Nvidia today formally introduced its K20 GPU family, supposedly the highest performing and most efficient accelerators ever built. If you recall, these new GPUs were featured in Titan, the world's fastest supercomputer that was unveiled by the U.S. Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL) a couple of weeks ago.
With CPU-integrated graphics upping their game and handling most casual gaming and HD video watching tasks without too many problems, AMD and Nvidia could find the bottom dropping out of the low-end discrete graphics market before long. The question is, how do the companies make up for it? One answer lies on the opposite end of the spectrum: high-end super-duper-computers often take advantage of the raw computational power of GPUs. That same processing efficiency is opening up new (final) frontiers -- Nvidia's helping a Lunar X team in their quest to go to the moon, Alice!
Steve Scott is an accomplished man. For the past 19 years, Scott's been working at Cray, where he served as vice president and CTO for the past 6 years. Prior to that, he was the chief architect of the Cray X1 scalable vector supercomputer and helped design several other Cray systems, including the XT, XE, and "Cascade" systems. Scott holds 27 U.S. patents in a variety of geeky areas (cache coherences and scalable parallel architectures to name just two), and is now bringing all that brain power to Nvidia.
With the Cold War a thing of the past, Russian scientists are free to concentrate their efforts on projects other than the space race and building a stockpile of nuclear arms. These days Russian scientists are studying computationally heavy topics like global climate change, ocean modeling, post-genomic medicine, and galaxy formation, and they're tapping into Nvidia's Tesla GPUs to do the heavy lifting.
Tesla Motors CEO Elon Musk isn't buying into the theory that the world will end in 2012. How do we know that? Because the ambitious CEO is making some bold predictions for the next two decades.
According to Musk, long range vehicle battery backs will be ultra affordable by 2020, and if you find that hard to believe, get this. By 2030, Musk predicts that all cars sold in America will be electric, egmCarTech.com reports.
Musk's time-frame might be overly optimistic, but it's understandable given his company's desire to push electric vehicles. Tesla plans to unveil the Model X SUV later this year, which will be followed by an affordable electric sedan priced around $30,000.
With more than a little help from Nvidia, Amazon hopes to bring supercomputing to the masses through a new Elastic Computer Cloud (EC2) offering called "Cluster GPU Instances."
By "masses," we're referring to enterprises and start-ups, not everyday Joes looking to run Crysis or improve their Folding@Home score (go Team 11108). Amazon's new Cluster GPU Instance is a server with two quad-core Intel Xeon X5570 processors, two Nvidia Tesla M2050 GPUs (Fermi), 22GB of memory, 1.7TB of storage, and a 10Gb/s Ethernet connection, Amazon's Jeff Barr said.
"With Amazon Cluster GPU Instances, our customers now have the power of high performance computing, the efficiency and speed of GPUs, and the highly scalable and affordable cloud environment our customers have come to expect fro Amazon Web Services (AWS)," said Peter De Santis, GM of Amazon EC2. "We're excited to help our customers access the raw power of GPU technology and look forward to the innovation this will enable."
Using all that GPU power can be tricky, but Nvidia says hundreds of applications have already been ported to the Nvidia CUDA architecture, making it easy for programmers to dive right in.
Nvidia on Tuesday announced that Hewlett Packard's flagship Z800 workstation computer is configurable with up to two Nvidia Tesla GPUs.
"The adoption of Tesla GPUs is the fastest of any new processor technology in the history of HPC," said Andy Keane, general manager of Tesla business at Nvidia. "We are delighted to see a leader such as HP begin to ship Tesla GPU-enabled systems into the market and to help accelerate the work of their customers."
Nvidia's Tesla GPUs differ from standard graphics chips in that Tesla is built using the company's massively parallel CUDA architecture, featuring 240 cores per processor. Tesla-based hardware solutions are designed for CAD/CAM,CE, computational finance, computational fluid dynamics, geographic information services, imaging, life sciences, and other high performance tasks.
In addition to up to two Tesla GPUs, HP's Z800 comes configurable with an Intel Xeon 5500 quad-core processor, Intel 5520 chipset, 4MB or 8MB of processor cache, and up to 192GB of DDR3-1333 memory.
Nivida and Super Micro have worked together in order to create a 1U server that ties together the power of massively parallel Tesla GPUs with multi-core CPUs. The system is said to deliver 12 times the performance of a traditional quad-core CPU-based 1U server.
The SuperServer 6016T-GF-TM2 is on display at Computex this week. “Our new Tesla GPU-based SuperServer 6016T-GF Series delivers a much higher performance-per-watt and per-rack than any other 1U solution in the market today," said Don Clegg, Super Micro‘s Vice President of Marketing. "This 2-Teraflop SuperServer meets the most demanding enterprise data center requirements for reliability and manageability."
Reportedly, Brazilian energy company Petrobras has already installed a cluster of 190 Tesla GPUs and is seeing a 5x to 20x improvement over their previous, multi-core CPU-based clusters.
To those looking for another venue to get their very own supercomputer, you’re in luck! Nvidia has recently announced that their CUDA-based Tesla C1060 GPU is available in Dell’s Precision R5400, T5500 and T7500 workstations effective immediately.
If you’re worried that just one of these GPUs isn’t enough to handle your hardcore needs, worry not – just one C1060 has enough power to control the main system of the European Extremely Large Telescope project (reportedly the world’s largest). According to Jeff Meisel with National Instruments, a workstation “equipped with a single Tesla C1060 can achieve near real-time control of the mirror simulation and controller, which before wouldn't be possible in a single machine without the computational density offered by GPUs."