On the Stanford Folding pages, todays fact of the day is:


Folding fact #18
The evolution of molecular mechanics:
1985 Myoglobin .30 nanoseconds (ns) simulation completed in 50 days on Vax Computer;
1992 HIV protease .10 ns simulation completed in 100 hours on a Cray comp.;
1998 villin 1000 ns simulation completed in 4 months Cray;
2001 B- Hairpin 38000 ns simulation completed in 8 days 5000 computer cluster of foldingathome.



Which when you work it out, means the 1985 comp. would have taken 63333 days i.e. 173 Yrs to do what the 2001 comps. did

Thats if the 1985 comp. worked out 30ns, but if you look it actually has a '.' before the '30'...
so that would mean my calculation is 100X to small (IIRC you don't use zero after the decimal place so it's a bit ambiguous)

And when you factor in the speed of todays comps. compared with 2001 models the numbers get even more amazing.


Fold on Mighty Brothers.

Pete

Different proteins take different amounts of time to simulate, and I'm sure the algorithms have evolved, so it's hard to compare the different proteins, but one could use Moores Law to "guestimate" how much more powerful a modern processor is to a circa 1985 processor.

I'm too embarassed to reveal how I calculated this, but I come up with 3000 times more powerful, or 8 days x 3000 = 65 years to do on a 1985 Vax, what can be done today in 8 days with a bunch of PC's.

Regardless of the numbers, what I find surprising is that they were able to do these simulations in 1985, that the knowledge was there to do it. On the other hand, the fact that Tinker is written in Fortran should provide some clue that molecular biologists have been doing these simulations for awhile.

David


Edit: As sbabb points out, I failed to take the difference in simulation time into consideration, so my number is too small by a factor of 38000 ns / 0.30 ns = 126666.67

38000 ns / 0.30 ns = 126666.67

If we're assuming that the proteins are equivalent in computational difficulty per unit of time, then that's the ratio of work between the VAX and the 5000 machine F@H cluster.

Since the VAX took 50 days to complete its protein, 126666.67 X 50 = 6333333.50 days is how long it should take a VAX (Model 780? 785?) to complete the 38000 ns fold. Divide that by 365.25 days per year and you get 17339.72 years of work.

If we take the 126666.67 computation ratio and divide it by the 5000 machines in the cluster we get the average "compute power" of a 2001 PC being 25.33 times that of the 1985 VAX.

Computationally intense applications are still written in Fortran. Fortran works well for these types of applications and the decades of work in optimizing Fortran compilers and libraries for HPC (High Performance Computing) applications usually makes them second only to hand-coding in assembly language for speed.


Scott

sbabb, Do you know any more info on the VAX?

I take it , it was a supercomputer in it's day.


Pete

The VAX (officially I think it stood for "Virtual Address eXtension" but those of us who had to program them often said "Virtual Architecture: eXperimental!) was Digital Equipment Corporation's follow-on to the successful PDP-11 line of minicomputers. DEC often referred to is as a "superminicomputer" but it was really no comparison to the real supercomputers of the day like the Cray-1, CDC Cyber 205, Cray X-MP, etc.

The original VAX 11/780 showed up in the late 1970s and it was the first commercial 32-bit computer. It was generally considered capable of executing one million instructions per second (MIPS) and the "VAX MIPS" became a de facto standard against which other computers CPU performance was measured. The "Virtual Address eXtension" bit meant that the VAX had memory management designed to use the disks as an extension to main RAM memory: Virtual memory. Physically a VAX 780 was at least 4 feet wide (more once you add disks, tapes, extra memory, etc.), 5 feet tall, and 3 feet deep. It lived in computer rooms on raised floor tiles.

The original VAX operating system was VMS (Virtual Memory System). It's been said that Dennis Ritchie took all the good stuff from the old (and huge) Multics operating system and created Unix, then Digital took what was left and created VMS. If you liked typing, VMS was a wonderful OS for you. I remember getting a directory listing by typing stuff like "DIRECTORY/FULL SERVER::USER$DISK:[PATH.TO.DESIRED.DIRECTORY.SEPARATED.BY.DOTS]FILENAME.EXTENSION;VERSION" Seriously.

Ken Olsen (founder and CEO of DEC) hated Unix and took every opportunity to trumpet the superiority of VMS. He often used odd measures like the size of the documentation set, proudly stating that the VMS doc set took up over 15 feet of shelf space while with Unix "you only get that one thin manual." He went through a bit of shock and denial when he found out that 1/2 the VAXes going out the door didn't have VMS licenses (and VMS service contracts, and the attendant big revenue for both) included because people were loading Unix on them. Grudgingly he decided that DEC should do their own Unix for the VAX, and "Ultrix" was born. Armando Stettner was the project manager for Ultrix and he hated the name. He told me (yes, I've been around that long) that he thought it sounded like a toothpaste brand. He had a military gas mask that he used to wear whenever he went to meetings with the marketing folks, just to let them know how he felt.

The VAX 780 was followed by the lower cost, lower performance 750 and 730. Then came the higher performance 782 and 785, the MicroVAX line of desktop and deskside systems, and finally the 8000 line. The later (around 1990) VAX 6000 line mixed in the new 64-bit Alpha CPU, so the waters got a bit muddy there.

VMS was eventually renamed OpenVMS to try to convince the masses that a simple name change made it a player in the buzzword-compliant area of "open systems" which was dominated by Sun, Apollo and HP with Unix-based systems. In the 1990s Microsoft was working on a replacement for Windows 3.x and Windows 95 with IBM (actually IBM was doing all the real work, since Microsoft had never actually written anything from scratch) that eventually became OS/2. One of the lead VMS designers left DEC with the source to the new version of VMS he was working on and went to Microsoft. OS/2 quickly got kicked to IBM and Windows NT was born from VMS (again Microsoft wrote nothing from scratch) with the Windows 3.1 GUI sitting on top of it. Later this morphed through a few versions of Windows NT before becoming Windows 2000, then Windows XP, and currently Windows 2003. The Windows 3.x line died, replaced by Windows 95. Win95 got updated to Windows 98, Windows 98 SE, and finally Windows ME (Millenium Edition) before dying off. Some feel that Windows ME was already brain-dead.

So the VAX was a fairly revolutionary and powerful machine in it's day, but it was no supercomputer, despite DEC's marketing efforts. While the VAX was crunching out 1 MIPS (Meaningless Indication of Processor Speed?) as a general-purpose system, real supercomputers of the time focused on floating-point performance and the Cray-1 did 133 peak megaflops (million floating point operations per second) in pre-VAX 1976. During the mid-1980s VAX heyday the Cray X-MP (1982) did 500 megaflops and the the Cray-2 did 1900 (1985) did 1900 megaflops. Quite a performance difference over the VAX, but also a big price differential. The Crays would have made interesting folding machines. Generally they were used by the government for weather simulations and whatever "other stuff" the NSA, CIA, and Department Of Defense did with them.

Now isn't that more than you wanted to know about Mesozoic computing?


Scott

So... I take it (reading between the lines) that's a yes then?

Jesting aside, Thanks, that was an extraordinary reply.

I think we will dub you, 'sbabb, SPCR's font of VAX knowledge.'



Pete

You do me proud Scott!

I started working with VAX/VMS in the early 80's, and only in the last two years have made the switch to Unix. When DEC got bought by Compaq, the company I work for saw the writing on the wall, that VMS would eventually not be supported, and thus decided to convert from VMS to Unix (True64). The project has taken 2 years. We convert this coming weeking.

David

OUCH! The fine folks at H-P must have known that you were going the Tru64 route and planned accordingly when they decided to kill off the best CPU they owned!

Can I interest you in a Sun system running a fine shade of Solaris?

And does anyone have any news on if the SPARC/Solaris version of F@H will EVER show up?