Sites and Affiliations
Research and Applications
Major Accomplishments
Frequently Asked Questions
Knowledge and Technology Transfer
Education and Outreach
Media Resources
Technical Reports and Publications
Parallel Computing Research - Our Quarterly Newsletter
Contact Information
CRPC Home Page

UniGuide Featured Site


10.03.97 by Steve Fisher, managing editor, HPCwire

Steve Wallach, co-founder of Convex Computer Corporation -- bought out by Hewlett-Packard in 1995 -- recently resigned his position as Chief Technologist for HP/Convex. He has subsequently joined CenterPoint Venture Partners, a venture capital firm specializing in high technology projects. Wallach will continue to be a member of President Clinton's advisory committee on High Performance Computing and Communications and chairman of CRPC's external advisory committee.

In this first installment of a two part interview, HPCwire questioned Wallach on a number of issues including his reasons for leaving HP/Convex, the direction that company is taking and his plans for the future. The following are excerpts from that conversation.


HPCwire: What was the rationale behind your break with HP/Convex?

WALLACH: Well, it was simple. I felt I had met my obligations. I told HP when they acquired Convex, that I was signing up to get the products delivered, which at that point were not announced (the Exemplar Series) and to make sure the division had its proper role within HP.

Once that happened, I wasn't saying I was going to leave, but I told them I would assess the situation and make a decision. After assessing the situation I said that basically, I want to start a company all over again. It wasn't so much that I didn't like what I was doing there or things like that, but a more free-spirited, entrepreneurial environment was more to my liking.

HP is a fine company and the Convex division has certainly done well. It was just time. It's not that unusual that the founders of a company bought out by a bigger company leave. That story is told over and over again, and I guess that's what happened. They were not surprised, nobody was really surprised. Most of the comments were 'Gee, we thought you were going to leave earlier not later.' But I kept my word, and that was it.

HPCwire: Just time for a change then?

WALLACH: That's all, there was nothing malicious or anything like that.

HPCwire: Is it true that you will remain a consultant to HP's high performance computing business?

WALLACH: Well, first of all, people there still call me up and ask my opinions on certain things, but legally, I'm not a consultant; in the sense that I don't have a contract with them. But if they call me up, I certainly help them out. I get e-mails from various people, but I'm not a lawyer where I round things up to the nearest fifteen minutes or whatever, it's all gratis. I'm very appreciative of what HP has done with the Convex division.

HPCwire: What direction do you feel HP is taking with its high performance computing business? Do you agree with it?

WALLACH: Fundamentally there is always some degree of 'you think this, I think that' but it's becoming fact for preparation for this interview I was.... a lot of people use Linpacks and Top 500's for various reasons. Like anything else, it's a set of data. But what I like is how you can show trends. Larry Smarr has a set of charts where he takes the top 500 and shows the trend of which are vector machines and which are RISC based machines. It shows, no great surprise, that vector machines are decreasing in terms of the percentage of the top 500 and RISC machines are substantially increasing. So you take whatever conclusion you want, but he's just taking the pure data and doing the calculations.

The Top 500 is published twice a year, for the Mannheim conference in June and for SC97, coming up shortly. In the June 19th report, 17 of the 500 are under the column called area installations, are called industry database; as opposed to academic or research, or things like that. I would venture to guess that if you looked at it a year ago, that number would have been less, and if you looked two or three years ago it could have been zero.

Another observation and again, you can print this out on the Web, Sun's biggest system, called the Ultra HPC 10000, which is a 64 CPU system...if you ignore what's at Sun itself, one of the biggest installations is at Allstate Insurance in Chicago, for database...commercial applications as opposed to technical ones. The largest installation of IBM SP/2 outside of government funded efforts is at Chase Manhattan Bank. The only reason I bring that up is, to answer your question, and I think this is going to be somewhat controversial, but I think more and more in high performance computing, what's happening especially as it all moves to RISC-based machines is, with the cost to develop these systems coupled with reduced government funding for high performance systems in general, that if the systems can't fit both the commercial and technical market, they just won't exist.

I think with HP, because they are such a major presence in the commercial market, just this week they announced the highest performance TPC-C benchmark on the Exemplar, the one designed by Convex. You're going to see over time that the right model in this high performance industry will be as follows... we design these high performance technical systems (for the technical market). They will be used by the NSF centers, the NCSA, San Diego, JPL, Caltech, the government labs etc. And then once they're out there and stable after a year or so I think you will find that the most successful companies will be the ones that can then take it into the commercial marketplace. Where rather than running finite element code, they'll run industry standard databases like Oracle and Sybase, Informix or codes like SAP, which manages the company.

So what has happened is that by the technical side willing to take the innovators and be the pioneers, what happens is, after everything is debugged and in the field for a year and you have hundreds of installations, you have the stability of the design and the stability of the software, you can then take it into the commercial marketplace. Because it's certainly not the beta test site, you can point to a hundred installations as opposed to just saying 'we've got thirty machines inside the company that we've been using to Q/A the software for the past year.' It's much more impressive to say we have thirty machines and in addition we have these hundreds of installations.

HP believes they can sell billions of dollars of the Exemplar series in the commercial marketplace. Which is consistent with the size of that effort. If you look in the last several months SGI has said they want to very aggressively go into the commercial marketplace. Sun of course is a very aggressive player in the commercial marketplace and I think I've seen, though I'd defer to IBM, that more than half of their installations of the SP-2 are in the commercial marketplace as opposed to the technical. So they start off in the technical first, not the commercial. I think that's what's happening.

I think the reason HP and the Convex division will do well is that model is working very well. That is the appropriate technology and market model. It's the appropriate technical model because the commercial market wants scalability, they want reliability and you have to incorporate those features if ultimately you're going to sell it to the commercial side. In fact there's a lot more focus now on the fault resiliency of the designs because of the commercial marketplace. I'm not saying we didn't do fault resiliency for the technical side, but when you're designing a car and all of a sudden it doesn't work for an hour, so you restart your high element code. I'm not saying people ignore it, but it's not as much of a disaster as if my website is down for an hour. So the term is mission critical and that marriage has worked very well.

HPCwire: So you're forming a new company?

WALLACH: Well, I'm in the process. I have a lot of different ideas.

HPCwire: What kind of firm are you looking at? What kind of products?

WALLACH: I'm not going to discuss that until as they say, the fat lady lady sings. I'm just looking all over the place.

HPCwire: So nothing is set with the new firm then.

WALLACH: That's correct.


Topics addressed in part II of HPCwire's interview with Steve Wallach include the battle between NT and Unix and the present and future state of the high performance computing industry. It will appear in next week's (October 10th) edition.


Steve Fisher is HPCwire managing editor. Email: