Hi everyone,

I've been lucky to attend NVIDIA's Nvision 08 event in California the last couple of days and wanted to report what I saw to the community. Nvision 08 is NVIDIA's first worldwide conference bringing together the many dynamic uses of their technology under the theme "visual computing."

It was organized in meeting rooms, theaters and three large halls: one hall for company exhibits for vendors involved in motherboards, graphics, automotive uses (huge!) and high performance (CUDA) computing; a very large hall for many hundreds of computers supplied by vendors and brought by participants involved in gaming of various kinds plus a Guiness world record attempt for the largest LAN party; and last, but not least, a huge hall for competitive gaming leagues with blow-by-blow commentary and reportage on screens 40 feet high. In addition to all that a variety of adjacent meeting rooms hosted professional conferences on CUDA/High Performance Computing, Automotive, Professional and Enterprise computing tracks. Ah, yes, and there were a variety of performance centers, HD theaters and other venues showing the incredible CG uses of NVIDIA that now power the film industry, many in 3D no less.

Gaming uses for NVIDIA are huge. We all know that, but to internalize how huge they have become so rapidly is not easy to internalize. One indicator is that the gaming industry is now tens of billions of dollars bigger than the world film industry. If anything, the world film industry, from Hollywood to Bollywood is seeming more and more destined to become an appendage of gaming given the extensive use of CG. Of the top ten revenue-grossing movies in the past year, nine could not have been created without CG. Actors speaking at events discussed how it has become routine for their synthetic doubles to be inserted into scenes and it is clear that fully synthetic actors will soon be used as non-gimmick replacements for live actors. But for all the 3D displays and expected evolution of striking graphics, the really amazing stuff seemed to be coming out of the massively parallel computing use of GPUs.

What surprised me and I think what surprised many of the attendees, a sort of collective realization that came upon us, was on the one hand the incredible surge in massively parallel architectures and on the other hand what they have now made possible. Many of us have been working on our own projects this past year and while it was obvious that many other things were going on it was still a big impact to see it come together in one place. Parallel computing is already huge. There are now tens of thousands of developers working with CUDA and many hundreds of truly significant applications. The amazing thing is how much of what is being done is truly transformative.

Cars are employing CUDA calculations within embedded NVIDIA GPUs to do real-time, highly sophisticated computer vision so that multiple cameras within the car can "see" other cars, obstacles, the road and terrain and make decisions about safety. It is not just simple as the car driving itself through traffic. It is things like the car knowing when a crash is unavoidable and being able to make physics decisions to minimize the damage of that crash. This can range from things like deploying air bags moments before impact to allow a softer, less dangerous deployment, to the car knowing that a pedestrian is going to be hit and the car making subtle changes in variable, electromotive bumper and hood component positions to dramatically lessen the harm to the pedestrian.

In consumer applications, simple cameras, like the digital cameras already routinely connected to computers, can provide imagery that can be analyzed using the speed of CUDA in real time to detect objects, including people and the motions of those people in view, to enable control of computers using gestures. Like the gestures and interfaces in the iphone? You can have that on your monitor as well if you like to enable new GUIs or to simply enable Wii-like games without any need to hold hardware.

It turns out we are getting remarkably close to simulating natural brain function by simulating the operation of many neurons in real time. We are many years away from doing humans, but simpler organisms are no longer a problem using relatively inexpensive (four GPUs, like four 280GTX cards... a little over a 1000 stream processors) hardware. Even if we are not talking "intelligence" or self-awareness or other sentience, a remarkable degree of autonomous function is now possible. There will only be more in the (near) future.

Anyway, while simulating the very largest systems is still a challenge requiring racks of NVIDIA GPU cards (albeit room sized racks and no longer football stadium sized racks), for an extraordinary range of applications requiring machine vision, machine reasoning, 3D everything, computation that used to be prohibitively expensive, can be accomplished with a single GPGPU card or chip and all that at such absurdly low costs that even the most trivial application, like an automobile back-up camera, can use it.

The backup camera is a good example of a nearly free CUDA application. People don't like the "fish-eye" distortion usually seen in backup camera displays. No problem. CUDA GPUs of the simplest sort can instantaneously recompute the warping to, in effect, re-project the fisheye distortion into a flat display.

Other examples include the new field of (I'm not making this up) apparel cloth physics. It turns out that while it is fairly simple in 3D packages to show some cloth draped over an object it is remarkably difficult to compute exactly how a real piece of cloth made up of given material of given thickness and wear will in fact appear in real life when cut into panels, stitched together to make clothing and worn by a human. It's a task that requires a substantial supercomputer, or a handful of NVIDIA CUDA-enabled GPU cards, to compute in real time. It is a critical task that now is done with CUDA and is spawning a new generation of computer-assisted garment industries, from designing and laying out and prepping new clothing designs for semi-automated manufacturing, to rendering those clothes in printed and online advertising or even virtual garment shopping.

On the high end there are many big applications that merit deploying rooms full of GPUs to do computations in seconds that formerly took days on supercomputers. These will increase the rate at which we are able to discover fundamental knowledge about biology, how our vision and thought work, how proteins fold, how the most complex processes of life proceed. Progress can now be made in minutes and days, with tens of thousands of simulations per day when just a year ago it would be just a simulation per week in some cases for a scientist to analyze.

I attended sessions on use of CUDA in biology, astrophysics, medicine, manufacturing, gaming and many other disciplines. What stood out in the reports and interactive discussions with folks who have been using CUDA the longest is the awareness among us all that this is just the very beginning. Looking at the inexorable advances in performance and lowered cost that we all know are coming, it is dawning on us all how incredibly transformative the emergence of truly free, literally zero cost, computation at effectively limitless rates will be on our society.

As David Kirk, NVIDIA's chief scientist, remarked in a very insightful talk, "faster is not just faster" meaning that up to some level, say 2 or 3 times faster, nothing changes, but when you get to 10 or 20 times faster things start changing as people become willing to adapt what they are doing and at 100 times faster new things become possible and people completely transform how they go about what they are doing. We can see how it will be 500 or 1000 times faster very soon now with prices for such computation going to near zero in the not too distant future.

I'm not a futurist and won't pretend to know where this is going. But that the technology is accelerating at a rate that is outstripping the imaginations of even those who had taken supercomputers (and I mean the really big iron of the last couple of years) for granted. That such power soon could be so cheap it could be inside every coffee pot, that I feel, as I have rarely felt anything so certainly before, will be true. It will be transformative in dazzling and unexpected ways and will change our world society in fundamental ways.

Anyway, in our very small part of this very big future it is clear the Manifold choice to take advantage of this new technology was the right one. Lucky for us, we got on the right track over a year ago and have built up a lot of expertise that we can utilize in new versions of Manifold. Like everyone else working with this technology, the more you do it the more you realize you are just scratching the surface and you have a wealth of ideas with which to go further.

By the way, on a personal note it was a wonderfully bizarre mix to have rooms full of maximum mathematics astrophysicists and the world's massively parallel supercomputing elite getting together just across the hall from a stadium full of gaming enthusiasts with hotrodded machines howling away at blood sport shootouts on screens two stories high. It was just wild and astonishing and perfect. All the gray beards and CUDA guys were thrilled and totally happy that the wildness and entertainment of gaming helps power getting 100 million CUDA-enabled GPUs out there so we can enjoy killer economy of scale to put a supercomputing BFG on our desktop. Cool! Now stand back as we frag that convolution matrix!

NVIDIA hopes to have full content for Nvision 08 online in the next few weeks.

Regards to all!

Cool. I wondered whether or not someone from Manifold would be going along to Nvision 08.

NVIDIA was great, very supportive. They used Manifold as an example in the enterprise presentations and asked us to participate in some of the round table and breakout sessions where CUDA users talked about their experiences and shared what they learned.

The conference had outstanding organization. NVIDIA did a super job for everyone. The free food was great, the facilities incredible (more high-end Christie projectors than Siggraph, it seemed...), free high speed WiFi and free Internet laptops everywhere and genuinely helpful "info" people in the corridors ready to help find meeting rooms, etc. Well done, NVIDIA!

How many other software developers in the geographic information field (in its broadest sense) were present or presented?

I was at an object-oriented image analysis conference a few weeks ago where Definiens Developer was giving a workshop, and none of their people had ever heard of GPUs or CUDA. Weird.

How many others from GIS? No other companies as far as I know. Autodesk was there for Nvision 08 for displays and conventional graphics but not CUDA. The closest thing to GIS were specialized seismic imaging folks, who don't really think of themselves as GIS.

The conventional image processing industry (driven by digital imaging, industrial inspection processes, CG, etc.) works with many functions within their programs that are analogous to some of what is done in remote sensing applications like ERDAS. Also, it's possible to rig up MatLab (which now has some CUDA plugins) do do work in remote sensing. There will also be plugins (if not already issued) from NVIDIA for PhotoShop. But none of those are either GIS or even remote sensing as understood by GIS markets.

It could be that the remote sensing companies do not yet realize this is a matter of life and death for them. That they weren't at Nvision 08 in a prominent way shows they may have already underestimated the matter and might have lost a precious year. It will be interesting horserace to see if they can master this new technology before new entrants can use it to take share from them.

On the vector side of things I have seen no interest at all in CUDA from classic GIS vendors, which is not surprising given their evolutionary path in the market. Converting your software to 64-bit operation is much easier by far than coding CUDA, yet they appear to have a ways to go even for that first step. Likewise for multicore operation, which most folks will try at least in part before venturing onto CUDA.

Plus, there is always the reality that vector work with CUDA is significantly more difficult than raster work. If your main business is rasters, like the remote sensing companies, it's less risky to dive into CUDA. But that's not the case for vector work where you have to grit your teeth and make a big investment in time and money just to test the waters. I can't blame anyone in a primarily vector GIS company for wanting to see if all this razzle-dazzle is real before investing years of time and tens of millions of dollars in giving CUDA a try.

We had other reasons in addition to CUDA for making a big investment into the capital improvements which now support CUDA as well. Since we were going to do it anyway we took the opportunity to design in good support for CUDA. Having a year's experience with CUDA made that a realistic proposition. But the motivations inside the legacy GIS companies are almost certainly very different.

We have committed this past year to improving the desktop customer experience with better performance and capacity and other improvements. ESRI has committed their past year of development into a return to time sharing by emphasizing ArcGIS server application development over modernizing their desktop products. That's why you see parts of ArcGIS Server going 64-bit but not desktop software like ArcView. That's a real parting of the ways and their investment into time-sharing via server-side applications takes them 180 degrees away from the use of CUDA.

Our mantra is that one hundred users sharing one processor is bound to be slower than one hundred processors serving a single user, so we strongly disagree with the direction ESRI is taking. We see it as an opportunity to pick up some of their customers who are not happy about missing out on high performance, modern desktop GIS experiences. It will be a very interesting year ahead to see how that works out. :-)

the town Valenciennes where i live have an europeean meeting call les E magiciens. Big studio specialized in 3D and video effect from USA come every years for meet /discover talentuous students from international art school ( seem they don't use paper ...but PC ...) . The meeting is open to professionnal but consumer people could come .

This year during meeting

1-Nvidia was here to show what there last graphic card could do during technical session !!. . ..amazing !!! the price is more than the card recommand by manifold !!!

2- like amiga-linux -party some demo have to achieve before the end of the meeting . Team of 5 people for flash ( audio-programming-flash designer...) in graphic session and same number for vidéo session ( photoshop- premiere-after effect ...) .Some ll be nominated and ..............

3-real free 3D projection film ocur.film make in 3D for some cinema that accept this format ( very few in france ) .Belgium team seem more advance than USA !! Flymetothemoon

4 see also in cinema some research make by "student" about animation and vidéo

For come back to nvidia : All the pc during the meeting was sponsored ( have nvidia graphic card ) by nvidia and AMD ....never see intel and ATI . So i have also a good feeling respect about nvidia . they were all professionnal and have many questions tooo............. . never see nvidia to microsoft mix perhaps they are in game/3D domain ?

don't know it and discover now that nvidia have sponsored a projection of the 3D animation film "Fly me to the moon" for NVISION 08 the 25 aout at santa clara !!!.

Make a mistake the first 3D Film was USA in november 2007 ‘Beowulf 3D’ make with autodesk maya by sony Pictures imageworks compatible with Dolby 3D Digital Cinema

registration from nvidia french site

Very interesting. I suspect that many software houses are betting the farm on the arrival of Larabee. But I am really surprised that companies like Leica makers of Erdas ERM and Erdas Imagine, and Microimages makers of TNTMIPs, or ITT makers of ENVI were not there.

Nick, you've got a point there, but the classic remote sensing companies have never been known for being early adopters of new technology, the transition to 64-bit operation and multicore parallelism being a case in point. Plus, products like ENVI support so many operating systems (LINUX ??) that it is not easy for them to move forward quickly. But if the usual inertia is not the issue and they are deliberately holding off jumping into the game thinking that the arrival of Larrabee will let them go GPGPU-parallel without a lot of effort, I think they will be disappointed.

Larrabee will be no automatic path to parallelism or to instant high performance without development effort. Just as in the present case of multicore processors, simply having a multicore processor does not mean one's application will be rewritten into either being a) multithreaded or b) that algorithms within it will be sensibly parallelized. Adding many Larrabee x86 cores won't do much, if anything, in general purpose computing unless the application is re-written to take advantage of parallelization. There is some significant validity to the idea that this will be easier to do if it is a more similar architecture to what is already there, but it doesn't save you from the bulk of the job, which is the fundamental parallelization of your application.

A further consideration is that GPUs are much faster at computation than x86 CPUs. A conventional CPU has most of the die given up to cache memory and to control logic for things like interrupts, etc. The many processors on a GPU have very little of that... they are all compute and darned near zero logic overhead. Massively parallel GPUs are set up to run zillions of threads with hardware management of those threads to do things like conceal memory latency. So, yes, it takes more thinking to write your application to use that but once you do it goes like crazy. In contrast, a conventional CPU takes less thinking to write your application but it might not go as fast. The flexibility a conventional CPU core offers is very good but it comes at the cost of a major performance hit at times, for example when you get a stall on a cache miss.

When Larrabee will be introduced, where it will fit in the GPU to CPU architectural continuum, at what price and with how many cores and with what performance is not something Intel has disclosed, so any speculations are premature. But, you know the crew at Manifold... nobody is going to complain if we get more cores from Larrabee! :-) When it starts shipping we will, of course, support it.

In the meantime, NVIDIA is here today with about 100 million CUDA-enabled GPUs already shipped and a very wide range of CUDA-enabled products at just about every price/performance point you might want. I'm glad we're supporting those today and if the remote sensing folks are waiting for Larrabee I think they're making a mistake.

Re: Futurists

I'm sure you know of Ray Kurzweil, Dimitri, but for those who don't he is a futurist with a pretty good track record (see http://www.kurzweilai.net/index.html?flash=1).

Ray points out that most everything around us is on an exponential timeline - either slowing down or speeding up. Computing of course is an obvious example. He's shown, by plotting various aspects of computing (cost, size, bang for the buck, power consumption, number of transistors, MIPs, FLOPs, etc) that we are moving through the sharpest part of the exponential curve right now.

He also shows that just as one technology runs out of steam (sometimes literally), another one 'magically' steps in to take its place.

We are clearly at the end of the road with process shrinks, threatening the exponential curve. But here comes massively parallel computing to take over.

It's quite humbling to see Ray's theories coming true around you. I guess the next step is carbon nanotubes and still further until we've been through the singularity and converted the entire universe into computronium where humanity is completely virtual. I guess that would be NV's fault

Seriously though, I highly recommend everyone check out the link.

Will Render (for?) Food

Fascinating Dimitri, thanks for the report.

Tragically, I can only imagine the military applications for all of this

Military applications, from what i've seen, don't rely on heavy processing of numbers. I think they're more concerned with getting timely data correct.

Nanotechfighterbots are another thing though

I think Colin is probably thinking of the modelling of big bangs

An of course there is always code breaking

Then again, the military might be in a different ball-park already:

http://www.eweek.com/c/a/Government-IT/IBM-HP-Building-New-Infrastructures-for-US-Military/

What would it take to get to 90 teraflops using NVIDIA CUDA cards?

A motherborard with ninety PCIe slots?

Seriously here is more to a fast supercomputer than lots of processors. There needs huge amounts of RAM (Tbs) so that the enitire project can be held in memory without the bottleneck of accessing harddisks.

What would it take to get to 90 teraflops using NVIDIA CUDA cards?

A single NVIDIA T10 Tesla card is almost a teraflop. Do a bin select to choose slightly higher clock rates and water cool it (ASUS makes a water block cooler for G200 NVIDIA GPUs and was demoing multiple water cooled cards installed in a system at Nvision 08) and you get a teraflop per card. Install four cards per computer and you need 23 computers to get 90 teraflops.

23 PCs should use much less electrical power than the 149 chassis (with 16 processors each... the Power6 is a dual core general purpose microprocessor) in the IBM design. It would probably be a twentieth of the cost as well.

More importantly, the actual throughput of the NVIDIA based system might likely be much greater than of the IBM system. The 240 processor G200 series GPU inside a T10 Tesla card is a much faster compute engine than a dual core Power6, and having 4 GB on the card with 960 processors in the same chassis working with 16 GB of on-card memory gives a lot of immediate processing power that does not have to go through networks between the computers.

The other interesting aspect of this is that NVIDIA performance roughly doubles every year (it's actually taking about 15 months). Next year that same performance level will require only 12 computers with 4 NVIDIA cards each, or the cluster built now could be upgraded with new NVIDIA cards to get 180 teraflops... because CUDA is transparent to the underlying architecture you could actually do an upgrade with new cards and not change your software or anything else. Cool!

As a matter of interest would it be possible to run Manifold on a cluster with Windows HPC Server 2008?

http://www.microsoft.com/HPC/default.aspx#datasheets

Release 8 does some multicore/multiprocessor CPU parallelization for things like DBMS access and image library rendering, and uses CUDA within the Surface - Transform dialog for GPGPU parallelization. But it will not parallelize across multiple separate systems configured as a cluster.

Parallelization plans for Manifold going back a few years originally looked at using clusters, because that was realistically the only way to go in those days if you wanted to go parallel. But now the compute density available from GPGPU computing is so much greater with such enormously greater bandwidth to memory than is possible across even fast networks that clusters are taking somewhat of a back seat for anything but the very largest applications (where one is forced to go to clusters). GPUs have become so fast that in most applications it is better to plug four GPGPU boards into a single machine than cluster out.

I can see, however, that as Manifold gets more and more parallel it could pay to add a third layer of parallelization (the first two layers being massively parallel GPGPUs supported by multicore / multiprocessor CPUs in the same box) by clustering out to multiple systems. Manifold has a few ideas in that direction for future releases. :-)

Correction: it is EVGA that makes the water block cooler, not quite yet available in distribution but within a few weeks. Supposedly it will be around $250, but it is very well made, machined out of a massive solid copper billet. It is designed to be plumbed into an existing water cooling system that one is presumed to already be using to cool the CPU.

Here's a picture from Nvision 08 of the EVGA water cooling installation with three 280 GTX cards and a CPU water cooler as well.

Attachments:
evga_coolers.jpg

Another picture, an even cooler one... (ahem...)

Also from the EVGA booth at Nvision 08 where they were using liquid nitrogen to cool down CPUs and GPUs. The photo shows a speed trial done with only the CPU being cooled. In the center background of the photo, right next to the small bottle of apple juice, one can see cooling blocks that were used earlier in the day to demo liquid nitrogen cooling of NVIDIA GPUs as well.

The use of liquid nitrogen and (also popular) dry ice bathed in acetone are stunts, of course, but they show clearly how it is the inability to prevent overheating that is part of the limitation on processor speed by limiting input voltages and clock rates.

Attachments:
nitrogen_cooling.jpg

The report on NVision 08 from Cadalyst.

http://management.cadalyst.com/cadman/Event+Report/Event-Report-NVISION-08-NVIDIArsquos-Inaugural-Vis/ArticleStandard/Article/detail/546824?contextCategoryId=6764

Manifold mentioned

Just read an interesting article in an old New Scientist magazine (July 2007) "Darwin and the generation game" about designing all sorts of stuff using evolutionary algorithms. Basically allowing a design to evolve in a computer simulation, rather than the usual "intelligent design" approach. It mentions this being made possible by advances in parallel computing. Must be really taking off now.

The conference had outstanding organization. NVIDIA did a super job for everyone. The free food was great, the facilities incredible (more high-end Christie projectors than Siggraph, it seemed...), free high speed WiFi and free Internet laptops everywhere and genuinely helpful "info" people in the corridors ready to help find meeting rooms, etc. Well done, NVIDIA!

Free food? everthing for free? No wonder you have such praises. Sounds cheap.

You are sounding like a salesman for Nvidia.

Everytime a graphics card company releases a new card, you get all the hype possible. Go back and search for the release of Nvidia 12 months ago and see whatever happens to it. You could get any "top of the line card" 12 months ago for less than $30!

This too will pass...sadly..without much impact on the everyday user....

Go back and search for the release of Nvidia 12 months ago and see whatever happens to it. You could get any "top of the line card" 12 months ago for less than $30!

The top of the line card from Nvidia 12 months ago was and remains the Quadro FX 5600 and cost around 2000USD (may be somewhat less if you were imaginative). Or, the Nvidia's Quadro Plex boxes containing up to 4 FX 5600's cost around 10000-15000 USD ea and the Tesla D870 was about 6000USD.

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

There are two kinds of people in this world my friend, those that know what they are talking about and those that dont. Once gain you have demonstrated your ignorance..

Oh yeah, I can now see manifold users waiting inline to buy the quadro plex boxes worth $10k.

maybe george lucas has one or maybe steve jobs at pixar has one...

do you know of anyone in the GIS industry who has a plex installed in their desktop?

the main reason given by users who use manifold is -

'oh it only cost $400 and does all the functions of arcgis that costs $10k.'

usually this means that they are not willing to spend $10k.

now you are telling me that these same people would buy a video card (or video box) that costs $10k?

$10k to run a GIS map with a file size of 1GB max? I can see te reaction of my purchasing officer when I give in a request for a quad plex card...

Also I may be interested to know why only Manifold as a GIS company was invited by Nvidia? Could it be that they are sensing that there is an overkill in design here? we are not rendering the entire globe you know. Even google earth can run on my lowly ati radeon card on 256 of shared memory at that.

i can now see the ad of nvidia...buy manifold bundled with nvidia quad for only $400 + $9,999*

*cards are sold separately and may require additional pcie slots & power supply and cooler pads ,etc...

"do you know of anyone in the GIS industry who has a plex installed in their desktop?"

I know that geophysicicsts and oil explorationists use this kind of hardware to enable visualisation of reservoir structure. Ditto anyone working with city models, and large terrain moels, who want fly-throughs. So in answer to your question, yes. However, my answer was to your question above, about top of the range Nvidia Graphics cards. You suggested a top of the range Nvidia grcard 12 months ago can now be had for 30USD. You clearly did not know anything about Nvidia's product range.

"the main reason given by users who use manifold is -

'oh it only cost $400 and does all the functions of arcgis that costs $10k.'

usually this means that they are not willing to spend $10k.

now you are telling me that these same people would buy a video card (or video box) that costs $10k?"

No, it means they have 10K that they would have otherwise spent on obsolete 32-bit software, to spend on the best hardware they can obtain (if they need it).

"10k to run a GIS map with a file size of 1GB max? I can see te reaction of my purchasing officer when I give in a request for a quad[ro] plex card..."

Yes he would think you are an idiot (but you might benefit from a 1500USD card, if you had the software to uitilise it). But if you were working on the structure of an oil reservoir and needed to understand where to put the next well, and if you did not make the right decision, it could be a 100million dollar fiasco, then he would be an idiot to deny you the hardware to allow you to visualise the rerservoir. Ditto if you were producing a 3D virtual reality model of London, for instance.

However, you have rather missed the point about graphics cards from Nvidia. It is this, they are not now restricted to performing graphics operations. With CUDA they can be used to peform just about any operations that requires huge numbers opf calculations to complete and which can be parallelised. A good example of this is the field of numerical weather prediction. Until recently if you wanted to run a NWP model, you required a supercomputer that filled an entire floor. You can nowrun such models on a desktop with multicore CPUs and I know that soon (perhaps already) by harnessing Nvidia CUDA enabled GP-GPUs.

"Also I may be interested to know why only Manifold as a GIS company was invited by Nvidia?"

Were they? My impression was that Dimitri went on his own accord - ask him. But why not be invited. The company is one of the first to take up CUDA and the only GIS manufacturer that has seen the advantages of using CUDA.

"Could it be that they are sensing that there is an overkill in design here? we are not rendering the entire globe you know."

Why not, if you want to you might soon be able to (if you have the data) :-)

Were they? My impression was that Dimitri went on his own accord - ask him. But why not be invited.

Very true - it is a nice feeling to be invited, but with NVIDIA it is easier than that, which is one reason they have become incredibly strong. NVIDIA gives open access, for free, to all the tools you need to work with CUDA. Anyone can download them at zero cost and begin working with CUDA right away. No need for secret handshakes, no need to form "partnerships" or any of that old-fashioned guff. Just download the tools and get to work and if you want to succeed it's all there for the taking. If you don't get into the game you have no one to blame but yourself.

NVIDIA plays no favorites: all are welcome. Everyone is encouraged to participate in the forums, to send in news of new applications using CUDA, etc. NVIDIA is happy to pass on news of wonderful new things accomplished with their technology, to cite remarkable new uses of CUDA in CUDA Zone and all that good stuff. It's all good and open to anyone who gets off their rear ends and gets into the game. It was that way a year and a half ago when we started and it is that way today.

Even from the very beginning there was no need for any secret handshakes to get started with CUDA, and there was plenty of discussion in computing circles about GPGPU. You either had the brainpower in your own head to realize this was important and something that should be done or you didn't.

For us it was a natural thing to jump into CUDA because of our team's extensive experience with parallelization, processor architectures and the like. So we went to work immediately, and were very pleased to see how easy NVIDIA made it for us by putting everything on their web site for immediate download.

Look, I want to make it clear I'm not arguing against any personal relationships. Those are great, but even in the best of relationships it is usually more efficient if an engineer can point and click and download something from a web site than if he or she has to ask a manager and that manager then calls up his or her counterpart at the other company to get something special. Even if there is an instant and eager response it is quicker for the engineer to simply point and click and get it even faster. NVIDIA made that possible.

I have to admit when we first started with CUDA we were glad to have personal relationships with NVIDIA just in case something came up. But the simple facts of the matter were that nothing came up that required any panic phone calls or emails. NVIDIA did such an outstanding job of putting everything out for anyone to use that we never needed anything special. We just got to work and did what we needed to do. That's something any GIS company could have done.

Why were we the only GIS company to dive in while the others hung around like a bunch of morons not invited to the dance? You have to ask those others how they missed such an important call, as the door was wide open to them as it was to us. But on the way to asking that you have to ask why they were so late to support Vista and still haven't done Server 2008, why they were so unthinkably late doing 64-bit anything, and why they totally missed the call on even elementary first steps to use multicore/multiprocessor conventional CPUs.

In all fairness since our first use of CUDA we've obviously grown closer to NVIDIA. But that's a relationship open to any software company writing huge code shipped to lots of customers and building a track record of success. We have so many customers in common now that it makes sense for Manifold and NVIDIA to work together in routine coordinations such as doing lookahead on updates so we don't end up with version skew that could easily be avoided with a few weeks' worth of forethought.

It's also the case that when you have products with millions of lines of code in play and hundreds of thousands of lines of NVIDIA-specific code those things don't happen overnight, and a company like NVIDIA that understands software development is going to make it easy for ISVs like us to commit to their technologies by giving us reasonable lookahead on things they are doing that will take us time to incorporate. But those are all matters which do not affect folks just getting started with CUDA as you have to get some ways down the learning curve before any of that specific stuff matters to you. NVIDIA already gives very good lookahead on the forums and in public statements for anyone who needs general guidance.

As far as Nvision 08 was concerned, the event was open to everyone and if anyone wasn't there they have only themselves to blame for missing a spectacular gathering. Not everyone can be a presenter, of course, and NVIDIA did a good job of getting presenters and other participants to attend who they felt would make it a worthwhile experience for attendees. There were very many companies mentioned or featured in the presentations, including Manifold. We are honored by that recognition, of course, but in no way did NVIDIA play any favorites. Had any other GIS company rolled up their sleeves, jumped into the fray and got something accomplished they too would have been eagerly welcomed by both NVIDIA and all the other participants.

Anyone who wants to jump in and start developing for CUDA can do so today. You don't need anything but a point and a click to download those tools that are open to all. Get those applications done and send them into CUDA Zone and come to Nvision 09 next year.

There are two kinds of people in this world my friend, those that know what they are talking about and those that dont. Once gain you have demonstrated your ignorance..

Nick,

I may be ignorant with regards to "new" trends but I have something better...I have common sense.

And common sense tells me that I will not be spending $10k for a video box when I can get a vc released 12 months ago for $30 that would do the same job...

I may be ignorant with regards to "new" trends but I have something better...I have common sense. And common sense tells me that I will not be spending $10k for a video box when I can get a vc released 12 months ago for $30 that would do the same job...

I dont believe I was advocating anything of the sort! If you need it, it is available. Most people dont need this kind of hardware, but you certainly wont be getting a workstation graphics card that costs 2000USD now, in twelve months time for thirty. A 30USD card certainly wont give the sam performance. You need a reality check and head examined if you think this.

It is important to realise why high end graphics cards cost more than ones for gaming etc, it is partly because they are superior hardware, faster GPUs, more and faster onboard memory etc, but it is largely because they use superior firmware (drivers) that are highly optimised.

Why were we the only GIS company to dive in while the others hung around like a bunch of morons not invited to the dance? - dmitri

Yes he would think you are an idiot (but you might benefit from a 1500USD card, if you had the software to uitilise it).- nick verge

Wow the vocabulary used by so called 'elite' of Manifold really shows the type of people behind the software.

What is the saying - birds of the same flock, feather together?

Fortunately, I am not a bird...

I have been searching the forums of Microsoft, ESRI, Oracle and so far I have not seen such language being used.

You are on your way to becoming the #1 software company in the GIS Industry. God Help Us.

Good night.

Nick, he's not willing to read 5 lines of text in a paragraph nor able to read subjunctive. Don't bother, don't feed the troll.