost fermi card and use that, but realize that you may be forced to develop multiple versions of your algorithms if you want the optimum fermi performance / capabilities for one version, but you will need a differently coded version if you want to be able to run a similar program on an older GT2xx GPU for instance.

If you're content with very basic GPGPU capabilities and learning to code for them, you could get started writing OpenCL code that can run on your main CPU even without an advanced GPU, or you might be able to use the device emulator capabilities for code testing / debugging that NVIDIA may still support for limited experimentation.

If you really want to use the architecture to its fullest potential, though, you'll probably end up writing CUDA code and not OpenCL since the former gives you a lot "closer to the metal" capabilities than OpenCL does.

For the forseeable future AMD's 58xx cards will actually handily outperform NVIDIA's fermis in double precision math heavy kernels due to the intentionally crippled DP performance in the fermis, though AMD has very immature OpenCL support compared to NVIDIA, and CUDA is better than OpenCL in terms of raw programming efficiency capability anyway.

I'd wait until the mainstream Fermis are out in 2-3 months, then use that, and in the mean while learn OpenCL now based on the CPU runtimes and study up on the Cuda docs so you have some clue about how to use it when you get the card.

Well the GPGPU landscape is still forming in significant ways, and the very architectures of the software/hardware are still very much evolving. What this means is that in most cases you can certainly write code that "in effect" accomplishes the same overall computational goals with an older card like many of the ones in the GT2xx series, but you'll be restricted architecturally in the ways in which you have to design the algorithms themselves since some constructs / capabilities that are present on the newer cards are either absent or are so inefficient that they're useless on the older cards. NVIDIA calls their architectural divisions "compute capability" classes and they revision them like 1.0, 1.1, ... 2.0, etc. depending on
the generation of card. The newer cards have things like unified memory spaces to better support pointer use, the ability to run multiple different compute kernels at the same time, better more efficient ability to share data across compute units / threads / kernels on the GPU, better double precision, and so on.
You should read the following documents to get an idea of the capabilities Fermi cards have that GT2xx cards do not.
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_FermiTuningGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/docs/NVIDIA_FermiCompatibilityGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_CUDA_ProgrammingGuide.pdf

In many cases the advantages are compelling, and can even mean the difference between algorithms that will run well on a GPU and ones that just won't, i.e. better caches, data sharing, kernel parallelism, better debugging, etc.

So if you want to have the most modern architecture available to program on, I'd say wait 3 months for a lower cost fermi card and use that, but realize that you may be forced to develop multiple versions of your algorithms if you want the optimum fermi performance / capabilities for one version, but you will need a differently coded version if you want to be able to run a similar program on an older GT2xx GPU for instance.

If you're content with very basic GPGPU capabilities and learning to code for them, you could get started writing OpenCL code that can run on your main CPU even without an advanced GPU, or you might be able to use the device emulator capabilities for code testing / debugging that NVIDIA may still support for limited experimentation.

If you really want to use the architecture to its fullest potential, though, you'll probably end up writing CUDA code and not OpenCL since the former gives you a lot "closer to the metal" capabilities than OpenCL does.

For the forseeable future AMD's 58xx cards will actually handily outperform NVIDIA's fermis in double precision math heavy kernels due to the intentionally crippled DP performance in the fermis, though AMD has very immature OpenCL support compared to NVIDIA, and CUDA is better than OpenCL in terms of raw programming efficiency capability anyway.

I'd wait until the mainstream Fermis are out in 2-3 months, then use that, and in the mean while learn OpenCL now based on the CPU runtimes and study up on the Cuda docs so you have some clue about how to use it when you get the card.

Well the GPGPU landscape is still forming in significant ways, and the very architectures of the software/hardware are still very much evolving. What this means is that in most cases you can certainly write code that "in effect" accomplishes the same overall computational goals with an older card like many of the ones in the GT2xx series, but you'll be restricted architecturally in the ways in which you have to design the algorithms themselves since some constructs / capabilities that are present on the newer cards are either absent or are so inefficient that they're useless on the older cards. NVIDIA calls their architectural divisions "compute capability" classes and they revision them like 1.0, 1.1, ... 2.0, etc. depending on
the generation of card. The newer cards have things like unified memory spaces to better support pointer use, the ability to run multiple different compute kernels at the same time, better more efficient ability to share data across compute units / threads / kernels on the GPU, better double precision, and so on.
You should read the following documents to get an idea of the capabilities Fermi cards have that GT2xx cards do not.
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_FermiTuningGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/docs/NVIDIA_FermiCompatibilityGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_CUDA_ProgrammingGuide.pdf

In many cases the advantages are compelling, and can even mean the difference between algorithms that will run well on a GPU and ones that just won't, i.e. better caches, data sharing, kernel parallelism, better debugging, etc.

So if you want to have the most modern architecture available to program on, I'd say wait 3 months for a lower cost fermi card and use that, but realize that you may be forced to develop multiple versions of your algorithms if you want the optimum fermi performance / capabilities for one version, but you will need a differently coded version if you want to be able to run a similar program on an older GT2xx GPU for instance.

If you're content with very basic GPGPU capabilities and learning to code for them, you could get started writing OpenCL code that can run on your main CPU even without an advanced GPU, or you might be able to use the device emulator capabilities for code testing / debugging that NVIDIA may still support for limited experimentation.

If you really want to use the architecture to its fullest potential, though, you'll probably end up writing CUDA code and not OpenCL since the former gives you a lot "closer to the metal" capabilities than OpenCL does.

For the forseeable future AMD's 58xx cards will actually handily outperform NVIDIA's fermis in double precision math heavy kernels due to the intentionally crippled DP performance in the fermis, though AMD has very immature OpenCL support compared to NVIDIA, and CUDA is better than OpenCL in terms of raw programming efficiency capability anyway.

I'd wait until the mainstream Fermis are out in 2-3 months, then use that, and in the mean while learn OpenCL now based on the CPU runtimes and study up on the Cuda docs so you have some clue about how to use it when you get the card.

Oh, for the love of Stallman...

wget -c http://us.download.nvidia.com/XFree86/Linux-x86/195.36.15/NVIDIA-Linux-x86-195.36.15-pkg1.run

Or, if you're running Ubuntu, just sudo apt-get install envyng-gtk or, if you're running Kubuntu, sudo apt-get install envyng-qt. You still don't need a web browser to install a driver - it's just a heck of a lot easier to find it that way.

Never kept more than a few computers that long. There is a certain energy savings if I upgrade my equipment every 3-4 years that has kept me ahead of any legacy hardware cutoff. Also the oldest video card I have an x300 works fine in Ubuntu, even in 3D and it is 5+ years old. Both NVidia and ATI have a large enough interest in Linux devices/workstations/platforms they would never stop support now.

What other architecture are you talking about, BSD and Solaris? Both supported by NV but ATI only does Linux.

Freedom for who, why should a company that spent 100's of millions of dollars in RD give away the keys to kingdom?

http://glew.sourceforge.net/ GLEW also comes with the NVIDIA SDK: http://developer.nvidia.com/object/sdk_home.html NVIDIA SDK is also a good place to start with OpenGL.

In both cases the user has to go to NVIDIA / ATI / Intel website and download one, this requires navegating a bunch of questions about what exact model of graphics card is in the machine, which Joe Sixpack isn't going to know

He may not need to know:

Option 1: Manually find drivers for my NVIDIA products.
Option 2: Automatically find drivers for my NVIDIA products. Download Drivers

Direct3D drivers tend to be better because a certain company oils the development process.

The hardware manufacturers talk to Microsoft. Microsoft talks to the hardware manufacturers. 93% of the market and a platform more open than Apple's and less fragmented than Linux gives it a very strong voice.

They're still a minority. I've only seen polarized glasses now that these 3D movies came again.

But hell, when it was a new thing I even played Left4Dead in 3D with NVIDIA's 3D Vision. Expect that I went the cheap way and just used red-cyan (or maybe some else) glasses as the feature is build-in with normal nvidia drivers too. The experience is definitely more cool and works really well with some games. Those incoming hordes of zombies really do scare you. If I had more cash I would probably buy that set and some supported monitor, as it does make a difference in games.

it sure does, if you have an 8xxx+ card and flash 10.1 beta... i have a 7600gt agp, 195.x nvidia beta drivers, and the vdpau libs installed, for some reason i can now play youtubeHD without stutter and the flashplayer load is very much spread even over my four cores.

http://www.nvidia.com/object/adobe_flashplayer_plus_nvidia.html

The second test seems to forget that Flash added GPU acceleration in Windows, which dramatically drops CPU usage. It's not even small amount, it's 60%->12% with YouTube 720p video and most likely even more with 1080p. They've been working a lot with NVIDIA on it, which means more bad news for HTML5. I also installed those new NVIDIA drivers and newest Flash beta and full screen video is considerably smoother.

And where's Opera in this test? They added HTML5 support in 10.5 final too and their whole drawing engine will be hardware accelerated, with websites also. Their canvas implementation is also faster than with any other browser.

This issue is related to automatic fan control not working due to improper registry keys, and so GPU's that run warm (9800 series for instance) can quickly overheat and potentially suffer damage. I'm having no issues with mine, but I set fan profiles manually as I'm using a machine that has a very hot MCH & fb-dimms (2008 Xeon) and don't want the gpu contributing more. However for anyone interested (and using a GT200 or at least G80/G92 on up) here's the fix: http://forums.nvidia.com/index.php?showtopic=161767

nVidia beat you to it nearly ten years ago, Intel... the GeForce 3 was used to create "Final Fantasy: The Spirits Within" big-screen movie.

actually you can get copies of every MS product right back to DOS 3 via technet.

Technet is not a retail channel for typical consumers to get a single product. Also, the older products aren't supported and generally don't work on the new hardware around today.

because some FOSS project is FAR FAR more likely to stop producing updates and go offline (because they got a life/job/girlfriend) then a company like MS or nvidia which has actual funding

NVIDIA has already shown they are willing to drop driver support for their products when they aren't interested anymore. And it's not just about the risk of if they will stop support, you also need to factor in the damage done - we don't have the option to fix the proprietary stuff ourselves even if we wanted to, but we could fix the abandoned FOSS stuff if we considered it worthwhile.

so you'll need use a better example

So you'll need [to] use a better excuse.

Note: You can only install this driver if you have the installed the latest GeForce Graphics drivers v196.21.

I guess they started focusing on releasing new drivers since the 3D craze started up again. It also only seems to be supported on 8 series and above graphics cards, whereas they had support for the LCD shutter glasses (including brands other than NVIDIA) with older video cards a while ago.

See this thread as an example of NVIDIA dropping support for the older hardware.

Sorry to hear your fingers are broken and cant use a search engine.

Here's the nvidia link stating what hardware and software is OpenGl 3.x capable. it's quite extensive. http://developer.nvidia.com/object/opengl_3_driver.html

FYI, my laptop has a Quadro FX 1700M.

A 310m is a mainstream graphics card for notebooks, not "high end".

Translation: this a mainstream laptop.

That being said I've yet to see a game using OpenGL that can render anything as well/as quickly as DX10.

That's cute and all but It's just a shame that it's blatantly and fundamentally wrong. Taken from TFA:

It's common knowledge that OpenGL has faster draw calls than DirectX (see NVIDIA presentations like this one if you don't want to take my word for it), and it has first access to new GPU features via vendor extensions.

The link to NVidia's presentation is here and the wikipedia article on the subject also states that OpenGL is faster than DirectX.

So, where exactly do you base your "OMG OpenGL is tHe sLowZz" claims?

let's just take Linux for example, drivers that work in one distro don't always work in another

If done right, they absolutely do. If you're talking about the 3D market, nVidia has one download per OS per CPU architecture. So the same 64-bit (or 32-bit) driver will work across all distros.

Having a stable/typical setup helps, as you can focus in more on your product than worrying about whether or not it will work on all these different platforms/drivers/whatever.

Yeah, it's great, until you want to do something that stable/typical setup won't support.

Sure, you could have an "open standard," but someone is controlling that, too.

That "someone" is a group of people from many companies, not just one. And while it's "controlled", it is much more freely licensed -- contrast to DirectX. How likely is Microsoft to license the DirectX APIs for Linux or OS X -- or worse yet, PS3 or Wii? How much would they charge for it?

The article notes about motion blurring, and links to NVidia's page about it's technology. The last figure shows a terrain with full-screen motion blur effect, which in my opinion is pretty important in games to create that feeling of speed. People usually object against this and bloom effects and just want a sharp picture, but maybe some games have taken it too far. It's important none the less, even if it's not all sharp picture, because your eye picture isn't all that sharp either and you experience the same blur.

The article notes about motion blurring, and links to NVidia's page about it's technology. The last figure shows a terrain with full-screen motion blur effect, which in my opinion is pretty important in games to create that feeling of speed. People usually object against this and bloom effects and just want a sharp picture, but maybe some games have taken it too far. It's important none the less, even if it's not all sharp picture, because your eye picture isn't all that sharp either and you experience the same blur.

Nvidia already bundles their graphics acceleration with their own processor. The Tegra processor has plenty of power for Internet Tablets, with way lower power and overall system cost. I think it would be a terrible mistake for Nvidia to enter the i386 compatible market. They should instead help us build a future without all that baggage.

What's with all the opposite caps? From their website, it looks like the proper casing for nVidia is either nVIDIA or NVIDIA.

I had an especially bad time with games using the Source Engine like Left 4 Dead. It seems they used some nasty programming in there. I had to disable multi-threading, sound, and a number of other things to get it to crash less, rather than not crash, in Vista 32-bit. If I played Dawn of War II with shadows enabled (the default) I would also consistently get display driver restarts.

According to this, it's wireless with a 6' USB recharge cable.

W

Combine that with Pixel Qi multi-touch displays and you've got a rock'in platform. This display does color video with the backlight on, but with it off it does E-Ink-like low-power black and white as good as any e-book reader. Power it with an Nvidia Tegra processor, and run Ubuntu Netbook Remix, and you've got one freaking awesome Internet tablet, with e-book reader being the killer app. Add a dynamically warping virtual multi-touch keyboard, and you've got a killer device. I think I'll need one of these babies for every member of my family.

Multiple graphics cards? That's a very exotic multiple-display configuration these days now that dual-head graphics cards are the standard. So if you act like your experience with your exotic setup is typical, you can expect shocked reactions from the 99% of multi-display users currently using a single dual-head graphics card with no problems or setup difficulty.

The most common problem is having to use nvidia's setup tool instead of the standard 'display preferences' control panel because nvidia is taking forever to implement xrandr 1.3. But on the upside, they have their own (proprietary) solution to support hardware acceleration with a Xinerama setup (with similar cards).

One word (well, acronym): UDK

It's essentially Epic releasing all their hard work for FREE for non-commercial use. That at least puts them on par with Valve and their free "Source SDK".

The Nvidia forums are rife with people who are saying that the 180 drivers aren't stable and that, in particular, they have problems with Fallout 3. This thread has more info.

If you read TFA you'll see that Crysis pretty much requires one of these very high end cards to run at high resolutions and better than 30 frames per second. That's not really what interests me though. The fact that this card has nearly a teraflop of double precision floating point processing power is the most impressive for GPGPU applications. GPGPU applications typically don't care about the single precision number that the card companies like to point out. To figure out what the integer capabilities are you really have to look at the specific problem.

Too bad AMD isn't putting much focus on GPGPU and has only released a beta SDK that still is more difficult to use than CUDA. Nvidia's Tesla only has 78 gigaflops double precision, but it has 4GB of RAM on it. If AMD put the focus on that a bit, they could crush Nvidia.

No you still don't get it. Please tell us how you can infringe on this:

"2.1.2 Linux/FreeBSD/OpenSolaris Exception. Notwithstanding the foregoing terms of Section 2.1.1, SOFTWARE designed exclusively for use on the Linux or FreeBSD operating systems, or other operating systems derived from the source code to these operating systems, may be copied and redistributed, provided that the binary files thereof are not modified in any way (except for unzipping of compressed files)."[1]

That is the license in question. There is no infringement, they are acting with in the license. There is a reason that the drivers are broken in to binary blobs with a stable abi, the the open source shim the connects that stable abi to the current kernel api.

[1]: http://www.nvidia.com/object/nv_swlicense.html