GTC 2013: Cortexica Vision Systems Talks About the Future of Image Recognition During the Emerging Companies Summit

Subject: General Tech, Graphics Cards | March 20, 2013 - 09:44 PM | Tim Verry
Tagged: video fingerprinting, image recognition, GTC 2013, gpgpu, cortexica, cloud computing

The Emerging Companies Summit is an series of sessions at NVIDIA's GPU Technology Conference (GTC) that gives the floor to CEOs from several up-and-coming technology startups. Earlier today, the CEO of Cortexica Vision Systems took the stage to talk briefly about the company's products and future direction, and to answer questions from a panel of industry experts.

If you tuned into NVIDIA's keynote presentation yesterday, you may have noticed the company showing off a new image recognition technology. That technology is being developed by a company called Cortexica Vision Systems. While it cannot perform facial recognition, it is capable of identifying everything else, according the company's CEO Ian McCready. Currently, Cortexica is employing a cluster of approximately 70 NVIDIA graphics cards, but it is capable of scaling beyond that. Mcready estimates that about 100 GPUs and a CPU would be required by a company like eBay, should they want to implement Cortexica's image recognition technology in-house.

The Cortexica technology uses images captured by a camera (such as the one in your smartphone), which is then sent to Cortexica's servers for processing. The GPUs in the Cortexica cluster handle the fingerprint creation task while the CPU does the actual lookup in the database of known fingerprints to either find an exact match, or return similar image results. According to Cortexica, the fingerprint creation takes only 100ms, though as more powerful GPUs make it into mobile devices, it may be possible to do the fingerprint creation on the device itself, reducing the time between taking a photo and getting relevant results back.

The image recognition technology is currently being used by Ebay Motors in the US, UK, and Germany. Cortexica hopes to find a home with many of the fashion companies that would use the technology to allow people to identify and ultimately purchase clothing they take photos of on television or in public. The technology can also perform 360-degree object recognition, identify logos that are as small as .4% of the screen, and identify videos. In the future Cortexica hopes to reduce latency, improve recognition accuracy, and add more search categories. Cortexica is also working on enabling an "always on" mobile device that will constantly be indentifying everything around it, which is both cool and a bit creepy. With mobile chips like Logan and Parker coming in the future, Cortexica hopes to be able to do on-device image recognition, which would greatly reduce latency and allow the use of the recognition technology while not connected to the internet.

The number of photos taken is growing rapidly, where as many as 10% of all photos stored "in the cloud" were taken last year alone. Even Facebook, with it's massive data centers is moving to a cold-storage approach to save money on electricity costs of storing and serving up those photos. And while some of these photos have relevant meta data, the majority of photos taken do not, and Cortexica claims that its technology can be used to get around that issue, but identifying photos as well as finding similar photos using its algorithms.

Stay tuned to PC Perspective for more GTC coverage!

Additional slides are available after the break:

GTC 2013: Pedraforca Is A Power Efficient ARM + GPU Cluster For Homogeneous (GPU) Workloads

Subject: General Tech, Graphics Cards | March 20, 2013 - 01:47 PM | Tim Verry
Tagged: tesla, tegra 3, supercomputer, pedraforca, nvidia, GTC 2013, GTC, graphics cards, data centers

There is a lot of talk about heterogeneous computing at GTC, in the sense of adding graphics cards to servers. If you have HPC workloads that can benefit from GPU parallelism, adding GPUs gives you computing performance in less physical space, and using less power, than a CPU only cluster (for equivalent TFLOPS).

However, there was a session at GTC that actually took things to the opposite extreme. Instead of a CPU only cluster or a mixed cluster, Alex Ramirez (leader of Heterogeneous Architectures Group at Barcelona Supercomputing Center) is proposing a homogeneous GPU cluster called Pedraforca.
Pedraforca V2 combines NVIDIA Tesla GPUs with low power ARM processors. Each node is comprised of the following components:

• 1 x Mini-ITX carrier board
• 1 x Q7 module (which hosts the ARM SoC and memory)
  • Current config is one Tegra 3 @ 1.3GHz and 2GB DDR2
• 1 x NVIDIA Tesla K20 accelerator card (1170 GFLOPS)
• 1 x InfiniBand 40Gb/s card (via Mellanox ConnectX-3 slot)
• 1 x 2.5" SSD (SATA 3 MLC, 250GB)

The ARM processor is used solely for booting the system and facilitating GPU communication between nodes. It is not intended to be used for computing. According to Dr. Ramirez, in situations where running code on a CPU would be faster, it would be best to have a small number of Intel Xeon powered nodes to do the CPU-favorable computing, and then offload the parallel workloads to the GPU cluster over the InfiniBand connection (though this is less than ideal, Pedraforca would be most-efficient with data-sets that can be processed solely on the Tesla cards).

While Pedraforca is not necessarily locked to NVIDIA's Tegra hardware, it is currently the only SoC that meets their needs. The system requires the ARM chip to have PCI-E support. The Tegra 3 SoC has four PCI-E lanes, so the carrier board is using two PLX chips to allow the Tesla and InfiniBand cards to both be connected.

The researcher stated that he is also looking forward to using NVIDIA's upcoming Logan processor in the Pedraforca cluster. It will reportedly be possible to upgrade existing Pedraforca clusters with the new chips by replacing the existing (Tegra 3) Q7 module with one that has the Logan SoC when it is released.

Pedraforca V2 has an initial cluster size of 64 nodes. While the speaker was reluctant to provide TFLOPS performance numbers, as it would depend on the workload, with 64 Telsa K20 cards, it should provide respectable performance. The intent of the cluster is to save power costs by using a low power CPU. If your sever kernel and applications can run on GPUs alone, there are noticeable power savings to be had by switching from a ~100W Intel Xeon chip to a lower-power (approximately 2-3W) Tegra 3 processor. If you have a kernel that needs to run on a CPU, it is recommended to run the OS on an Intel server and transfer just the GPU work to the Pedraforca cluster. Each Pedraforca node is reportedly under 300W, with the Tesla card being the majority of that figure. Despite the limitations, and niche nature of the workloads and software necessary to get the full power-saving benefits, Pedraforca is certainly an interesting take on a homogeneous server cluster!

In another session relating to the path to exascale computing, power use in data centers was listed as one of the biggest hurdles to getting to Exaflop-levels of performance, and while Pedraforca is not the answer to Exascale, it should at least be a useful learning experience at wringing the most parallelism out of code and pushing GPGPU to the limits. And that research will help other clusters use the GPUs more efficiently as researchers explore the future of computing.

The Pedraforca project built upon research conducted on Tibidabo, a multi-core ARM CPU cluster, and CARMA (CUDA on ARM development kit) which is a Tegra SoC paired with an NVIDIA Quadro card. The two slides below show CARMA benchmarks and a Tibidabo cluster (click on image for larger version).

Stay tuned to PC Perspective for more GTC 2013 coverage!

GTC 2013: TYAN Launches New HPC Servers Powered by Kepler-based Tesla Cards

Subject: General Tech, Graphics Cards | March 19, 2013 - 06:52 PM | Tim Verry
Tagged: GTC 2013, tyan, HPC, servers, tesla, kepler, nvidia

Server platform manufacturer TYAN is showing off several of its latest servers aimed at the high performance computing (HPC) market. The new servers range in size from 2U to 4U chassis and hold up to 8 Kepler-based Tesla accelerator cards. The new product lineup consists of two motherboards and three bare-bones systems. The S7055 and S7056 are the motherboards while the FT77-B7059, TA77-B7061, and FT48-B7055.

The TA77-B7061 is the smallest system, with support for two Intel Xeon E5-2600 processors and four Kepler-based Tesla accelerator cards. The FT48-B7055 has si7056 specifications but is housed in a 4U chassis. Finally, the FT77-B7059 is a 4U system with support for two Intel Xeon E5-2600 processors, and up to eight Tesla accelerator cards. The S7055 supports a maximum of 4 GPUs while the S7056 can support two Tesla cards, though these are bare boards so you will have to supply your own cards, processors, and RAM (of course).

According to TYAN, the new Kepler-based HPC systems will be available in Q2 2013, though there is no word on pricing yet.

Stay tuned to PC Perspective for further GTC 2013 Coverage!

GTC 2013: Jen-Hsun Huang Takes the Stage to Discuss NVIDIA's Future, New Hardware

Subject: General Tech, Graphics Cards | March 19, 2013 - 02:55 PM | Tim Verry
Tagged: unified virtual memory, ray tracing, nvidia, GTC 2013, grid vca, grid, graphics cards

Today, NVIDIA's CEO Jen-Hsun Huang stepped on stage to present the GTC keynote. In the presentation (which was live streamed on the GTC website and archived here.), NVIDIA discussed five major points, looking back over 2013 and into the future of its mobile and professional products. In addition to the product roadmap, NVIDIA discussed the state of computer graphics and GPGPU software. Remote graphics and GPU virtualization was also on tap. Finally, towards the end of the Keynote, the company revealed its first appliance with the NVIDIA GRID VCA. The culmination of NVIDIA's GRID and GPU virtualization technology, the VCA is a device that hosts up to 16 virtual machines which each can tap into one of 16 Kepler-based graphics processors (8 cards, 16 GPUs per card) to fully hardware accelerate software running of the VCA. Three new mobile Tegra parts and two new desktop graphics processors were also hinted at, with improvements to power efficiency and performance.

On the desktop side of things, NVIDIA's roadmap included two new GPUs. Following Kepler, NVIDIA will introduce Maxwell and Volta. Maxwell will feature a new virtualized memory technology called Unified Virtual Memory. This tech will allow both the CPU and GPU to read from a single (virtual) memory store. Much as with the promise of AMD's Kaveri APU, the Unified Virtual Meory will result in speed improvements in heterogeneous applications because data will not have to be copied to/from the GPU and CPU in order for the data to be processed. Server applications will really benefit from the shared memory tech. NVIDIA did not provide details, but from the sound of it, the CPU and GPU both continue to write to their own physical memory, but their is a layer of virtualized memory on top of that, that will allow the two (or more) different processors to read from each other's memory store.
Following Maxwell, Volta will be a physically smaller chip with more transistors (likely a smaller process node). In addition to the power efficiency improvements over Maxwell, it steps up the memory bandwidth significantly. NVIDIA will use TSV (through silicon via) technology to physically mount the graphics DRAM chips over the GPU (attached to the same silicon substrate electrically). According to NVIDIA, this new TSV-mounted memory will achieve up to 1 Terabytes/second of memory bandwidth, which is a notable increase over existing GPUs.

NVIDIA continues to pursue the mobile market with its line of Tegra chips that pair an ARM CPU, NVIDIA GPU, and SDR modem. Two new mobile chips called Logan and Parker will follow Tegra 4. Both new chips will support the full CUDA 5 stack and OpenGL 4.3 out of the box. Logan will feature a Kepler-based graphics porcessor on the chip that can “everything a modern computer ought to do” according to NVIDIA. Parker will have a yet-to-be-revealed graphics processor (Kepler successor). This mobile chip will utilize 3D FinFET transistors. It will have a greater number of transistors in a smaller package than previous Tegra parts (it will be about the size of a dime), and NVIDIA also plans to ramp up the frequency to wrangle more performance out of the mobile chip. NVIDIA has stated that Logan silicon should be completed towards the end of 2013, with the mobile chips entering production in 2014.

Interestingly, Logan has a sister chip that NVIDIA is calling Kayla. This mobile chip is capable of running ray tracing applications and features OpenGL geometric shaders. It can support GPGPU code and will be compatible with Linux.

NVIDIA has been pushing CUDA for several years, now. The company has seen some respectable adoption rates, by growing from 1 Tesla supercomputer in 2008 to its graphics cards being used in 50 supercomputers, with 500 million CUDA processors on the market. There are now allegedly 640 universities working with CUDA and 37,000 academic papers on CUDA.

Finally, NVIDIA's hinted-at new product announcement was the NVIDIA VCA, which is a GPU virtualization appliance that hooks into the network and can deliver up to 16 virtual machines running independant applications. These GPU accelerated workspaces can be presneted to thin clinets over the netowrk by installing the GRID client software on users' workstations. The specifications of the GRID VCA is rather impressive, as well.

The GRID VCA features:

• 2 x Intel Xeon processors with 16 threads each (32 total threads)
• 192GB to 384GB of system memory
• 8 Kepler-based graphics cards, with two GPUs each (16 total GPUs)
• 16 x GPU-accelerated virtual machines

The GRID VCA fits into a 4U case. It can deliver remote graphics to workstations, and is allegedly fast enough to deliver gpu accelerated software that is equivalent to having it run on the local machine (at least over LAN). The GRID Visual Computing Appliance will come in two flavors at different price points. The first will have 8 Kepler GPUs with 4GB of memory each, 16 CPU threads, and 192GB of system memory for $24,900. The other version will cost $34,900 and features 16 Kepler GPUs (4GB memory), 32 CPU threads, and 384GB system memory. On top of the hardware cost, NVIDIA is also charging licensing fees. While both GRID VCA devices can support unlimited devices, the licenses cost $2,400 and $4,800 per year respectively.

Overall, it was an interesting keynote, and the proposed graphics cards look to be offering up some unique and necessary features that should help hasten the day of ubiquitous general purpose GPU computing. The Unified Virtual Memory was something I was not expecting, and it will be interesting to see how AMD responds. AMD is already promising shared memory in its Kaveri APU, but I am interested to see the details of how NVIDIA and AMD will accomplish shared memory with dedicated grapahics cards (and whether CrossFire/SLI setups will all have a single shared memory pool)..

Stay tuned to PC Perspective for more GTC 2013 Coverage!

GTC 2013: Prepare for Graphics Overload

Subject: General Tech, Graphics Cards, Mobile, Shows and Expos | March 18, 2013 - 09:10 PM | Scott Michaud
Tagged: GTC 2013, nvidia

We just received word from Tim Verry, our GTC correspondent and news troll, about his first kick at the conference. This... is his story.

Graphics card manufacturer, NVIDIA, is hosting its annual GPU Technology Conference (GTC 2013) in San Jose, California this week. PC Perspective will be roaming the exhibit floor and covering sessions as NVIDIA and its partners discuss upcoming graphics technologies, GPGPU, programming, and a number of other low level computing topics.

The future... is tomorrow!

A number of tech companies will be on site and delivering presentations to show off their latest Kepler-based systems. NVIDIA will deliver its keynote presentation tomorrow for the press, financial and industry analysts, and business partners to provide a glimpse at the green team's roadmap throughout 2013 - and maybe beyond.

We cannot say for certain what NVIDIA will reveal during its keynote; but, since we have not been briefed ahead of time, we are completely free to speculate! I think one certainty is the official launch of the Kepler-based K6000 workstation card; for example. While I do not expect to see Maxwell, we could possibly see a planned refresh of the Kepler-based components with some incremental improvements: I predict power efficiency over performance. Perhaps we will receive a cheaper Titan-like consumer card towards the end of 2013? Wishful thinking on my part? A refresh of its GK104 architecture would be nice to see as well, even if actual hardware will not show up until next year. I expect that NVIDIA will react to whatever plans AMD has to decide whether it is in their interest to match them or not.

I do expect to see more information on GRID and Project SHIELD, however. NVIDIA has reportedly broadened the scope of this year's conference to include mobile sessions: expect Tegra programming and mobile GPGPU goodness to be on tap.

It should be an interesting week of GPU news. Stay tuned to PC Perspective for more coverage as the conference gets underway.

What are you hoping to see from NVIDIA at GTC 2013?

ASUS HD 7970 DirectCU II versus a dual linked Dell 3007WFP

Subject: Graphics Cards | March 18, 2013 - 03:17 PM | Jeremy Hellstrom
Tagged: 2560x1600, amd, hd7970 direct cu 2, asus, dell, 3007WFP

[H]ard|OCP has wanted to publish their review of the ASUS HD 7970 DirectCU II for a while but ran into a compatibility issue during their testing and ended up being a perfect example of what sometimes happens to review sites and enthusiasts on the bleeding edge.  [H] uses a Dell 3007WFP with a resolution of 2560x1600 which necessitates the use of a dual link DVI connection, which cause the issue you can see below.  No other setup seemed to reproduce this problem, even the same monitor on a single link DVI at 1920x1080 or at the higher resolution on Display Port would not display the issue.  So what began as a review of an HD 7970 with some nice extra features from ASUS became a long session of troubleshooting.  Take a read through the review as these cards should be back in stock over the next few months, very likely with a solution to this problem already incorporated.

"Today we have the ASUS HD 7970 DirectCU II strapped to our test bench for your reading pleasure. We will compare it to the AMD Radeon HD 7970 GHz Edition and to the NVIDIA GeForce GTX 680 to determine whether the custom VRMs and DirectCU II cooling solution are the droids you are looking for in your next graphics card purchase."

Here are some more Graphics Card articles from around the web:

Graphics Cards

• PowerColor PCS+ HD7870 GHz Edition 2GB GDDR5 @ LanOC Reviews
• HIS Radeon HD 7850 iPower IceQ Turbo 4GB @ Tweaktown
• ASUS Radeon HD 7850 DirectCU On Ubuntu @ Phoronix
• AMD Radeon HD 7870 GHz Edition vs. Nvidia GeForce GTX 660: frametimes @ Hardware.info
• Powercolor HD 7870 Myst Edition @ Bjorn3D
• AMD vs Nvidia: Radeon 7870 vs GeForce 660 Using Frame Times @ HCW
• Prolimatech MK-26 GPU Cooler Review @ Pro-Clockers
• GELID Icy Vision Rev. 2 VGA Cooler Review @ Hi Tech Legion
• Matrox DS1 review: a practical Thunderbolt dock @ Hardware.info
• Nvidia GeForce GTX TITAN 6 GB @ X-bit Labs
• Nvidia GeForce GTX Titan @ Techspot
• Gigabyte GeForce GTX Titan @ Legion Hardware
• EVGA GeForce GTX 670 FTW 2GB @ Hardware.info
• NVIDIA GTX TITAN vs. SLI & Crossfire @ Hardware Canucks

NVIDIA Allegedly Launching Quadro K6000 GK110 GPU For Professionals

Subject: Graphics Cards | March 8, 2013 - 09:17 AM | Tim Verry
Tagged: quadro, nvidia, kepler, k6000, gk110

Earlier this week, NVIDIA updated its Quadro line of workstation cards with new GPUs with GK104 “Kepler” cores. The updated line introduced four new Kepler cards, but the Quadro 6000 successor was notably absent from the NVIDIA announcement. If rumors hold true, professionals may get access to a K6000 Quadro card after all, and one that is powered by GK110 as well.

According to rumors around the Internet, NVIDIA has reserved its top-end Quadro slot for a GK110-based graphics card. Dubbed the K6000 (and in line with the existing Kepler Quadro cards), the high-end workstation card will feature 13 SMX units, 2,496 CUDA cores, 192 Texture Manipulation Units, 40 Raster Operations Pipeline units, and a 320-bit memory bus. The K6000 card will likely have 5GB of GDDR5 memory, like its Tesla K20 counterpart. Interestingly, this Quadro K6000 graphics card has one less SMX unit than NVIDIA’s Tesla K20X and even NVIDIA’s consumer-grade GTX Titan GPU. A comparison between the rumored K6000 card, the Quadro K5000 (GK104), and other existing GK110 cards is available in the table below. Also, note that the (rumored) K6000 specs put it more in like with the Tesla K20 than the K20X, but as it is the flagship Quadro card I felt it was still fair to compare it to the flagship Telsa and GeForce cards.

The Quadro cards are in an odd situation when it comes to double precision floating point performance. The Quadro K5000 which uses GK104 brings an abysmal 90 GFLOPS of double precision. The rumored GK110-powered Quadro K6000 brings double precision performance up to approximately 1 TFLOPS, which is quite the jump and shows that GK104 really was cut down to focus on gaming performance! Further, the card that the K6000 is replacing in name, the Quadro 6000 (no prefixed K), is based on NVIDIA’s previous-generation Fermi architecture and offers .5152 TFLOPS (515.2 GFLOPS) of double precision performance. On the plus side, users can expect around 3.5 TFLOPS of single precision horsepower, which is a substantial upgrade over Quadro 6000's 1.03 TFLOPS of single precision floating point. For comparison, the GK104-based Quadro K5000 offers 2.1 TFLOPS of single precision. Although it's no full GK110, it looks to be the Quadro card to beat for the intended usage.

Of course, Quadro is more about stable drivers, beefy memory, and single precision than double precision, but it would be nice to see the expensive Quadro workstation cards have the ability to pull double duty, as it were. NVIDIA’s Tesla line is where DP floating point is key. It is just a rather wide gap between the two lineups that the K6000 somewhat closes, fortunately. I would have really liked to see the K6000 have at least 14 SMX units, to match consumer Titan and the Tesla K20X, but rumors are not looking positive in that regard. Professionals should expect to see quite the premium with the K6000 versus the Titan, despite the hardware differences. It will likely be sold for around $3,000.

No word on availability, but the card will likely be released soon in order to complete the Kepler Quadro lineup update.