Miscellaneous Operating Systems/Hardware

I came across this material about some special Xeon coprocessor boards that SGI are now having as an option in some of their servers (even in 1U systems).

http://www.sgi.com/products/servers/accelerators/phi.html

Apparently these boards each have 60 Xeon cores running at over 1 GHz, with 8 GB of RAM, and the coprocessor can actually run independently as a Linux server if you want to use it that way (you can even SSH into it). Times are strange. Hardware technology is moving so incredibly fast!

I was watching a video recently of a data center from the early 1990's, which featured some old mini fridge sized Sun boxes. I was curious to find out how much actual processing power a classic SVR4 Unix box like that would have had in the early 1990's, and found out that the CPU on this model was less than 24 MIPS! Then I started looking through old SPEC CPU2000 integer performance benchmarks, to see how some classic Unix systems fared from the late 90's and early 2000's, vs. the PC architecture machines. Funny enough, the $8000 Power5 workstations were comparable in CPU power to the $1000 x86 workstations of their time. All the old Alpha chips as well look to be no more powerful than the early line of Pentium 4 chips.

http://www.spec.org/cpu2000/results/cint2000.html

It's kind of depressing to look at data that puts a number on an older piece of technology. Many of these systems were novel and engineered well, and large models could definitely scale to a high degree. A lot of thought was put into building these different platforms. It makes me realize, though, that in the late 1990's or early 2000's, at least the smaller Unix machines were already starting to look old and behind the times. For example, a review is still up for an RS/6000 workstation in the late 1990's. (Note that the price of the test model they received was over $80,000!)

http://www.drdobbs.com/ibms-rs6000-43p-model-260/199200782

For some comparisons to other machines in the workstation market:

Advances in technology are amazing, not only in cpu, but the same can be observed in memory density, storage, networking, gfx etc.

Pre-2000 our 7TB nfs storage array spanned over 2000 drives; it was running on sun (tatung) clones, and the clients were freebsd hosts; it felt fast and large back then, now you can have roughly the same thing in a desktop; for a mere grand.

Now, the real question isn't how fast it is, but how you use it, if it's to run World of Warcraft or watch movies, then I think it might be somewhat overkill, but if it's to run a clustered filesystem, it might be worth it.

I see some people doing stuff on their 16Mhz arduino that's probably more useful than someone's $500 microsoft surface. Oh and I forgot to mention, Apollo 11's computer was 2Mhz; they made it to the moon and back; about 40 years ago (they probably wouldn't have made it back if they had had a microsoft surface instead).

The Phi is what became of Larrabee . The 60 cores are not modern Xeon cores (Sandy Bridge) but Pentium cores with some extensions.

Tilera have chips with 100 cores.

Exactly! And that speed is in the eye of the beholder. On a very responsive application (i.e using asynchronous non-UI thread blocking coding) you can turn many devices with much less MHz/ghz, slower IO into a great platform. It's that most software these days is written with the mindset of "I've got tons of processing power, SSDs, gigabit Ethernet etc, why should I care about performance?" I see it all the time in my part of the country. Typically the answer from them is, buy better hardware not look at the code and make it better. Often times it's them simply not using the right approach for the problem, taking the shortcut.

Went through some very underperforming C#/SQL yesterday, the original coder on top of calling a pretty complex SQL twice (once in the code behind to a variable he wasn't doing anything with and the other in the page itself), he was doing a SELECT * on the query when he only needed 2 columns out of 20.

What's funny (and sad) is that those same folks, don't code with more than 1 CPU in mind. As we know on here, most of the time, the more cores, the lower clock rate, so in a single threaded application you might actually get worse performance than that of a less core/higher clock server/client.

Very true, I guess what matters most is whether the machine fulfills its purpose. When I was a teenager, I didn't have much money, but I managed to get some used 486 business PC's for free. For what I needed the machines for, learning how to use Unix, ten year old PC's were completely sufficient.

Later when I was upgraded to a Pentium 2 and had a 14.4 kbps external modem, I felt that was a huge upgrade. I could even use X11 and the Mozilla suite -- living in luxury! The big change came a few years later when I wanted to use Firefox, play music and videos, and multi-task more. At that point, even 512 MB of RAM and an Athlon 2500 seemed like they were just "okay."

I get wear just fine out of my G5. Once it gets to the point where it is not an effective browsing machine (it's nowhere near that yet), then I'd probably have the Intel mini run Linux in a VM and allow Firefox on the VM to connect to X11 on the G5 so I can still use the rest of the stuff the G5 works perfectly well for (Photoshop, Final Cut HD, games, Mac OS 9 Classic apps, basic Microsoft Office).