Brian's Place

SC2011 - Seattle

2011-11-15T13:03:00.000-08:00

Annual pilgrimage to SC currently underway. This year a large presence for WestGrid in the Compute Canada booth, with an IQStation running 3D demos for the week. Spend the first three days setting up the hardware and software with WestGrid colleagues... A bit of a job, but demo was ready to go for the Gala Opening last night. Had lots of interest at the Gala opening with many people interested in the demos...

GPU Ray Tracing - nVidia tech talk

2011-08-08T17:17:00.000-07:00

GPU ray tracing is available as a main stream tool. Five years ago considered "the future" but impossible on GPU (challenge/claim made to GPU community). NVidia solution is Optix, a library API for abstracting the RT core from the hardware. There are a bunch of commercial apps for RT on GPU but Optix as an API for developers looks interesting. Beta of new version available soon.

Lunch with the Aussies

2011-08-08T16:07:00.000-07:00

Heard from Kevin Smith and Matt Adcock from my ANU/Australia days. They are also in town for SIGGRAPH and we went to lunch. Talked about old times, caught up with what the Canberra (ANU/CSIRO) gang is up to. Also talked a lot about my PhD research, turns out Matt and one of his colleagues (Viveka Welley) are interested in similar areas. Not too surprising since I worked so closely with them in the "ol' days".

Cory Doctorow on DRM

2011-08-08T12:34:00.000-07:00

Great talk from Cory Doctorow on the evils of DRM. I suspect it will be up on YouTube soon. Cory explicitly gave open license to record his talk, despite the fact that SIGGRAPH has recording restrictions. An interesting community for such a talk.

SIGGRAPH papers fast forward

2011-08-07T19:29:00.000-07:00

Fast forward summarizes all the SIGGRAPH papers with each presenter given one minute to describe their work. I think the speakers spent as much time on their fast forward talks as their paper. Pretty entertaining and lots of great looking talks.

Hot3D panel at HPG 2011

2011-08-07T19:13:00.000-07:00

Interesting panel at the end of HPG 2011. David Luebke (nVidia), Mike Houston and Mike Mantor (AMD), and William Mark (Intel) talked about futures of graphics hardware. Interesting that all speakers talked about the impact of other forces that are driving GPU development (mobile computing/power/sales volumes, getting the GPU closer to the CPU). This, combined with other comments during the conference (did you know that the PS3 gpu is five years old and isn't going to be updated for another couple of years) and the influence of HPC on GPUs makes you wonder if high performance graphics will be the driver of GPU development in the future.

High Performance Graphics (HPG) 2011

2011-08-07T18:33:00.000-07:00

Back in conference mode so better update the blog. Spent the last couple days at HPG 2011, a conference that is running in parallel to SIGGRAPH 2011. Brings back memories of early to mid nineties when I was working in this area. Back to my teapot days as my brother would say. Pretty interesting to see what is going on. In some ways the same stuff (voxels, bounding volume hierarchies, and frames per second) and in some ways very different (20 ms per frame, and coaxing flops from GPUs). Lots of interesting papers!!!

Bonus was that I got to meet with a couple of old colleagues, including Eric Haines (from my Ray Tracing News days circa 1992) and Mark Bolstad (from IEEE Vis circa 1993).

SAGE - Pushing the Envelope, like usual!

2010-11-16T10:39:00.000-08:00

The SAGE BOF was interesting. EVL and the SAGE group pushing the boundaries like usual! Nice work Jason, Luc, Maxine, Sungwon, and Ratko (and the rest of the SAGE community).

Some very impressive streaming results from SARA and some interesting work with CalIT2, CINEGrid, and Disney. Nice integration with ParaView and SAGE to stream parallel ParaView session to SAGE wall. Also, nice discussion of user interface issues in designing for wall based use cases.

Factors:

Thin bezels important to usability.
Reachable for interaction.
Visiblity for large audiences.
Mobility

Touch screens - good for detail, localized interaction, tiring for longer work.
Gyromouse - standing but no need to walk back and forth

Comfort - use of laptops, sitting around, interacting with desktops
Video conferencing

Mapping content between remote sites

Touch screen interface over very large tiled display - VERY nice!

Argonne and nVidia booth on demo

Definitely worth going to the booths for Kaust (Booth 2247), SARA, Argonne (Booth 2513) and others to see the SAGE demos.

Whats Next? SAGENext - Single node SAGE portals makes it usable to MANY users! Enabling cloud applications such as Google Services (like anything cloud, I am not quite sure what this means 8-)... Coming to you for SC 2011!

What's in a name - Ultrascale Visualization

2010-11-15T07:54:00.000-08:00

The 5th Ultrascale visualization workshop at SC 2010 is focused on visualization at the exascale. What a great name - Ultrascale Visualization is always current - visualization of data at the largest scale and never requiring a name change. The key question: How do you deal with exascale computation when it is too expensive (in terms of time and power) to move the raw data off of the machine. Is in-situ visualization the way of the future?

Jazz on Bourbon Street

2010-11-14T14:20:00.001-08:00

I suppose one can't spend time in New Orleans without mentioning the great Jazz and crazy atmosphere on Bourbon Street. Arrived late Saturday night but managed to find a great live band, playing for free in a small plaza just off Bourbon Street. Great tunes!

Paraview Coporcessing Library

2010-11-14T14:15:00.000-08:00

Paraview now has a co-processing library, designed to allow computational codes to perform "in-situ" visualization "easily". I didn't sit through this tutorial in it entirety, but it looks like an interesting capability for those that want to visualize without writing data to disk and/or transferring data to a local visualization machine.

Still requires the author of the computational code to transform data to VTK objects, but given the fact that data sets are becoming larger we may see this more and more.

CUDA Tutorial

2010-11-14T14:09:00.000-08:00

CUDA Programming + Optimization - my head hurts!!!

Blog is back - SC 2010 in New Orleans

2010-11-14T14:04:00.000-08:00

I'm backkkkk... This time at SC 2010 in New Orleans.

First things first, here is a list of the activities that I think are interesting at SC this year. Of course none of this includes stuff on the exhibition floor. Not sure how I will fit it all in!

Al Gore - Speaking at SC09

2009-11-19T07:34:00.000-08:00

Vice President Al Gore is speaking at SC this morning. Got here early to make sure I got a seat - lots of seats left at the moment, but its going to fill up fast I suspect...

A quiet SC 09 show floor?

2009-11-18T09:14:00.000-08:00

It is hard to ever describe the SC show floor as quiet, but all things are relative. The overall feel I get from the show floor this year is that things are much more sedate this year, with less people in the booths and less technology on the floor. This seems to be particularly true across my domain of interest, visualization. None of the system vendors have a large visualization presence on the show floor, with only HP having a "visualization pod". The biggest company presence in visualization is the nVidia booth (not too surprising), and they do have a lot of nice technology on show. Comparably, the ATI presence within the AMD booth (WRT visualization at least) is relatively small.

The "big" academic players also seem to have scaled down booths, with what appears to be less people and less technology. It is hard to quantify if this is really true, but it is certainly my initial feeling... From a visualization perspective, I was expecting lots of high res displays (4K, tiled displays), high res visualizations, visualization clusters, and high resolution video streaming. Certainly, this all exists at SC this year, but my guess is that there is less on the show floor this year than there was last year.

I always come to SC looking for the "next big thing", and quite often I come away with some ideas and thoughts in this direction. This year, so far, not so much... Two more days to go - we will see.

Coast-to-Coast live at SC 09 in Portland

2009-11-18T09:02:00.000-08:00

Richard Vaughn's Coast-to-Coast (C2C) seminar was seen live at the Compute Canada and CANARIE booth at SC 2009 in Portland. It was timely that the C2C seminar was being held during a time slot that did not have any demos in the booth, so we put it up on the big screen. An impromptu live research seminar on "World-mediated robot intelligence", attended live by 15 sites and 50 people across Canada, not including the audience that attended at SC09...

3D Web Future of HPC?

2009-11-17T09:08:00.000-08:00

Justin Ratner, CTO of Intel talked about the 3D web and the future of HPC. Hypothesis was that the HPC industry needs to step outside the box and not rely only on the big science funding model. He mentions three fundamental aspects of what he describes as the 3D web:

Continually simulated
Multi-view animated
Immersive and collaborative

Some key issues and technologies mentioned:

ScienceSim (www.sciencesim.org)
- 3D web platform for scientific simulation
- Based on OpenSim (www.opensimulator.org)
Demonstrated/discussed several interesting applications

Bio-diversity and fern population distribution
Fashion design - cloth simulation

Discussed several computational challenges in physical simulations
- Physically accurate cloth simulation
- Physically accurate sound generation
Interoperability is key
Merging of HPC and Cloud worlds
Architecture - collision of CPU and GPU

Larrabee

Computational co-processor
Many cores, many threads
Memory management
Showed actual code running on hardware

Ct - High-level data parallel programming (RapidMind)
CPU-GPU interaction inefficiency - copy over PCI-Exp

Shared memory between Xeon and Larrabee
Mine-yours-ours (MYO) sharing
Demo of physics simulation (soft body dynamics) running on Larrabee using MYO sharing

SGEMM performance - overclocked Larrabee 1006 GFlops - we think it was real hardware!!!

ParaView/Sage Integration

2009-11-16T09:43:00.000-08:00

EVL and TACC presented an interesting integration of ParaView with Sage. Takes advantage of ParaView's parallel rendering capabilities and SAGE's ability to stream visualization to very high resolution tiled displays. Remote rendering of ParaView applications, rendered at TACC and displayed on the EVL tiled display.

nVidia CUDA - Fermi chipset

2009-11-15T09:22:00.000-08:00

A couple of interesting features on the Fermi chipset from nVidia.

512 CUDA cores
ECC memory available on Fermi - This is an important one...
ECC can be turned on or off
- I wonder if this has a performance impact?
- If it didn't, why wouldn't it just be turned on all the time???
8x double precision performance increase
IEEE 754-2008: Fused multiply-add for single/double precision support

Welcome to my new blog

2009-11-15T09:16:00.000-08:00

Hello,

Welcome to my new blog... I decided to create this as a mechanism to easily post info about the cool stuff that I see at the SC09 conference. We will see where it goes from there...