Parallella: A Supercomputer For Everyone • Hacker News Board

Re: Parallella: A Supercomputer For Everyone

#81

Post by microtherion » Thu, Oct 11, 2012, 1:14 AM UTC

adapteva wrote: post #72

microtherion wrote: post #63

Earlier quoted context omitted.

As one of the HN contributors who mocked the GHz performance spec in an earlier discussion thread, I welcome the fact that you're engaging in debate about this. That said, I still think that the GHz stat is just about as BAD a metric as any (I suppose "pin count times # of cores" would be worse :-). About the only positive inference I can draw from this is that you have the thermal situation in your system under cont…

Thanks for posting a very valid concern! Does the following FFT based image processing demo address the concern about communication bottlenecks in the approach? http://www.adapteva.com/white-papers/using-a-scalable-parall... Corner turns for 2D FFTs are usually quite challenging for GPUs and CPUs.[ref] Yaniv, our DSP guru, completed the corner turn part of the algorithm with ease in a couple of days and the on chip d…

Yes, this looks like a much more appealing argument to a distributed computing audience. As for marketing to "the general audience", it's not clear to me that this is a realistic aspiration.

I suspect that the people who would be happiest buying something like this are going to be very technical, not just USING Linpack, FFTs, neural networks, or HMMs on a regular basis, but used to IMPLEMENTING them as well. This audience is definitely going to want red meat like the paper you're linking to.

With the Kickstarter campaign, you may also get customers who just think it's cool to own a supercomputer, but when they realize they can't run Crysis on it, they may be disappointed.

Re: Parallella: A Supercomputer For Everyone

#82

Post by browndeertech » Thu, Oct 11, 2012, 1:32 AM UTC

compilercreator wrote: post #43

DeepDuh wrote: post #34

Earlier quoted context omitted.

Sounds pretty good, thanks for the heads up. Now I'm curious to see some benchmarks as soon as someone puts 20 or 30 of these on a board with lots of GDDR 3 Ram :).

Sorry apparently I was wrong about OpenMP support, not there right now but there is a C/C++ compiler with OpenCL.

OpenMP support would be interesting, and should be possible by extending what we did for the OpenCL support. The basic machinery is very similar. Also, someone mentioned Fortran. There are Fortran bindings for the STDCL API that is built on top of OpenCL, so this could help interface to existing Fortran codes and provide a partial solution for Fortran programmers.

Re: Parallella: A Supercomputer For Everyone

#83

Post by dfox » Thu, Oct 11, 2012, 1:55 AM UTC

DeepDuh wrote: post #80

dfox wrote: post #77

Earlier quoted context omitted.

GPUs are primarily designed to be good for graphics, which implies completely different internal architecture. While GPUs have some graphics-oriented functional units the main factor is that all these cores have to access pretty large chunk of shared memory (textures, frame buffer...) and do that uniformly fast (and also support some weird addressing modes and access patterns). I suspect, that large part of die area…

Sorry, but I have to correct a little bit here. Today's GPUs are - not simple SIMD. NVIDIA calls it SIMT (single instruction multiple thread), mostly since you can branch a subset of them, so for the programmer it does feel somewhat like threads. - not just optimized for Graphics anymore. E.g. since Fermi, the Tesla cards have DP performance = 50% of SP - which has been specifically introduced for HPC purposes. They…

My point is that GPUs have significantly different architecture from most of these "many cores on a chip" designs. Original reason for that was clearly that such architecture was necessary for graphics, coincidentally it works better for many interesting HPC workloads. It's clear that manufacturers are introducing technologies that are not required for graphics, but they cannot be expected to do modifications that will make their GPUs unusable for graphics.

And as for SIMD/SIMT, I mentioned SIMD mostly in relation to operations on short vectors done by one thread, which is mostly irrelevant to overall architecture of the core, as it can very well be implemented by pure combinational logic in one cycle given enough space. My mental model of how modern GPU core (physical, not logical) actually works is essentially some kind of simplistic RISC/VLIW design with large amounts of registers with compiler and or hardware interleaving instructions of multiple threads into one pipeline, which may or may not be how it actually works but it looks probable to me.

In my opinion most of chips like Epiphany IV or XMOS or whatever, in contrast to GPUs, are useful for only limited classes of workloads as they tend to be memory starved.

Re: Parallella: A Supercomputer For Everyone

#84

Post by vidarh » Thu, Oct 11, 2012, 10:36 AM UTC

throwaway1979 wrote: post #25

I picked up a raspberry Pi a few days ago. Initially, I was blown away by the low price point. Since then, I've been reflecting on what makes a computer useful. For personal computers - desktops and laptops - I think we don't have a shortage of processor cycles. The minimal specs of the Raspberry Pi make it useable - 256MB of RAM, 700 MHz CPU, a few GB of storage and enough MB to saturate a home broadband connection.…

It's a dev platform. A way for people to experiment with a new architecture. The roadmap is to eventually get to PCIe cards with thousands of cores.

Re: Parallella: A Supercomputer For Everyone

#85

Post by jacques_chester » Thu, Oct 11, 2012, 11:18 AM UTC

scott_s wrote: post #55

jacques_chester wrote: post #46

Earlier quoted context omitted.

And what sets those upper limits? In general: money. Buying more of the most performant equipment available. So. It's an economic designation.

Sure, you're limited by money. But if someone magically was able to produce a machine for $100 that was on par with Titan ( http://en.wikipedia.org/wiki/Titan_(supercomputer) ) I would still call it a "supercomputer" until it became ubiquitous.

Except that someone would buy $10 million of such machines and gang them together. And we would call that a supercomputer, not whatever you put under your desk.

Defining it on any architecture or performance metric is just pointless because the march of time renders such things utterly moot. Remember when PlayStation 2s were "supercomputers"? Please.

Re: Parallella: A Supercomputer For Everyone

#86

Post by vidarh » Thu, Oct 11, 2012, 11:42 AM UTC

adapteva wrote: post #76

wtracy wrote: post #67

Since I thought I saw an Adapteva person posting here earlier: If the Kickstarter falls through, what options could you still make available to hobbyists? Is there some version of your current prototype setup that you could sell, even if it's not one convenient board?

We would rather not think of that option:-) if the ks project fails, we'll do our best, but seems unlikely that we could support selling kits to hobbyists and they would certainly cost thousands of dollars each due to a lack of volume.

If you don't reach it in time, collect pre-orders. Seriously. Getting an escrow setup in place that in effect gives you a similar payment mechanism as Kickstarter (money handed over to you once the $750k is met; returned if criteria are not met) does not need to be expensive. Even without Escrow I think that if you get close to the target, a substantial number of those of us who've committed on Kickstarter will be ok with taking the risk. And it'd let you set longer/more flexible terms to make reaching it easier.

Re: Parallella: A Supercomputer For Everyone

#87

Post by scott_s » Thu, Oct 11, 2012, 8:12 PM UTC

jacques_chester wrote: post #85

scott_s wrote: post #55

Earlier quoted context omitted.

Sure, you're limited by money. But if someone magically was able to produce a machine for $100 that was on par with Titan ( http://en.wikipedia.org/wiki/Titan_(supercomputer) ) I would still call it a "supercomputer" until it became ubiquitous.

Except that someone would buy $10 million of such machines and gang them together. And we would call that a supercomputer, not whatever you put under your desk. Defining it on any architecture or performance metric is just pointless because the march of time renders such things utterly moot. Remember when PlayStation 2s were "supercomputers"? Please.

Note that I said it was a sliding performance target. That is, I agree that defining it as any particular architecture or absolute performance is pointless. It's a designation relative to what is currently possible.

Re: Parallella: A Supercomputer For Everyone

#88

Post by jacques_chester » Fri, Oct 12, 2012, 4:41 AM UTC

scott_s wrote: post #87

jacques_chester wrote: post #85

Earlier quoted context omitted.

Except that someone would buy $10 million of such machines and gang them together. And we would call that a supercomputer, not whatever you put under your desk. Defining it on any architecture or performance metric is just pointless because the march of time renders such things utterly moot. Remember when PlayStation 2s were "supercomputers"? Please.

Note that I said it was a sliding performance target. That is, I agree that defining it as any particular architecture or absolute performance is pointless. It's a designation relative to what is currently possible.

What doesn't slide is that supercomputers cost more.

Here's a simple test.

Suppose that I give you these two measurements:

SPECmark: 1.2 million.

SPECmark: 3 million.

Which one is the supercomputer? Without knowing the date, there's simply no way for you to tell.

Suppose instead I write:

System cost: $20 million 1990 dollars

System cost: $400 1990 dollars

Which one is the supercomputer? I think most people will be able to pick which one is which.

Re: Parallella: A Supercomputer For Everyone

#89

Post by scott_s » Fri, Oct 12, 2012, 3:34 PM UTC

jacques_chester wrote: post #88

scott_s wrote: post #87

Earlier quoted context omitted.

Note that I said it was a sliding performance target. That is, I agree that defining it as any particular architecture or absolute performance is pointless. It's a designation relative to what is currently possible.

What doesn't slide is that supercomputers cost more. Here's a simple test. Suppose that I give you these two measurements: SPECmark: 1.2 million. SPECmark: 3 million. Which one is the supercomputer? Without knowing the date, there's simply no way for you to tell. Suppose instead I write: System cost: $20 million 1990 dollars System cost: $400 1990 dollars Which one is the supercomputer? I think most people will be ab…

Relative performance is how people who design and use supercomputers define them. Cost is a secondary effect because if you're going to shoot for the limits of what we can do, it's going to be expensive.

The major difficulty with your second list is that expensive computers don't need to be high performance. Consider the computers that go into satellites and spacecraft. They are extremely expensive, but not high performing.