Excitement about NatAmi

[AJCopland]

I thought I'd chip in and say something since, despite my lack of time to really chip in very much, I am still on the Natami team.

Everyone seems to think that the 68050/070 or 3DCore development has delayed the release somehow. They're completely separate.

The 3DCore has a lot of research and going back to the books because, especially for me, it was the first time we'd tried to really plan out something like a GPU. Gunnar has an existing rasteriser with texture mapping but we wanted something more powerful. In effect we have a version 1.0 of the 3DCore, and most of our research and planning is for a version 2.0.

NEITHER of those 3DCore's are likely to be in the first release because it could delay launching the Natami and no-one wants that.

The 68050 however is a very advanced project that Gunnar, Jens and AndinG have all been working on. That doesn't mean it will be ready for launch but again that doesn't mean it won't be used later on just like the 3DCore.

Gunnar has also written a couple of games whilst all this has happened!

All of this has happened in parallel and without interrupting Thomas who has worked on nothing but the board designs and SAGA implementation. If you looked at the sites hardware page:
http://www.natami.net/hardware.htm
You can see that he has used a real 68060 for testing and development of the LX board. If you've read the forums you'd also know that the MX board will be able to use one as well but that it will be optional so that the Natami can be made cheaper when the 68050 is good enough (which it nearly is now).

If the 3DCore(s), 68050/070 etc didn't exist then Thomas would still be at this stage now with the board development because he has a life and job outside of this and is doing the hardware design and SAGA development alone.

Everyone else is the community/team support for other things, such as acquiring and testing coldfire dev boards, setting up the site, and for those who live locally like Gunnar just being a helpful friend who keeps things going and keeps all of the flame wars off his back.

All of the boards right from the first one have been about advancing the idea to find it's flaws and refine it. Thomas doesn't want to release a broken and half-arsed implementation he wants to do it properly so the Natami30, LX and the two CPU cards have all been about figuring out what that really means. That doesn't include a lot of the stuff that doesn't get shown or talked about outside of IRC and the team pages.

I've had doubts about my own ability to contribute but I do NOT doubt Thomas's desire to finish and release Natami with SAGA at all. Or Gunnar's dedication to the project, nor Jens desire to get the N68050 running in the final machine.

Andy

[AJCopland]

I'd say in some ways the Cell is more "multi-core" than other CPUs. Those SPEs run autonomously to each other. The only real interaction between them is explicit DMA between memory. Other than that, those CPUs can do pretty much any kind of computation other CPUs can, except they are pretty short on local memory.

In order to leverage all those Cell SPEs you have to design your software architecture and code REALLY well (e.g. tasks/jobs instead of the traditionally used threads). A byproduct of this is that the task/job code you design for this is usually easier to scale up to hundreds of cores as compared to a threaded model.

Once you have code that will run on Cell, it is relatively easy to port it to the kind of multi-core you find on 360 and PC. The other way around is not true though if you didn't take PS3 into account. So in some ways while I hate Cell, it is also probably a good heads up on how software needs to be designed moving forward. So I think tech will be more ready for hundreds of core than we might think, if that ever happens. Maybe with the Epic thing, they aren't sure if the hardware will be powerful enough.

No-one is going forward with a design like the Cell for any platform, it's an evolutionary dead end. Better to spend more on the CPU than be stuck with something so awkward to develop for. Besides tasks/jobs are a common way of feeding multiple threads which is simply what the task/job system for the SPEs hid from you by using the SPURS library.

Instead of SPEs we'll see more fully formed AMP systems where the SPEs can actually access main memory and handle more than 32k in or out at once.

Anyway, not that this has much to do with Natami. Thinking of SMP on there seems very ridiculous at this point when its not even being done on AROS/Morphos/AOS4.x. I think if anything it would make more sense to extend the CPU with SIMD first.

This is the relevant part, extending it with SIMD, pipelining the N68050/70 better, making it OoO etc are probably all  a better idea right now than making the system SMP capable.

There's good argument for having another CPU to offload tasks to like the GP2X's second ARM cpu be used for or the PlayStation2's IOP. That's more like treating the second cpu as a co-processor.