NVIDIA Launches the Tesla K40 GPU Accelerator

[Cristian_25H]

NVIDIA today unveiled the NVIDIA Tesla K40 GPU accelerator, the world's highest performance accelerator ever built, delivering extreme performance to a widening range of scientific, engineering, high performance computing (HPC) and enterprise applications.

Providing double the memory and up to 40 percent higher performance than its predecessor, the Tesla K20X GPU accelerator, and 10 times higher performance than today's fastest CPU, the Tesla K40 GPU is the world's first and highest-performance accelerator optimized for big data analytics and large-scale scientific workloads.





Featuring intelligent NVIDIA GPU Boost technology, which converts power headroom into a user-controlled performance boost, the Tesla K40 GPU accelerator enables users to unlock the untapped performance of a broad range of applications.

"GPU accelerators have gone mainstream in the HPC and supercomputing industries, enabling engineers and researchers to consistently drive innovation and scientific discovery," said Sumit Gupta, general manager of Tesla Accelerated Computing products at NVIDIA. "With the breakthrough performance and higher memory capacity of the Tesla K40 GPU, enterprise customers can quickly crunch through massive volumes of data generated by their big data analytics applications."

Ultimate Performance for Science, Big Data
Based on the NVIDIA Kepler compute architecture -- the highest performance, most efficient architecture ever built -- the Tesla K40 GPU accelerator surpasses all other accelerators on two common measures of computational performance: 4.29 teraflops single-precision and 1.43 teraflops double-precision peak floating point performance.

Key features of the Tesla K40 GPU accelerator include:

• 12 GB of ultra-fast GDDR5 memory allows users to process 2X larger datasets, enabling them to rapidly analyze massive volumes of data.
• 2,880 CUDA parallel processing cores deliver application acceleration by up to 10X compared to using a CPU alone.
• Dynamic Parallelism enables GPU threads to dynamically spawn new threads, enabling users to quickly and easily crunch through adaptive and dynamic data structures.
• PCIe Gen-3 interconnect support accelerates data movement by 2X compared to PCIe Gen-2 technology.

In a related announcement, the Texas Advanced Computing Center (TACC) at The University of Texas at Austin -- one of the leading advanced computing centers in the United States -- plans to deploy "Maverick," a new interactive, remote visualization and data analysis system powered by NVIDIA Tesla K40 GPU accelerators. Maverick is expected to be fully operational in January 2014.

"The Tesla K40 GPU accelerators will help researchers crunch through massive volumes of big data and gain new insights through large-scale, sophisticated visualizations," said Kelly Gaither, director of Visualization at TACC. "With NVIDIA GPUs, Maverick will provide researchers powerful interactive capabilities to advance their most complex scientific challenges."

he Tesla K40 GPU accelerates the broadest range of scientific, engineering, commercial and enterprise HPC and data center applications. Today, more than 240 software applications take advantage of GPU acceleration. The complete catalog of GPU-accelerated applications is available as a free download.

More information about the Tesla K40 GPU accelerator is available at NVIDIA booth 613 at SC13, Nov. 18-21, and on the NVIDIA high performance computing website. To learn more about CUDA or download the latest version, visit the CUDA website.

Users can also try the Tesla K40 GPU accelerator for free on remotely hosted clusters. Visit the GPU Test Drive website for more information.

Availability
Shipping today, the NVIDIA Tesla K40 GPU accelerator is available now and in the coming months from a variety of server manufacturers, including Appro, ASUS, Bull, Cray, Dell, Eurotech, HP, IBM, Inspur, SGI, Sugon, Supermicro and Tyan, as well as from NVIDIA reseller partners.

View at TechPowerUp Main Site

 

[librin.so.1]

clocks pls

 

[xvi]

clocks pls

http://www.anandtech.com/show/7521/nvidia-launches-tesla-k40

tl;dr
Stock: 745MHz
Boost (doesn't work like you think, apparently): 810MHz, 875MHz

 

[librin.so.1]

So in other words, the K40 is to the K20X as what the GTX 780ti is to the GTX TITAN, hardware-wise?
WOW. Not impressed much

 

[The Von Matrices]

Well, there is one thing interesting about this. This card is the first time I've seen 4Gbit GDDR5 chips used. Get ready for another doubling of graphics memory next video card generation.

 

[xvi]

So in other words, the K40 is to the K20X as what the GTX 780ti is to the GTX TITAN, hardware-wise?
WOW. Not impressed much

Exactly.

 

[HumanSmoke]

So in other words, the K40 is to the K20X as what the GTX 780ti is to the GTX TITAN, hardware-wise?

Not really.
The GTX 780 Ti has half the onboard memory of the GTX Titan
The Tesla K40 has twice as much onboard memory of the K20X

WOW. Not impressed much

Based on your erroneous assumption and the fact that you aren't the target demographic for the board I am not in the least surprised.
There is a "contact us" tab at the Eurotech site should you wish to give these supercomputer manufacturers the benefit of your experience.

 

[librin.so.1]

Not really.
The GTX 780 Ti has half the onboard memory of the GTX Titan
The Tesla K40 has twice as much onboard memory of the K20X

I know this. I simply decided to handwave this fact out in my comparison. As I was pretty much only taking the GPU itself for that comparison.

Based on your erroneous assumption and the fact that you aren't the target demographic for the board I am not in the least surprised.

Well, the way they presented it, it felt as it was supposed to be something heaps faster than the previous top Tesla product. They said it as if it was so fast, it would be more than enough to revolutionize computing. But meh, it's just marginally better [not counting the 2x memory amount]. Shame on them for getting me too excited for a moment there.

 

[The Von Matrices]

Well, the way they presented it, it felt as it was supposed to be something heaps faster than the previous top Tesla product. They said it as if it was so fast, it would be more than enough to revolutionize computing. But meh, it's just marginally better [not counting the 2x memory amount]. Shame on them for getting me too excited for a moment there.

Have you not read press releases before? Even the most minor of improvements is "revolutionary" if it's in a press release.

 

[Death Star]

My primary research focus includes a ton of computational electrodynamics work (mainly FDTD in OpenCL) and while the raw crunching power of the GPU is only a moderate upgrade the 12GB of RAM almost justifies the price tag. It's VERY easy to fill up 6GB of RAM, so for sims requiring more RAM you need to get creative with memory management to avoid the large overhead from excess PCIe transfers. 12GB is also pretty easy to fill up, but it gives quite a bit of extra headroom to more creatively manage host-device transfers. Storing the relevant information of a sim across both GPU and system memory can cripple performance by an order of magnitude or more

I'd kill for even a mid-range gaming GPU with 16GB+ of RAM...

 

[beck24]

K40

This is a nice chunk more performance for those special users and to keep the same TDP is impressive. GPGPU is becoming a much more mainstream tech and Nvidia is at the forefront.