# Personal supercomputing, as long as it's under $10k USD The John’s (West and Leidel) at InsideHPC.com did a nice study on personal supercomputing at the site. It is worth a read. In short, they found people would find such boxen useful. But they don’t want to spend more than$10k for them.
This is interesting at many levels. Matches up very well with informal/anecdotal data we have from conversations with users.

We noticed that users wanted personal supers many years ago. They wanted something that could give immediate feedback, even if it wasn’t as fast as the heavy metal down the hall.
This is part of what drove us to build our Pegasus many-core systems. The purpose of these are to localize a significant amount of power. And they easily hit below the $10k price point, even for the 16 core systems. While these are shared memory single units, you can run them with MPI as 8-16 way computing systems. We have quite a few customers doing exactly that. The desktops are so powerful, our customers are using them to run reasonable sized CFD and other simulations on them. This said, distributed resources are more complex. You need N motherboards, N CPUs, and N memories. If you don’t mind inexpensive motherboards (aka desktop motherboards), and some strange cases, you can keep the cost per system pretty far down. Assume you can get a quad core MB + 4GB memory + CPU for about$800. The problem with most desktop motherboards are that they don’t include video out, and don’t take kindly to booting without this. You can find them though.
So for $800/unit, get 9 units, and put disks on one. Get a small switch. The issue is that the case and mounting issues will be hard. Your BOM costs would be close to$9000 when done. Under $10k for this type of home-brew system may be possible, but it would be hard. Ok. What are the other options? This is why this is interesting. Accelerators. GPUs specifically. With the right platform, you can add multiple units. With less complexity than the cluster. I’ve been making the point for a while that if you give people the choice to manage a cluster or a single box as powerful as the cluster, the latter will win hands down. And this is why the price profile of accelerators is so important. You can create an awesome computational engine in well under$10k. We have. You can build a cluster in under $10k but you are going to need to skimp on resources to get there. Most folks won’t want to build their own. Which is why the CX-1 and the SiCortex machines are interesting. And hopefully why Pegasus is interesting. We have been showing them for 2 years, and selling them for as long. End users want many-core and parallel machines at their desk. This is known, and the study helps nail that point. They don’t want to spend more than$10k.
I am sure others will enter this market and get their products noted on HPCwire. Seems to happen a bit these days. Maybe we should figure out what they are doing, and do some of that ourselves. Our customers who know about Pegasus like them. Just like JackRabbit.

### 7 thoughts on “Personal supercomputing, as long as it's under $10k USD” 1. I agree that at the professional level the <$10k single machine price point is very attractive for deskside HPC. Far more so than a small cluster which I’ve found to just be more hassle to manage and program.
However, I’d also love to see more choice around the $3k price point for machines oriented towards compute and data intensive work. Not workstations but small headless deskside servers with reasonable chunk of memory, at least 500MB/s disk IO from 6TB or more, plus 8 cores compute. A GPU accelerator is also welcome. Not rack mount and not to use as a workstation – given the constraints how close can we get for$3k?

2. Like this but with 8 cores (would add about $250 to the cost here) We are getting ~650 MB/s out of it. Not at$3k.
We could build a deskside version of ΔV that would come quite close to hitting all of these targets. I have not thought that this would be a viable market though. From a vertical market perspective (Financial, Oil and Gas, medical imaging, etc) where do you think it might play?
I am curious. I will look into it. We have some quotes and proposals to finish up the weekend, I might run the numbers and see if it could work.

3. BTW: I should point out that the picture in the link is of the machine that runs our day job web site, among other things. We eat our own dog food, so to speak. This is a product we sell and support.

4. I had smaller financial firms and sw dev houses in mind. Having that size and speed of IO available per developer can be transformative when it comes to analyzing larger data subsets (eg. a 5TB sized subset of a 100TB main data set) when it comes to research or algorithm development. Its my own bias but I, like many programmers, discover life isn’t so simple when doing larger runs over web data etc.
The kinda folks who want to do netflix recommendation prize hacking, tim bray’s widefinder[1], or generally trying to do development with much larger data sets and concurrency in mind. While there is definitely a market for the hign end personal hpc box, its just my opinion (no business case put together obviously) that there is a market for midrange development boxes for this.
Far more cost effective to have developers use these and then have larger servers for the even bigger runs rather than have developers contending even during small runs. Or heaven forbid some kind of shared dev SAN.
When I looked to buy a machine recently I found it hard to get the raw deskside power in the places I needed it. I guess your deskside jackrabbit fits the bill if I can find an extra 2k.
[1] http://wikis.sun.com/display/WideFinder/The+Benchmark

5. No news on this one?

6. Should have it soon. Dealt with a cluster break-in last week that used up lots of my time, and got lots of people mad at me for telling them what happened, what went wrong, and how to fix it.

7. Looks like we can build the Pegasus unit with 8 cores (2GHz Intel E5405) 16 GB ram, 12x 500 GB SATA drives in hot swap units, hardware accelerated RAID, and get it out for about $4963. Dropping to 4 cores and 8 GB ram, gets us to$4257. This is with 2 PCIe-16 (though one of the two is a PCI-e x4 in that slot). This system should sustain about 700 MB/s+ to disk in RAID6.
If we pull the hardware accelerated RAID, and make it look like a ΔV (software RAID), the same unit will come in at $3745 with 8 cores and 16 GB ram, and$3027 with 4 cores and 8 GB ram. Here we may be able to hit about 400 MB/s or so in RAID6.
If we switch to an AMD platform, we can shave it a little more.
You would need a graphics card for this unit, no integrated port. If you add in a decent nVidia card into the PCI-e slot (say a GTX 260, see the video 1 above to see me playing with one) it would add ~300-350 to the cost (depends upon which card).
We could up the performance of these quite a bit as well. Specifically, this unit would let us add in up to 32x 2.5 inch drives, and we could use SAS and fast WD SATA units together with the hardware accelerated RAIDs. Likely we could exceed our 1.6 GB/s measurement for the 24 bay device (which is interestingly our most popular hardware product) with these units. Depending upon the drives used, we could hit some pretty interesting price points with excellent performance.
Please send me an email (joe _at_ scalability _dot_ org or landman _at_ scalableinformatics _dot_ com) if you want it in a more formal manner. And if you think people are interested in this, please, by all means, we are happy to speak with them.