| NVIDIA Editors Day 2008 |
| News - Featured Website News | |
| Written by Olin Coles | |
| Saturday, 24 May 2008 | |
NVIDIA Editor's Day 2008Each year NVIDIA hosts a invite-only event for industry editors. Everyone from reviewers to magazine editors were on hand to learn the reasons behind NVIDIA's latest graphics processor launch. At this editor's day event, the heads of technology converge for a day of demos and discussion - and a few nice surprises. Benchmark Reviews was pleased to receive an invitation into this exclusive event, and looked forward to the opportunity to share thoughts with the biggest name in the graphics market. Prior to actually experiencing the NVIDIA Editor's Day event, I was a little bit concerned that my time would taken up with buzz-word marketing pitches and hard-line product comparisons. To my complete surprise, there wasn't a single moment that had me feeling green-washed. Actually, the opposite was true; at several points throughout various topic discussions the NVIDIA staff would often times point out where a competitors product might have done better than their own in a last-generation product lines. In fact, there were many times where the discussion felt more like a tech discussion than the voice of an industry leader. At its core, the NVIDIA Editor's Day event was about a sharing concepts and developmental ideas with the theme Gaming Beyond: Beyond Gaming. Heterogeneous Parallel ComputingUjesh Desai: GeForce General ManagerNVIDIA Editors Day 2008 began with a warm welcome from Ujesh Desai, the GeForce General Manager. His introductory overview highlighted the "GPU Beyond" theme, which dispelled the rumor mill of an Intel vs NVIDIA rivalry and replaced it with a more heterogeneous computing environment that matches the CPU with a relevant and more appropriate GPU.
The message Ujesh wanted us to walk away with is that for the same cost, any custom-configured tier one system could dramatically improve performance for many of the most common applications by simply reducing the processor option one level and increasing the GPU option up one level. The benchmark results were very impressive, showing that while the first system was CPU heavy it didn't exactly translate into higher performance overall. Conversely, for the same price a matched CPU-GPU computer system performed several levels better in everyday applications: from Windows Vista, to Adobe Photoshop, to playing video games. Demonstrations began with an interesting third-party feature that enforces the importance of heterogeneous computing: the Adobe PicLens plug-in for Google Images sub-site. I know that from personal experience my search for images on this site has felt a lot more like browsing pages and less like finding what I want. With PicLens, the NVIDIA GPU lends itself to creating a scalable mosaic of photos which allow the end-user to pan, tilt, and zoom groups of images as a collection without needing to browse through many pages.
Mr. Desai gave the floor away to John Mack of Adobe, who took some time to introduce a few new features found in Photoshop CS4, and leaked some details during his demo of Adobe Stonehenge. While graphic professionals will undoubtedly have a lot of interest in the new Adobe products, the primary message was that Adobe finally recognizes the need for powerful GPUs; with the same level of importance that the industry has placed on CPUs. Using a very large 2 GB file for demonstration, the CPU became less utilized and takes a distant back seat to the more critical GPU, and everyday functions such a free transforms and manipulating angles becomes a lightning fast operation with the right tools.
Beyond Traditional GraphicsDavid Kirk: Chief ScientistAs an enthusiast, I have a hard a hard time adapting to technologies that I don't interface with on a regular basis. Technologies such as Ray Tracing, Global Illumination, and Radiosity all play an important role in delivering first order reflections of light by the GPU. David Kirk, the Chief Scientist within NVIDIA, gave a worthy effort towards explaining the free CUDA (Compute Unified Device Architecture) programming tool to a crowd of editors more hell-bent on video game frame rates. Thankfully, there were relevant ties into the consumer markets, and the discussion on enterprise computing harnessed by the GPU transitioned into a discussion on how the industry can use Ray Tracing to make more realistic graphics in video games. Keeping in mind products like the Sony Playstation 3 utilizes an NVIDIA GPU, and because of this there's a new opportunity for developers to take advantage of Ray Tracing in future platform products; although there was very little speculation that there might be a widespread adoption during the Playstation 3's lifetime.
Beyond GamingJohn Milner: Director of CUDA DevelopmentJohn Milner, the Director of CUDA Development, stepped up to help bridge the gap in consumer understanding of the compute technology and what specific applications it benefits. CUDA helps transform GPUs from the GeForce, Quadro, and Tesla platforms into a compute environment. It became clear despite the extraneous use of programming terms that almost anything that uses a GUI or graphical display could ultimately benefit from the CUDA profiles. From white-matter brain scans to CAD-level applications, the 100 GBps worth of streaming compute bandwidth, CUDA offers the opportunity for basic code developers all the way up to HPC program writers to take advantage of the free programming. Later in his briefing John demonstrated how Hanwech Associates Volera, one of CUDA's early independent adopters, has now been given the ability to compute worldwide market financial valuations in mere minutes (using last generation hardware), whereas previously it took weeks. Another key demonstration showed how CUDA could lend itself to speeding AutoDock cancer-research scans up to 12x faster than previous results, allowing for a much quicker response and diagnosis. So far the information was very exciting; but not nearly as exciting as how these developments were all done via free SDK download without NVIDIA assistance. The private-sector industries really seemed to have given CUDA the proper porting and made out ten-fold for their efforts. Engineers will rejoice in a 17x improvement in MATLAB processing times using last generation (8800-series) GPU hardware, for example. The theme Mr. Milner impressed was the fact most developers obtained their results for free, and only submitted their results after the fact... meaning NVIDIA released it, and everyone else took the technology and ran with it. CUDA is available for FREE from NVIDIA here: https://www.nvidia.com/object/cuda_home.html What's the Graphics Architecture?Tony TamasiAs the day wore on, I began to feel that the GeForce 9800 series was really more of a clearinghouse for the G92 GPU. When Tony Tamasi returned to give his lecture on the architecture of the GeForce GPU, my feelings were confirmed. It amazed me how different the end-goal was between a CPU manufacturer, and the goals of a GPU manufacturer such as NVIDIA. While Intel keeps loading the processor with a growing cache buffer and refining their fabrication process to achive higher speeds, NVIDIA designs a GPU that is really made to compute to a higher level by increasing the number of processor cores.
The Optimized PCSam Blackman: Elemental TechnologiesNVIDIA's development partner Elemental Technologies offered an insightful transcoding demonstration, spear-headed by CEO Sam Blackman. Sam briefly described the encoding experience for Premiere Pro and FinalCut Pro, and showed how a top-shelf Intel Quad-Core processor would take more than five hours to transcode the same file in only 35-minutes with the upcoming GeForce GPU. In the first demonstration, RapiHD technology allowed the end user to create and encode media at speeds far faster than previously available. Using the BodaBOOM media converter, Sam successfully encoded a two-minute video clip in mere seconds whereas Apple iTunes was barely 10% complete. In the second demonstration, Adobe Premier Pro (using H.264) took a random video clip and encoded the 1440x1080 file at an astounding 48 frames per second! Anyone who works with Premier knows one fact to be true: encoding takes an excessive amount of waiting time. This is all a necessity of the past, as the new CUDA architecture paired to the upcoming GeForce series GPU (but also includes past generation products as well) has revolutionized the encoding experience. Unlike in the past, CUDA will remove the down-time associated with the transcoding process. While CUDA is multi-GPU aware (via PCI-Express bus), developers are still working to implement the new architecture as a standard. What this ultimately means is that professionals working with video can immediately see phenomenal gains out of CUDA with the upcoming GPU series... and an outstanding boost in performance from previous GeForce 9600 series GPU's and earlier. Essentially, previous generations show a linear gain across the number of GPU cores available. Compute dependency also comes into play, and professionals can determine what percentage of processing is dedicated to each particular need. Folding @ HomeVijay PandeDuring the discussion of CUDA, which was easily the focus of NIVIDIA Editor's Day, Stanford's own Vijay discussed how Folding @ Home could take advantage of the 70-million GPU's currently in existence and pair them with CUDA. During the Folding @ Home demonstration, compute performance was impressive while using the optimized GPU. Now I'm a huge fan of F@H (just look up Das Capitolin), and it made an immediate impact on me how quickly a GPU has outpaced the CPU in molecular simulation.
Last year ATI made headlines when they created a GPU with compute capability to assist in folding for the F@H project. Some may have perceived this as a market gimmick, but compute ability on the GPU is something that Intel and AMD are going to be facing as real competition if they don't address it soon. Mr. Pande went on to discuss some of the upcoming changes for the project, such as beta releases of the CUDA-powered client for the F@H project. More detail on the Folding @ Home project are available here: https://folding.stanford.edu/ Gaming Beyond with NVIDIA PhysXManju Hegde: VP PhysX SolutionsPhysX has an outstanding name built from such a short duration of existence. With over 150 different games available to all platforms, and over 25,000 registers/active developers, PhysX titles have become some of the most realistic displays of simulated physics ever found. Video games are now seeing an added dependence on physics processing, just as well as GPU and CPU. Last-generation GPUs from NVIDIA's GeForce graphics processor series have already begun the shift of processing dependency of video games onto the GPU, so it won't be very much longer before the CPU really offers no added level of performance to video games. NVIDIA acquired PhysX only a few months ago, however it now takes mere months to produce PhysX titles instead of the year long process previously experienced; thanks to the ease of adoption with the CUDA programming interface. Physics reproduction in graphics development isn't limited to PhysX alone; their closest competition is Havoc. While both offer the same functionality across the different platforms, only PhysX is is going to benefit from CUDA and future advances in the NVIDIA GPU.
NVIDIA NVISION 08Bob Csongor started the next discussion on NVISION 08 by asking who in the room had attended a GeForce LAN event. Ironically, there were only a few hands raised in an packed wall-to-wall room; mine wasn't one of them. The meaning behind his question was to demonstrate how the enthusiast community operates and thinks. He enforced his point by explaining the NVIDIA is a whole lot more than video games. Describing commercials using graphics manipulation, or the simulated first-down line during a broadcast football game, Bob was able to convey that graphics are well beyond gaming. This was more or less the entire theme of the Spring 2008 Editor's Day: while gamers are the core of NVIDIA, the company offers gaming beyond, and go beyond gaming. Mr. Csongor led the group into a discussion that spanned from upcoming attempts to set a Guinness world record at the next LAN event, into the technical and marketing material that will be offered in the next NVISION 08 event. To my surprise, there was an enormous diversity between the topics. Visual computing topics fro the enthusiast, to the professional, and everywhere in-between. Of course there will be the GeForce LAN, and case modifications events, but training workshops and exhibitions fill the other half of the agenda. Industry speakers who shake ground with their bleeding-edge research are expected to offer an insight beyond the level we rumor in our writings. Of course, there are also the big developments such as Microsoft's DirectX 11 and up-and-coming PhysX developed titles. Of course, the theme again was focused on gaming, and everything not gaming. Some of the names on the itinerary carried an NVIDIA tag, but it was refreshing to see the vast majority of the key speakers be outside of the company... even retired Space Shuttle Commander Eileen Collins to share her experiences on how space flight has changed with technology intersections. More information on NVISION 2008 is available here: https://www.nvision2008.com/ Closing Thoughts: NVIDIAAs the first half of the day came to a close, lunch was being prepared and a line was forming. I wasn't in a particular hurry to eat (perhaps 'feed' might be the appropriate term for the cattle drive stampede I witnessed), so I remained back in the conference room to talk with David Kirk about CUDA and the Folding @ Home project. For background, the topic of CUDA seems relatively new to me; there was a lot more talk of TESLA in our SC07 SuperComputing Coverage of NVIDIA than there was talk of CUDA. It confused me as to why a nearly two-year-old project has suddenly received a double-dose of adrenaline in the past month. Mr. Kirk put things into greater perspective when he mentioned that until recently NVIDIA wasn't known for a program architecture, and that with the recent dramatic-yet-true success stories paired with college courses becoming available for programming in CUDA there was finally a reason to tout the project.
So I joined the end of a very long lunch line full of knowledge, and an empty stomach. Ironically, being one of the last people to sit down in the cafeteria played to my advantage. Mr. Jen-Hsun Huang, President of NVIDIA, sat across from my table and began an inviting dialog. Now, I have to stop for a moment here and confess: up to this point, I have heard the name and seen a picture or two, but when Mr. Huang sat down next to me I had no idea who he was. Perhaps it was because I was the only person in a suit and tie at this event, or because I was the last to take a seat for lunch, but he and I began to discuss the industry as a whole and within ten minutes my seemingly empty table filled with other editors. It took a few more questions before it all clicked with me who this was (he does look very young for a CEO, after all), and it impressed me that we were encouraged to talk with unrestricted candor. More than anything else, I was surprised at the level of respect given back to the editors. Some manufacturers see the technical writer as a necessary evil, and despite our power over consumer influence we still go very much unappreciated. This wasn't the case at all while visiting NVIDIA. At the end of the the editor's day event, Mr. Jen-Hsun Huang spent nearly a half-hour taking on-the-spot questions. After receiving almost a dozen random questions it occurred to me that the answers were not only on the mark, but they were legitimate and honest. Part of me already understands that when you truly grasp a topic, there isn't a single question that can shake you; but the other part of me thought that there has to be the underlying purpose of swaying opinions. Sure, Mr. Huang has an agenda, but his agenda is so purpose-driven it made his discussion on heterogeneous computing seem just as noble as saving the planet or conserving energy.
After catching Jen-Hsun alone in the parking lot after the event, he wasn't at all apprehensive of what I might ask or hurried in his response. My question: "When do you foresee the next NVIDIA GPU to be produced using a smaller fabrication process?" To my complete surprise, Mr. Huang answered: "We count our blessing every time a product is designed." He went on to explain that GeForce GPUs already ride the edge of technical boundaries, and that a reduced die process doesn't always translate into better performance or higher efficiency. It caught me a little off guard to hear the top NVIDIA rank tell me a completely genuine answer, without the slightest tone of marketing gloss. It's difficult to gauge a person and their agenda in only one day, but if Jen-Hsun Huang is as genuine towards his goals as he states, the industry is going to be tossed on its ear in a short matter of time. The biggest problem I see with so many large companies is that they have no voice. Take for example Intel and AMD: ask yourself to answer who they are and what they stand for. See my point? I couldn't even begin to tell you their mission statement... but my insider observation lends me to believe it's all about profit. That's not to say that NVIDIA isn't taking their cut - they are publicly traded after all, but every single thing they do inside their design rooms is to solve a consumer problem. I've been building and selling computers for almost nine years now, and only once have I ever really felt that the big-name manufacturers were working for the everyman (that was back when AMD introduced the Athlon series of processors). AMD might be trying to re-live that moment with affordable low-end solutions or neutered tri-core processors, but it's too little perhaps too late. Intel on the other hand seems to be back on their Pentium 4 collision course where it's now about cores like it used to be about clock speed. I don't believe Intel is answering any real consumer need by producing ever-expanding cores for a limited software canvas. It seems that NVIDIA is filling the long-needed void, and offers a revolutionary change for computing with the CUDA programming architecture and graphics that feature extremely efficient processors. The future looks green, without a doubt, and the changes taking place right now are already changing the way the world does business. When I upgraded from a dual-core to quad-core Intel CPU, I noticed barely any change in performance and all of my benchmark test scores seemed to confirm only single-digit improvement; this isn't exactly what I call progress. Giving businesses ways to improve their productivity by over 200% using a free CUDA software development kit, a technology that doesn't give NVIDIA a single dime, is more towards the definition of progress. Questions? Comments? Benchmark Reviews really wants your feedback. We invite you to leave your remarks in our Discussion Forum.
Related Articles:
|
|
