Physical Architecture

The physical architecture of Titan is just as interesting as the high level core and transistor counts. I mentioned earlier that Titan is built from 200 cabinets. Inside each cabinets are Cray XK7 boards, each of which has four AMD G34 sockets and four PCIe slots. These aren't standard desktop PCIe slots, but rather much smaller SXM slots. The K20s NVIDIA sells to Cray come on little SXM cards without frivolous features like display outputs. The SXM form factor is similar to the MXM form factor used in some notebooks.

There's no way around it. ORNL techs had to install 18,688 CPUs and GPUs over the past few weeks in order to get Titan up and running. Around 10 of the formerly-Jaguar cabinets had these new XK boards but were using Fermi GPUs. I got to witness one of the older boards get upgraded to K20. The process isn't all that different from what you'd see in a desktop: remove screws, remove old card, install new card, replace screws. The form factor and scale of installation are obviously very different, but the basic premise remains.

As with all computer components, there's no guarantee that every single chip and card is going to work. When you're dealing with over 18,000 computers as a part of a single entity, there are bound to be failures. All of the compute nodes go through testing, and faulty hardware swapped out, before the upgrade is technically complete.

OS & Software

Titan runs the Cray Linux Environment, which is based on SUSE 11. The OS has to be hardened and modified for operation on such a large scale. In order to prevent serialization caused by interrupts, Cray takes some of the cores and uses them to run all of the OS tasks so that applications running elsewhere aren't interrupted by the OS.

Jobs are batch scheduled on Titan using Moab and Torque.


If you're curious about why Titan uses Opterons, the explanation is actually pretty simple. Titan is a large installation of Cray XK7 cabinets, so CPU support is actually defined by Cray. Back in 2005 when Jaguar made its debut, AMD's Opterons were superior to the Intel Xeon alternative. The evolution of Cray's XT/XK lines simply stemmed from that point, with Opteron being the supported CPU of choice.

The GPU decision was just as simple. NVIDIA has been focusing on non-gaming compute applications for its GPUs for years now. The decision to partner with NVIDIA on the Titan project was made around 3 years ago. At the time, AMD didn't have a competitive GPU compute roadmap. If you remember back to our first Fermi architecture article from back in 2009, I wrote the following:

"By adding support for ECC, enabling C++ and easier Visual Studio integration, NVIDIA believes that Fermi will open its Tesla business up to a group of clients that would previously not so much as speak to NVIDIA. ECC is the killer feature there."

At the time I didn't know it, but ORNL was one of those clients. With almost 19,000 GPUs, errors are bound to happen. Having ECC support was a must have for GPU enabled Jaguar and Titan compute nodes. The ORNL folks tell me that CUDA was also a big selling point for NVIDIA.

Finally, some of the new features specific to K20/GK110 (e.g. Hyper Q and GPU Direct) made Kepler the right point to go all-in with GPU compute.

Power Delivery & Cooling

Titan's cabinets require 480V input to reduce overall cable thickness compared to standard 208V cabling. Total power consumption for Titan should be around 9 megawatts under full load and around 7 megawatts during typical use. The building that Titan is housed in has over 25 megawatts of power delivered to it.

In the event of a power failure there's no cost effective way to keep the compute portion of Titan up and running (remember, 9 megawatts), but you still want IO and networking operational. Flywheel based UPSes kick in, in the event of a power interruption. They can power Titan's network and IO for long enough to give diesel generators time to come on line.

The cabinets themselves are air cooled, however the air itself is chilled using liquid cooling before entering the cabinet. ORNL has over 6600 tons of cooling capacity just to keep the recirculated air going into these cabinets cool.

Oak Ridge National Laboratory Applying for Time on Titan & Supercomputing Applications


View All Comments

  • mayankleoboy1 - Wednesday, October 31, 2012 - link

    Is there any scope of a FPGA or a group of FPGA's that replace standard algorithms with hardware implementations ?

    Example : Fourier transforms, matrix multiplication.
  • prashanth07 - Wednesday, October 31, 2012 - link

    Yes, there is significant research going on. In our lab we had a pretty big group working of using FPGAs for HPC. The RC based supercomputer is called Novo-G. It was the worlds biggest publicly known RC super computer.

    It is very small in physical size compared to some of the top conventional super computer, but for some specific compute requirements it comes close to beating top supercomputers. There was a major upgrade planned (around the time I was graduating) so it might even better now.
    What exact type of computations? I don't remember very well (I didn't work on RC, I was mostly s/w guy in conventional HPC part of lab), you might be able to get some info by checking out few posters or papers abstract.

    See: (very outdated, We didn't had anyone updating this site regularly) (very abstract, low on specific info)
  • Guspaz - Wednesday, October 31, 2012 - link

    Just think, if Moore's law holds for another few decades, you'll see this performance in a smartphone in 20-30 years... Reply
  • Montrey - Saturday, November 03, 2012 - link

    According to the paper, it takes 6 to 8 years for the #1 computer on the list to move to #500, and then another 8 to 10 years for that performance to be available in your average notebook computer. Not sure on notebook to smartphone, but it can't be very long.
  • Doh! - Wednesday, October 31, 2012 - link

    This kind of article keeps me coming back to Awesome stuff. Reply
  • bl4C - Wednesday, October 31, 2012 - link

    indeed, i was thinking:
    "now THIS is an article"

    great, thx !
  • gun_will_travel - Wednesday, October 31, 2012 - link

    With all the settings turned up. Reply
  • dragonsqrrl - Wednesday, October 31, 2012 - link

    Anand, I just want to confirm the core count on the Tesla K20. So this means one of the 15 SMX blocks is disabled on the K20? Reply
  • Ryan Smith - Wednesday, October 31, 2012 - link

    We're basing our numbers off of the figures published by HPCWire.

    For a given clockspeed of 732MHz and DP performance of 1.3TFLOPs, it has to be 14 SMXes. The math doesn't work for anything else.
  • RussianSensation - Wednesday, October 31, 2012 - link

    The article only states a range for DP of 1.2-1.3 Tflops.

    The specification could be 705mhz with GPU boost to 732mhz x 2496 CUDA cores ~ 1.22 Tflops

    Not saying it can't be 2688 CUDA cores but you are using the high-end of the range when the article clearly lists a range of 1.2-1.3Tflops. I don't think you can just assume that it's 2688 without a confirmation given the range of values provided.

Log in

Don't have an account? Sign up now