Can Architecture Be Liberated from the von Neumann Style? (pdf)

[Shade]

Abstract: proposed an asynchronous functional architecture that executes declarative code in the graph form and supports laziness and parallelism by default. It's shown that such an architecture has very important advantages over the von Neumann architecture.

Link to the article: github.com/shadenova/Nova/blob/main/Can Architecture Be Liberated from the von Neumann Style.pdf

[Ghideon]

15 minutes ago, Shade said:

It's shown that such an architecture has very important advantages over the von Neumann architecture.

I’m curious; what are the advantages? How do the proposed architecture differ from known methods for mitigating the Von Neumann performance bottleneck*?

 

*) if this is what OP wishes to discuss, more details may be required. I have not opened the link.

[studiot]

1 hour ago, Ghideon said:

I’m curious; what are the advantages? How do the proposed architecture differ from known methods for mitigating the Von Neumann performance bottleneck*?

 

*) if this is what OP wishes to discuss, more details may be required. I have not opened the link.

I found the 5 page pdf safe to download and interesting.

However the text reads more like a technical advertising document than a technical information document.

Long on rhetoric and short on detail.

Looks as though it could be a developemnt of mathematical flowgraph techniques, but with a call to manufacturers to implement the necessary hardware.

Doesn't the modern GPU go some way towards this goal ?

Edited September 9, 2022 by studiot

[Ghideon]

5 hours ago, studiot said:

I found the 5 page pdf safe to download and interesting.

Thanks, I might take a look at it!

 

5 hours ago, studiot said:

Looks as though it could be a developemnt of mathematical flowgraph techniques, but with a call to manufacturers to implement the necessary hardware.

Doesn't the modern GPU go some way towards this goal ?

That sounds reasonable, but I need to read more about flowgraph to have a valid opinion*. 

Recent GPU's have specific features applicable to for instance artificial neural networks, my guess is that OP is suggesting/looking for an architecture that is more generalised. Let's hope they return with some comments and details.

 

*) I've seen reports about using GPU's for graph databases, if that (at least partially) addresses your question. The programming model of GPU allows for some interesting modifications of typical algorithms used in programming a CPU, but that would be a separate thread I think. 

[Doctor Derp]

I think the last major effort to deviate from von neumann architecture was the cell processor which was co developed by sony, toshiba and IBM sometime before 2010.

There's nothing sinister or particularly remarkable about VNA. Its simply the easiest and most expedient method to connect standard computer components we have on hand. Its a basic BUS networking architecture.

[Phi for All]

On 9/9/2022 at 8:36 AM, Shade said:

Abstract: proposed an asynchronous functional architecture that executes declarative code in the graph form and supports laziness and parallelism by default. It's shown that such an architecture has very important advantages over the von Neumann architecture.

Link to the article: github.com/shadenova/Nova/blob/main/Can Architecture Be Liberated from the von Neumann Style.pdf

!

Moderator Note

We're a science discussion forum, and we aren't here to promote articles you've written elsewhere. If you wish to discuss what you've written with the rest of the members, then give us some discussion points. The article can be used as a reference, but members should be given enough information so they don't have to go offsite or click links they don't trust. So far you've posted twice, with links to articles you've written. We need you to respond as if you were discussing your ideas rather than lecturing or blogging.