Doug O’Flaherty (AMD)—Geophysics is one of the earliest adopters of new processors. Only 18 months after the first Opteron there are 12 demos of Opteron-based software and hardware at the SEG. AMD’s $/trace advantage has helped our penetration of seismic processing. The new direct connect chip architecture lets users tune math algorithms.
Isn’t math industry independent?
Yes. But because geophysics is an early adopter, it is pushing the envelope.
We aren’t slow to adopt new technology!
Not when it is non-disruptive. AMD’s core philosophy is customer centric, non-disruptive change. End users shouldn’t even know about changes like low latency and NUMA-aware operating systems. They will notice the possibility of having 64GB system memory systems like HP’s DL 585.
~
Eldad Weiss (Paradigm)—We have just announced Release 2 of Epos, our integration platform and now have several hundred Epos installations. Our solutions bring real productivity and easy to use visualization. Applications are tuned to new hardware, offering 64 bit Linux and high end graphics on a $15k desktop. These provide essentially the same performance as a high end machine. Our 3D Canvas integrates all data—seismics, interpretation, mapping and reservoir. Another key offering is our production tools for automated fault picking including Coherence Cube and the BP Center of Visualization fault propagation technology. We also automate the interconnection of horizons and faults—streamlining the interpretation process.
How do you integrate an OpenWorks or GeoFrame shop?
Interoperability is important but it doesn’t make sense just to integrate features. Customers are ready to buy into a new interpretation platform—a bit like buying a new word processor!
Does Paradigm really offer an integrated environment—a complete solution?
Yes. Last year we completed our integration—it’s no longer wishful thinking!
So you are targeting users of mainstream systems now?
Absolutely! Customers are dissatisfied with current solutions which have their roots in 1980s technology. They are looking for second generation solutions.
That’s not true for Petrel or Geoprobe!
OK—Petrel, Geoprobe and Epos are the new game for companies looking for a best in class solution. Landmark is switching to Geoprobe, SIS to Petrel.
What’s Paradigm’s take on OpenSpirit?
Epos and Schlumberger’s OpenSpirit (OS) are similar products—they can coexist. But the Epos footprint is larger than the current OS offering. There will always be a problem with a ‘standard.’ The reality is that you cannot limit a data model.
~
Paul deGroot (de Groot Bril)—Open dTect now supports 2D and 3D seismic attributes and new dGB-developed tools track faults and horizons simultaneously. The Open dTect has had 4,000 downloads since going open source. dGB is working on a business model of commercial plug-ins such as dip steering, filters, attributes, seismic neural networks and the chimney cube. Seismics can be imported from OpenWorks and GeoFrame. We used to be an OpenSpirit member but found it too costly and complex. For us is is easier and faster to do direct access through the vendor’s APIs.
~
Bill Bartling (SGI)—SGI has developed an interesting metric, the data throughput of a single person at a workstation. In 1993 this was 800kb/month: in 2004, 400GB/month! In this context, PCs and desktops have their limitations. Marathon for example is running Geoprobe with 500GB in memory over multiple graphics pipes with no decimation. SGI systems scale in any direction. Up to 4TB memory, but also to multiple pipes and massive data storage.
But the trend is to COTS hardware…
SGI’s new graphics supercomputer ‘Prism’ combines Linux on Intel, ATI graphics accelerators with SGI’s ‘sauce’ as the differentiator. The aim is to make the technology accessible and to lower the price/performance threshold for SGI technology. Many new and returning software houses are coming from desktop Linux to SGI mostly because of our libraries. These offer seamless scaling to huge addressable memory, large numbers of CPUs and multi-pipe graphics. All in a single operating system instance. In contrast, COTS clusters require an operating system and total commuting architecture for every 2 processors. Numaflex interconnectivity requires less copies of Linux and less graphics cards - the whole system looks like a single PC! VoxelVision’s GigaViz, originally designed for COTS Linux clusters has been ported to the Altix. The old GigaViz used Ethernet interconnect and PC graphics—not really a high-end solution from the graphics standpoint.
Does industry really need 400GB of data?
If you start with a 100GB dataset then create continuity data for co-rendered probes and for AVO etc.—you are easily looking at 400GB of co-rendered data.
~
Nick Weston (Sun Microsystems)—Sun’s AMD Opteron-based machine has been validated by all major upstream ISVs. We can also run Windows on Sun hardware! Petrel now runs on Windows on Sun. You can boot to Windows, Linux (Suse and RedHat) or Solaris. As companies move towards multi-processors and clusters, users are re-discovering the complexity of the old ‘big iron’ solutions. There are many non-trivial management issues associated with clusters, so it makes sense to outsource operations. We now offer CPU ‘power on demand’ from Sun-owned and operated centers with costs as low as $1/cpu/hour. Atos Origin is an early adopter in our Dallas data center, offering seismic processing of 2TB datasets. Data and application are encrypted. Processing is carried out remotely and data is de-encrypted at the client end. Another project with the University of Austin TACC involves a 100 CPU machine, with ½ TB RAM running big models for interpretation. Utility computing is also available for large scale visualization. This is especially interesting as the ROI of an in-house visualization center has never been great because of under-use—making it a good candidate for outsourcing.
© Oil IT Journal - all rights reserved.