Groundhog

Ramon

#1623 mention Amdahl, and say there is no one size fits all solution.

#1982 mention Amdahl again, and say that 8 is the sweet spot.

Maybe you can explain why 8 is sweet spot (and not 10, 12, 16 or 32). What are the technical reasons? you usually provide technical reasons, right?

I am sorry, I will not wait the answer. Bye! I am tired of this gratuite bullying from some people to those that are trying to maximize parallelism, determinism and bandwith for valid real world examples.

And recently there are even people saying that what "they just want" is one fast cog. Please, don't lie. That is not what is being discussed here.

I am sorry to be rude. When we get the chip in our hands, we well meet again and be friends. But now I am too busy to read what is being said in recent posts.

I was the first one to mention about grounhog day (yes with that typo, apologize for it). So I think that is a good tittle to announce that I am now free !

Dave Hein

Ramon, rather than leave the forum just ignore the posts about P1+. Only read the posts from Chip. They are the only ones that actually mean anything. Pretend like there is no such thing as a P1+ in development until Chip posts an update on it. That's pretty much the tactic that I've accepted. Thinking about a super cog or hub-slot sharing techniques will just drive a person crazy. It's pretty clear that's what has happened to some of the forum members.

User Name

I don't pretend to know what you're sore about, Ramon. Of course that's because I've been paying almost no attention to what has transpired recently. The conversation got too intense and emotions too high, so I bugged out. Not criticizing those who stayed. Not at all! We all have our own level of comfort. So to you, Ramon, I heartily endorse your decision to do something else until the grand unveiling.

Hopefully when all is done we will have something more than just the old round robin scheme (which is a great default behavior, but come-on guys, its 2014!). I just have to trust Chip. And I do.

Heater.

Ramon,

You should not take all this internet debate so personally. From my point of view there is no "bullying" going on. Only an expressions of different points of view and, hopefully, solutions to problems. It's an interesting debate even if it leads nowhere. At the end of the day whatever Chip does is what Chip does.

Now, you are obviously referencing my comments so I will do my best to answer.

Maybe you can explain why 8 is sweet spot (and not 10, 12, 16 or 32). What are the technical reasons? you usually provide technical reasons, right?

Firstly, I don't recall saying 8 was exactly the "sweet spot", of course nothing is so cut and dried.

The thing is Amdahl's law is as real as Newton's laws of motion. Let's try a thought experiment to clarify:

Imagine you have a bunch of processors working on some problem. Let that number be N.

Clearly as they are working on the same problem they have some shared resource in common. Let's say it's just RAM. Only one of those processors can access that RAM at a time.

That implies that if any of those processors need to access the RAM they have to somehow wait in-line to get their turn to access it.

OK. So what if we have only one processor, N = 1? It works as fast as it can and has zero wait time for the shared RAM. Life is good.

What if we have two processors, N = 2? Should be twice as fast right? No, at some point they will want the same RAM at the same time and one will have to wait for the other. That causes a delay for one of them. Performance increase is therefore not *2 but something a bit less.

What if we have three, N = 3? Similarly performance is not *3 because now any one guy has to wait for the other 2.

So as we see, every time you add a processor you do not get another processors worth of performance, only some fraction of it. Eventually you are adding a processor and only getting 1/10 or 1/100 or 1/1000 performance increase.

In the extreme you have an infinite number of processors and adding one more has zero performance increase.

In fact I would argue that realistic performance goes down to zero as you approach an infinite number of processors. After a processor has made a RAM access it has to wait an infinite amount of time to make the next one. Who is going to wait for that?

At some point increasing N is a waste of silicon space, power, and money.

So, there is the "sweet spot", somewhere between 1 and infinity. Is it 4? Or 8? or 16? or a million?

Where is the "sweet spot"? You tell me.

Note: This is all true no matter what arbitration scheme is used, round robin, whatever.

Invent-O-Doc

I don't blame him. I've thought of doing the same thing myself.

jazzed

Everyone has a tolerance level. Some accept junk, some disappear, others post replies, and others post nasty replies. Ignoring irritating things is usually best.

I'm not sure what Parallax gets out of all this, but the forums can actually be very helpful when people are trying to help instead of chasing some agenda.

msrobots

well said @jazzed.

Thank you.

Mike