Cray T94 Supercomputer on Ebay (ebay.com)
"I am liquidating some of my personal collection of unique stuff and my Cray T94 Super Computer must go. What you see in the photos is what I have and what you get. I know that you need more than what I have to make it work. You should consider inspection the item to make sure it is what you want as it is sold as is with no returns."
58 comments
[ 9.8 ms ] story [ 134 ms ] threadThe system chassis weighs ten tons, contains four tons of fluorinert coolant, and is approximately the shape and size of a very large chest freezer
And yet people complain about their mobile phone not being thin enough. It's amazing how far we've come in only 15 years.
http://www.parallax-tech.com/fluorine.htm#price
The CRAY T94 mainframe is 5 ft wide, 3 ft deep, and 4 ft high (1.5 m x 1 m x 1.5 m). It has a dry weight of 2000 lbs (909 kg).[1]
So only 120 times thicker than an iPhone 5, but 8,000 times more massive.
[1] https://cug.org/5-publications/proceedings_attendee_lists/19...
http://www.avonaero.com/
http://www.vdsracingengines.com/aboutus.asp
Wow, that's...uhh...impressive. I assume most people will just buy it to put in their lobby as an expensive couch.
edit
I was mistaken. The T94 doesn't have the same floor plan as the XMP.
Perhaps it'll make a good beer dispenser then.
I sometimes find it sad that pieces like this T94 wind up as museum pieces or conversation pieces, but it's just one of the casualties of our rapid rate of progression. I mean, the Qualcomm Snapdragons in cell phones are probably more powerful, and they run off a battery worth a couple thousand mAh.
The 32 processor version cost $39 million.
Of course, in its day, intel and AMD were still in the middle of the mflops.
[1] http://www.ebay.com/itm/RARE-PROTOTYPE-MACINTOSH-128k-COMPUT...
Market price is what someone will actually pay for it.
Add to this the Sony floppy "innovation" "magic", and it turns out, that AAPL is just a branding outfit, as The Register never misses to point out.
The T94 has a peak theoretical performance of about 8 GFLOPS, 1 gig of ram, half a meg of L2 cache and an SSD up to 4 gigs in size.
http://www.craysupercomputers.com/downloads/CrayT94/CrayT94_...
Compare this to the Galaxy S III, which has at least 1 GB of ram, 64 GB of flash storage, up to 8 MB of L2 cache per core (4x), and nearly 20 GFLOPS of real world performance (on the GPU).
It seems as though bringing memristor technology to market is a sure thing at this point. And that will again revolutionize computing technology in several steps. First, what happens when durable storage has the speed of RAM? Well, everything gets faster, of course. But then sleep functionality gets a whole crap-ton better (because waking from sleep can be effectively instantaneous). Which means longer battery life for mobile devices, yadda yadda.
Then there's the 2nd memristor revolution, when you use memristor technology in FPGA like devices. What happens when you can reconfigure a high-density integrated circuit with clock speeds in the gigahertz range and reconfiguration speeds on the scale of memory writes? Well, now you get everything the FPGA could have been, a jack-of-all trades ASIC device which can serve as GPU one moment, as CPU the next, as signal processor the next, etc, fluctuating between jobs perhaps thousands of times per second, as necessary. Imagine what you could do with that? As a simple thing, imagine an LLVM or Javascript virtual machine implemented in reconfigurable hardware and how much faster that would make it.
Then there's the 3rd memristor revolution, where memristor's are used directly to implement logic instead of transistors. How this could play out is anybody's guess, but when I think about it even a little the idea of the technological singularity quickly comes to mind. The implications for raw computing power and for machine based learning are truly astounding.
http://en.wikipedia.org/wiki/Computronium
http://www.eetimes.com/electronics-news/4229171/HP-Hynix-to-...
We don't have the benchmarking data, but I strongly suspect that these kinds of implementations will actually be significantly slower than what compilers are capable of doing on x86-64. This is almost certainly going to be true for stack-based VMs (stack operations are ridiculously slow compared to registers, and because the push/pop sequence done by instructions is implicit and linear you can't take advantage of any parallelization techniques like superscalar execution, branch prediction, out-of-order execution, and even pipelining is less effective).
Even for register-based ones the obvious things like type checking are actually extremely efficient on modern processors (I wrote an explanation on Quora: http://www.quora.com/What-is-a-lisp-machine-and-what-is-so-g...).
To get the same level of performance as modern CPUs you really need to take advantage of all the parallelization techniques I mentioned above. This is extremely difficult to design because you need to ensure all the permutations that are possible in valid instruction sequences produce the correct results in the presence of all the reordering/parallel execution going on. Modern CPUs actually have a lot of bugs that are found relating to this, but they only occur in very unusual code, and these bugs are fixed either by patching the microcode on the CPUs or by the OS, so you rarely encounter them. And this despite the amazingly thorough testing and huge amounts of formal verification that go into CPU designs.
I think FPGA-based designs will continue to be very algorithm specific for those reasons, even if we get FPGAs everywhere.
) Cray lists the fully loaded 32-cpu T90 has doing 800 GB/sec, a 8 GFLOPS T94 has 4 CPUs
The coolest part was a decade later, I ran into the guy that bought it... on Hacker News. And that guy was Steve Blank, who turned out to be a hero of mine.
http://steveblank.com/2009/11/19/closure/
And then...
>It sat in my barn next to the tractors and manure for five years. I had the only farm capable of nuclear weapons design. Cray called two years ago and bought it back for parts for an unnamed customer still running one.
That's kind of sad. But he did help ensure another Cray remained in working condition, so I guess that's a positive aspect of the story!
(http://www.chrisfenton.com/homebrew-cray-1a/)
I think he'd be interested in talking to anyone with Cray documentation or software.
We have come a long way in a short time. But I do miss the outlandish design of the Crays, and the blinking lights of Thinking machines. A bunch of racks just isn't the same.
You better be able to handle the heat dissipation, too :-)
EDIT: just to expand on this a bit, this means that there are workloads that these old-school supercomputers will run much faster than a modern high-end desktop. This particularly applies to workloads which have a lot of shared memory access at random locations - a very difficult case for the modern systems, which depend on high cache hit rates. Also, the GFLOPS ratings of these supercomputer processors are in many ways more real than the ratings of commodity processors, which depend on the pipelines being filled in very specific ways. So no, you can't replace this system (at least the 32-processor version) with a smartphone or even a desktop. Which is not to say it would be cost-effective at 39 million of 1996 dollars.
Another case which I saw personally was a simulation of a part of the visual cortex of the brain; you had neurons which were connected to their neighbors, but you also had a bunch of connections to far-away neurons, and the bandwidth between processors which simulated different parts of the cortex became a limitation (and the huge supercomputers which had the bandwidth were (a) expensive, and (b) had relatively slow processors for the number crunching in each region).
Except in this case, I found that the physical delay which existed on a long connection between neurons allowed us to buffer the messages and send a notification about the whole train of impulses, effectively compressing the data. Together with some other simple changes, the simulation ran 10 to 100 times faster, and could use clusters instead of supercomputers.
In general, there are not that many cases in which you really can't get rid of the requirement of fast non-local memory access; if there were, these supercomputers wouldn't have died out. But they were useful in some cases, and were also good for freeing people from thinking about how to localize their memory accesses - this speeded up development.