1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA Pioneers New Standard for HPC With Tesla GPUs Built on Kepler Architecture

Discussion in 'News' started by btarunr, May 15, 2012.

  1. btarunr

    btarunr Editor & Senior Moderator Staff Member

    Joined:
    Oct 9, 2007
    Messages:
    28,537 (11.25/day)
    Thanks Received:
    13,642
    Location:
    Hyderabad, India
    NVIDIA today unveiled a new family of Tesla GPUs based on the revolutionary NVIDIA Kepler GPU computing architecture, which makes GPU-accelerated computing easier and more accessible for a broader range of high performance computing (HPC) scientific and technical applications.

    The new NVIDIA Tesla K10 and K20 GPUs are computing accelerators built to handle the most complex HPC problems in the world. Designed with an intense focus on high performance and extreme power efficiency, Kepler is three times as efficient as its predecessor, the NVIDIA Fermi architecture, which itself established a new standard for parallel computing when introduced two years ago.

    [​IMG] [​IMG] [​IMG]

    "Fermi was a major step forward in computing," said Bill Dally, chief scientist and senior vice president of research at NVIDIA. "It established GPU-accelerated computing in the top tier of high performance computing and attracted hundreds of thousands of developers to the GPU computing platform. Kepler will be equally disruptive, establishing GPUs broadly into technical computing, due to their ease of use, broad applicability and efficiency."

    The Tesla K10 and K20 GPUs were introduced at the GPU Technology Conference (GTC), as part of a series of announcements from NVIDIA, all of which can be accessed in the GTC online press room.

    NVIDIA developed a set of innovative architectural technologies that make the Kepler GPUs high performing and highly energy efficient, as well as more applicable to a wider set of developers and applications. Among the major innovations are:
    • SMX Streaming Multiprocessor -- The basic building block of every GPU, the SMX streaming multiprocessor was redesigned from the ground up for high performance and energy efficiency. It delivers up to three times more performance per watt than the Fermi streaming multiprocessor, making it possible to build a supercomputer that delivers one petaflop of computing performance in just 10 server racks. SMX's energy efficiency was achieved by increasing its number of CUDA architecture cores by four times, while reducing the clock speed of each core, power-gating parts of the GPU when idle and maximizing the GPU area devoted to parallel-processing cores instead of control logic.
    • Dynamic Parallelism -- This capability enables GPU threads to dynamically spawn new threads, allowing the GPU to adapt dynamically to the data. It greatly simplifies parallel programming, enabling GPU acceleration of a broader set of popular algorithms, such as adaptive mesh refinement, fast multipole methods and multigrid methods.
    • Hyper-Q -- This enables multiple CPU cores to simultaneously use the CUDA architecture cores on a single Kepler GPU. This dramatically increases GPU utilization, slashing CPU idle times and advancing programmability. Hyper-Q is ideal for cluster applications that use MPI.
    "We designed Kepler with an eye towards three things: performance, efficiency and accessibility," said Jonah Alben, senior vice president of GPU Engineering and principal architect of Kepler at NVIDIA. "It represents an important milestone in GPU-accelerated computing and should foster the next wave of breakthroughs in computational research."

    NVIDIA Tesla K10 and K20 GPUs
    The NVIDIA Tesla K10 GPU delivers the world's highest throughput for signal, image and seismic processing applications. Optimized for customers in oil and gas exploration and the defense industry, a single Tesla K10 accelerator board features two GK104 Kepler GPUs that deliver an aggregate performance of 4.58 teraflops of peak single-precision floating point and 320 GB per second memory bandwidth.

    The NVIDIA Tesla K20 GPU is the new flagship of the Tesla GPU product family, designed for the most computationally intensive HPC environments. Expected to be the world's highest-performance, most energy-efficient GPU, the Tesla K20 is planned to be available in the fourth quarter of 2012.

    The Tesla K20 is based on the GK110 Kepler GPU. This GPU delivers three times more double precision compared to Fermi architecture-based Tesla products and it supports the Hyper-Q and dynamic parallelism capabilities. The GK110 GPU is expected to be incorporated into the new Titan supercomputer at the Oak Ridge National Laboratory in Tennessee and the Blue Waters system at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign.

    "In the two years since Fermi was launched, hybrid computing has become a widely adopted way to achieve higher performance for a number of critical HPC applications," said Earl C. Joseph, program vice president of High-Performance Computing at IDC. "Over the next two years, we expect that GPUs will be increasingly used to provide higher performance on many applications."

    Preview of CUDA 5 Parallel Programming Platform
    In addition to the Kepler architecture, NVIDIA today released a preview of the CUDA 5 parallel programming platform. Available to more than 20,000 members of NVIDIA's GPU Computing Registered Developer program, the platform will enable developers to begin exploring ways to take advantage of the new Kepler GPUs, including dynamic parallelism.

    The CUDA 5 parallel programming model is planned to be widely available in the third quarter of 2012. Developers can get access to the preview release by signing up for the GPU Computing Registered Developer program on the CUDA website.
     
    Recus, 1c3d0g, erixx and 4 others say thanks.
  2. xrealm20

    xrealm20 New Member

    Joined:
    Jan 29, 2010
    Messages:
    467 (0.28/day)
    Thanks Received:
    129
    Location:
    Houston
    Impressive. Wonder how many cores this one is carrying. 10 Racks for a 1P/flop HPC cluster. I can see a lot of research companies loving this...
     
  3. Crap Daddy

    Crap Daddy

    Joined:
    Oct 29, 2010
    Messages:
    2,739 (1.93/day)
    Thanks Received:
    1,044
    K10 is GK104x2... GTX690. K20 will be the Big K... later this year.
     
  4. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    3,293 (1.89/day)
    Thanks Received:
    1,513
    Location:
    Glasgow - home of formal profanity
    K10 is two GK104's, giving 320 GB/s.

    K20 is the daddy. Ladies and gentlemen, GK110 has arrived. What are it's specs? EDIT: http://www.nvidia.com/content/tesla/pdf/NV_DS_TeslaK_Family_May_2012_LR.pdf

    And also... :cry:

    It's as Crap Daddy's been hinting at, it's a Tesla part first and foremost - coming to TESLA line in Q4 2012.

    (beat me to it Crap Daddy... :laugh:)

    That means GK104 is as good as it gets this gen. Disappointed.
     
    Last edited: May 15, 2012
    Crap Daddy says thanks.
  5. KainXS

    KainXS

    Joined:
    Sep 25, 2007
    Messages:
    5,600 (2.20/day)
    Thanks Received:
    501
    oh , everyhing is TBA on the specs still then
     
  6. TheMailMan78

    TheMailMan78 Big Member

    Joined:
    Jun 3, 2007
    Messages:
    20,977 (7.87/day)
    Thanks Received:
    7,532
    btarunr this news made my manhood move.
     
  7. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    3,293 (1.89/day)
    Thanks Received:
    1,513
    Location:
    Glasgow - home of formal profanity
    Rising from the depths like a nightmare from H.P Lovecraft's writings.
     
  8. Crap Daddy

    Crap Daddy

    Joined:
    Oct 29, 2010
    Messages:
    2,739 (1.93/day)
    Thanks Received:
    1,044
    Well, that's it folks. You know, NV has some business to attend. To be honest during the whole Kepler rumors before GK104 was launched I was thinking that NV has other priorities than discrete GPUs (Tesla, Tegra and such) and partially I am right. They have only one chip, the GK104, which we will see in 3 maybe even 4 variants and that's about all regarding Kepler for gaming. GK106 is nowhere to be seen, I'm starting to doubt that it exists somewhere and GK107 seems to be low end.
     
    erixx and the54thvoid say thanks.
  9. TheMailMan78

    TheMailMan78 Big Member

    Joined:
    Jun 3, 2007
    Messages:
    20,977 (7.87/day)
    Thanks Received:
    7,532
    What? Everyone knows AMD's days are numbered because of the fabled vapor chip! :rolleyes:
     
  10. the54thvoid

    the54thvoid

    Joined:
    Dec 14, 2009
    Messages:
    3,293 (1.89/day)
    Thanks Received:
    1,513
    Location:
    Glasgow - home of formal profanity
    It's a shame. I think even AMD diehards were 'curious' about GK110 and now it's been revealed as a Tesla part. I'm a little underwhelmed. I'm happy with my current card but I wanted to see a 'Big Daddy' Kepler gaming part that wasn't a ridiculous and exceptional dual gpu.

    Oh well. Next rumour frenzy - Sea Islands, AKA HD8xxx.
     
  11. Crap Daddy

    Crap Daddy

    Joined:
    Oct 29, 2010
    Messages:
    2,739 (1.93/day)
    Thanks Received:
    1,044
    Nah, they're doing fine. They must sell heaps of the 7850.
     
  12. TheMailMan78

    TheMailMan78 Big Member

    Joined:
    Jun 3, 2007
    Messages:
    20,977 (7.87/day)
    Thanks Received:
    7,532
    Quiet! They will hear you! Common sense is not so common.
     
  13. Crap Daddy

    Crap Daddy

    Joined:
    Oct 29, 2010
    Messages:
    2,739 (1.93/day)
    Thanks Received:
    1,044
    OK, let's get back on topic. Here's what NV is preaching right now:

    "Kepler is world's first gpu designed for the cloud, to be deployed into cloud data centers worlwide. it does this with:
    --virtualized gpu
    --no longer does it need to connect to a display, it can render and stream instantaneously right out of chip to a remote location
    --super energy efficiency, so it can be deployed in a massive scale

    Every command buffer is now virtualized. we can now discern which virtual machine were to send us a graphics command. at the end, we can stream frame buffer to that spsecific virtual machine. One GPU can be shared with countless users.

    Who's got a GTX680? Care to share some?
     
  14. TheMailMan78

    TheMailMan78 Big Member

    Joined:
    Jun 3, 2007
    Messages:
    20,977 (7.87/day)
    Thanks Received:
    7,532
    Sounds like some epic folding applications could be had.
     
  15. mamisano New Member

    Joined:
    Feb 27, 2007
    Messages:
    43 (0.02/day)
    Thanks Received:
    5
    A few things...

    1- Kind of sad that the K10 has 1/5 the double precision compute power of a 7970.
    2- I can see the K20 being GPU compute only, similar to the Intel MIC cards.
     
  16. eddman New Member

    Joined:
    Mar 9, 2012
    Messages:
    35 (0.04/day)
    Thanks Received:
    10
  17. theoneandonlymrk

    theoneandonlymrk

    Joined:
    Mar 10, 2010
    Messages:
    3,384 (2.05/day)
    Thanks Received:
    565
    Location:
    Manchester uk
    yes indeed , but thats exactly why i feel so slapped about the face by nvidia with the GK104 i personally wanted a 660 with decent folding power , not to have to consider a 560:rolleyes:, they seem to be essetially moveing towards a point where they will start selling only speciallised cards, gamer or folder, but not the two.

    as this increases the profitability of their high end compute cards and closes the door on using cheaper Nv cards in compute intense applications and servers, its Gay:wtf:

    plus they are the biggest money milking tech gets ive seen, they squeeze a third more money out per coin spent then any other co(to be fair in some way a credit to them) but its from the customers:ohwell:
     
  18. TheMailMan78

    TheMailMan78 Big Member

    Joined:
    Jun 3, 2007
    Messages:
    20,977 (7.87/day)
    Thanks Received:
    7,532
    Well.......AMD's 7970 I hear is a nice folder :laugh:
     
  19. Fluffmeister

    Fluffmeister

    Joined:
    Dec 22, 2011
    Messages:
    651 (0.65/day)
    Thanks Received:
    185
    Sounds like smart business to me :D
     
  20. theoneandonlymrk

    theoneandonlymrk

    Joined:
    Mar 10, 2010
    Messages:
    3,384 (2.05/day)
    Thanks Received:
    565
    Location:
    Manchester uk
    smart business , yes ,as i implied.

    Good for me the customer, No no it isnt

    im only interested at all in Nv, for a hybrid physx card + perma folder card(1 off and probably a 560 now), as my next render card isnt out yet;) or for that matter even speculated about yet;) as my main rig is fine at this time(Fx8350 next Up:)).
     
  21. Maban

    Maban

    Joined:
    Mar 6, 2008
    Messages:
    2,398 (1.00/day)
    Thanks Received:
    1,020
    I'm surprised even the Tesla GK104 is locked at 1/24th DP power. That's shameful if you ask me.
     
  22. theoneandonlymrk

    theoneandonlymrk

    Joined:
    Mar 10, 2010
    Messages:
    3,384 (2.05/day)
    Thanks Received:
    565
    Location:
    Manchester uk
    thats why they need two of them on their first next gen compute card lmfao:roll: ,,you buying this im not and wasnt so to them i matter not but this dosnt scream performance crown to me and double its performance (GK110) and what do you get,,,thats right, it again but finally on 1 chip,,,, Epic Fail imho, though yes they will be economical, just shit.

    this shines with the Failness of a billion suns
     
  23. HalfAHertz

    HalfAHertz

    Joined:
    May 4, 2009
    Messages:
    1,888 (0.96/day)
    Thanks Received:
    378
    Location:
    Singapore
    It's not locked it - it's how it was designed. They reduced the number of advanced functional units in each SM in favor of more simpler ones to drive SP FP performance (and thus gaming) while reducing the power requirements.

    GK104 consists of 4 blocks, but only one of the four can do DP FP calcs. From AT:

    Looks to me like it was never meant to be a number crunching beast.
     
    Last edited: May 16, 2012
    1c3d0g and Maban say thanks.
  24. Maban

    Maban

    Joined:
    Mar 6, 2008
    Messages:
    2,398 (1.00/day)
    Thanks Received:
    1,020
    Thanks. I went over and read a little on that page. The world makes sense now.


    But still disappointing.
     
  25. Protagonist

    Protagonist

    Joined:
    Sep 7, 2010
    Messages:
    718 (0.49/day)
    Thanks Received:
    123
    Intel MIC

    So Intel MIC is supposed to rival this, i see the back panel looks like the one on Intel MIC no display outputs
     

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)

Share This Page