A Nicely-Built 40-Core Raspberry Pi Cluster

[schmidtbag]

So tell me, just how many of you WTF'ers and whiners happens to have a 40 core cluster sitting on your desk?

None of you. That's what I figured. So please keep your petty comments to yourself. It must be hard to be so small minded.

To the gentleman who built this wonderful cluster.....

Well done Sir. Keep up the good work. You have a bright future before you. And please ignore the WTF'ers and the Why Not'ers. They are just jealous of what you have accomplished.

I don't think anyone here was whining about it. It's more of a "what was the point of this?" than anything. A single GTX 680 or a R9 290X costs a lot less, takes up less space, has more practical uses, and probably has more processing power than that entire cluster (theoretical and actual). Of course, that takes away a lot of the fun of making such an elegant machine. Obviously if the creator seriously cared about performance, he wouldn't have chosen the Pi. I think we can all easily admit the cluster was amazingly well designed, but what we don't understand is why the Pi was used when it was probably the worst possible system (within reason) that could be used for this. To put it in another way - the cluster was obsolete and cost ineffective before it was even made. The Pi is cheap but spend another $400-$600 (which is really negligible for a 40-system cluster) and the system could've been at least 3x more powerful, and maybe could've been a little smaller.

[georgezilla]

Here's the real question ....

So tell me, just how many of you WTF'ers and whiners happens to have a 40 core cluster sitting on your desk?

None of you. That's what I figured. So please keep your petty comments to yourself. It must be hard to be so small minded.

To the gentleman who built this wonderful cluster.....

Well done Sir. Keep up the good work. You have a bright future before you. And please ignore the WTF'ers and the Why Not'ers. They are just jealous of what you have accomplished.

[schmidtbag]

You can always start your own project with discrete components or 3d print your design. Or use FPGAs. It's delusional to think that x86 is the be all and end all of architectures.

Delusional is a bit strong of a word. x86, PPC, and sort of SPARC were the only architectures with a modular system on a consumer level. PPC and SPARC dropped out of the PC market a long time ago, so that just leaves x86. All other architectures are either under-performing, single-board systems, or SoCs. FPGAs are not ideal for everyday computing purposes. They make good hobbyist projects, not so much high-end enthusiast rigs.

I personally own 3 different ARM platforms and I'm interested in getting a MIPS or PPC based system - I'm probably a lot more open to other architectures than most people on these forums (or in general). But at the end of the day, I want to sit down and use a computer that is well supported. I also want a computer that was put together with parts I picked out, and I just don't see that happening in the future, x86 or otherwise. When you look at how tiny things like the odroid U2 are and how capable it is, you know the future of enthusiast hardware is coming to an end.

[caligula]

I agree - as a PC enthusiast, its somewhat sad that whenever I intend to build a new rig, it's probably going to be no bigger than a micro ATX with 1 stick of RAM, a cpu, a single SSD, and a single GPU. There won't be a need to put in anything else, but the only advantage to the computer vs an OEM brand is the ability to overclock. Even overclocking has become a pretty boring market lately. Anyone can breach 4GHz these days without even trying, and there's a good chance you don't even need the extra speed to begin with.

I'm thinking 10 years from now, all PCs will basically be the same platform, they'll all be pocket sized, and the only large computers will be either mainframes to handle unfathomable amounts of data, or quantum computers (which will basically do the same thing). Technology is going to look boring pretty soon.

You can always start your own project with discrete components or 3d print your design. Or use FPGAs. It's delusional to think that x86 is the be all and end all of architectures.

[curaga]

You can always go retro, and build a Beowulf cluster out of 2010 parts in 2050.

Of course there's new areas of exciting tech too. For example 3d printing, the moment you can print metal items affordably, I'm so in
The current plastic items aren't that useful, the print area is small, and the quality bad. Some of the new resin-based printers are starting to have acceptable quality vs the filament-based ones though.

But once you can print big metal items with good quality, the things you can do will skyrocket. Yes I'll download a car.

[schmidtbag]

Agreed. This is a very slick build, both in terms of "contents" and execution.

I find building a modern PC kind of depressing -- you put a few components together and everything just works. Not like the old mainframes... I miss the miles of beautifully managed cabling, aisles of components, and the big panels of blinky lights showing activity in each page of memory

I agree - as a PC enthusiast, its somewhat sad that whenever I intend to build a new rig, it's probably going to be no bigger than a micro ATX with 1 stick of RAM, a cpu, a single SSD, and a single GPU. There won't be a need to put in anything else, but the only advantage to the computer vs an OEM brand is the ability to overclock. Even overclocking has become a pretty boring market lately. Anyone can breach 4GHz these days without even trying, and there's a good chance you don't even need the extra speed to begin with.

I'm thinking 10 years from now, all PCs will basically be the same platform, they'll all be pocket sized, and the only large computers will be either mainframes to handle unfathomable amounts of data, or quantum computers (which will basically do the same thing). Technology is going to look boring pretty soon.

[bridgman]

Agreed. This is a very slick build, both in terms of "contents" and execution.

I find building a modern PC kind of depressing -- you put a few components together and everything just works. Not like the old mainframes... I miss the miles of beautifully managed cabling, aisles of components, and the big panels of blinky lights showing activity in each page of memory

[TeoLinuX]

I think the project is great.

Sure, RPi is a poor choice, but let's consider it's a test build, made mostly out of fun, in which the main goal is the number of NODES (not cores!) and that computational power is not the ultimate purpose (otherwise the author is right:go for a i7 + GPU)

The DIY acrilic/hardware case is really badass AND good looking

My major concern for using such a rig for testing massively parallel software is the tremendous bottleneck caused by the internal ethernet network.

In any case, congrats on your build; and I'm sure those 3 grands gave you not only lots of fun, but a good "visibility" and a nice story to add to you resum?.

[droidhacker]

Well, sure, but a 486 won't be able to do a whole lot of concurrent operations. That's like comparing a CPU to a GPU--they're different beasts, with different purposes.

That said, he was doing it sort-of for fun, but it started out as a thesis project. According to his blog post, he's going to use it to test distributed software.

486 is perfectly capable of process concurrency.
But my point was partially in jest anyway, for reasons as pointed out by the builder of the system. For raw power, it just isn't a huge bang for the buck, yet it is still a fun project.

[Kano]

For the price of a 40 core PI cluster you could easyly buy 2x Intel Xeon E5 with 10 cores, which means 20 cores in one box with over 2 ghz each. Maybe for massive parallel apps a Xeon Phi with 61 cores active would be interesting as well. Using clusters of extra slow ARM v6 system leads to nothing, it is just crazy spent money. I have nothing against ARM in general and a PI can be used as low cost (VDR) server or XBMC client or for lots of other fun projects but it was never designed for HPC usage.