That is why I keep saying that a G5 (let's say PA-Semi's dual core) should be the MAIN CPU on a motherboard and the Cell should be a co-processor.
The Cell isn't 100% instruction compatible with a G4 nor G5, I believe. Does the Cell have Altivec?
I continue to harp on about Altivec, simply because people were led to believe that since it's inclusion, that it was there for the long haul, and are coding with that segment of the CPU in mind.
Support Amiga Fantasy cases!!! How to program: 1. Start with lots and lots of 0's. 10. Add 1's, liberally. "Details for OS 5 will be made public in the fourth quarter of 2007, ..." - Bill McEwen Whoah!!! He spoke, a bit late.
That is why I keep saying that a G5 (let's say PA-Semi's dual core) should be the MAIN CPU on a motherboard and the Cell should be a co-processor.
This idea sounds strange, because there is already a PPC-processor inside the CELL. I also don't think that this architecture works. It doesn't make sense and is too expensive.
Quote:
The Cell isn't 100% instruction compatible with a G4 nor G5, I believe. Does the Cell have Altivec?
Yes, the CELL has Altivec, because there is a PPC unit inside the CELL.
Quote:
I continue to harp on about Altivec, simply because people were led to believe that since it's inclusion, that it was there for the long haul, and are coding with that segment of the CPU in mind.
As nbache already said, it's all a matter of using it or taking the time to learn coding for fully optimizing for the CPU. The times where the programming peoples knew every register or particularity of the CPU are over I think. It all depends how good the development tools are and how good they implement every possible optimization. Maybe it is best to look at Apple Macintosh when it still was PPC. Only very few programs had full Altivec optimizations! It wasn't used that much from what I read. Now think of the AmigaOS, there are even less peoples who can and will do that. It's all nice to have I think, but will there really be that much real-world advantages in the end?
That is why I keep saying that a G5 (let's say PA-Semi's dual core) should be the MAIN CPU on a motherboard and the Cell should be a co-processor.
Cell as a co-processor? What would be that use for? It would be the same thing like you have a G3 as main processor and a G4 as coprocessor.
Quote:
The Cell isn't 100% instruction compatible with a G4 nor G5, I believe. Does the Cell have Altivec?
It is a stripped down G5, instruction compatible on user level. (Although it is lot slower than G5, as far as I heard, because many instructions were implemented via microcode.) It does have AltiVec unit, but the SPEs are lot more capable of doing the data crunching instead of AltiVec. (IBM calls it VMX, because AltiVec is the registered trademark of Freescale, transferred from Motorola.)
Of course, you could always port it to "just" the PPE (I think that's what the PPC compatible part of the CELL is called), but that would be a real waste, wouldn't it? Sort of like driving your 4.2 litre SUV half a kilometer down the road to the baker's and back Sunday afternoon? (Oh wait, people actually do that nowadays ...)
Well, the PS3 will be available in millions of units and porting OS4 to it is possible.
It still would not be a cheap box and the display for it costs a little more than a generic 17" TFT. Due to the restriction of the hypervisor and that it's not that easy to make good use of the SPE's it would not be as fast as it could be, at least not from the start.
But still, the hardware would be available to buy from a major player in the business for several years to come.
I see only two real problems right now. The one is a company that is believed to still exist somewhere for real and which may need to be asked if they want to make some money for each unit sold although they had nothing to do with OS4 at all.
The other problem is how to ensure that only licensed copies are installed but Sony may already have the answer to that question.
There is no helpful features such as prefetching, branch prediction etc. as has been the norm on previous generations of modern CPUs. Everything has to be done by the programmer.
There is no helpful features such as prefetching, branch prediction etc. as has been the norm on previous generations of modern CPUs. Everything has to be done by the programmer.
or by the compiler (?).
Yeah, that's what I would have expected too. But I'm only going by what I heard yesterday, and it was quite clear that they were using C, but still had to plan out by themselves how to distribute and schedule the various work units optimally between the various threads.
Maybe more help will become available with later generations of compilers? But more waiting time for tools before being able to start such a project is not exactly helpful either.
I didn't hear all of it, but part of it was a walkthrough of the history of the POWER architechture and how it led up to the CELL. In some ways similar to the talks Mikael Haglund has been giving at the last couple of AmiGBGs. I couldn't help thinking that he should be persuaded to invite her over for the next one to assist/replace him.
At one point during the talk, she said something like: "You know, I'm really not happy doing the presentation on this Windows laptop. Why don't we switch to some decent hard- and software?" and pointed to a PS3 down in the middle of the hall running Linux, which then took over for the rest of the slides and some live demos. There was a cheer from the crowd not unlike what could have been imagined if Amigans had been present at that famous occasion way back when IBM(?) used Scala on an Amiga for an important business presentation (I forget the exact occasion).
She did also show a couple of neat demos, one where a terrain was rendered in real time (Vista-like) into several windows, one was only using the PPE doing app. 1 fps, another had Altivec switched on to give 3 or 4, I think, and finally it got all the various units and pipelines working together to boost it to a smooth 30 fps. I'm not sure if it was pre-recorded, or they actually ran them all at the same time (or it could of course have been a total fake, but ...). Anyway, quite impressive.
Another one was a Julia Set calculation where the result was shown as a reflective, ever-changing "blob" sort of thing, floating in mid-air in a stylish room with doors, windows, columns, furniture etc., which was all reflected in real time onto the "blob". Could probably have dropped some jaws at a demo party ...
Edit: But of course I couldn't help sitting there wondering what might have happened if the OS4 project had been a couple of years quicker and funding had been abundant and so on ... could it have been OS4.x running those demos? (Sorry for the fantasy warp; we shall now promptly return your to your own reality.)
IBM must have had Linux on the PS3 without the Hypervisor or could the PS3 do a demo like that with all the Sony limitations? If the latter is the case I think the PS3 would be more than good enough for OS4.
Well, maybe what that video shows was the basis for what I saw, but they had it running in several windows simultaneously. The screen was more or less evenly divided in quarters, and the top right had the slow PPC version running all the time for comparison. Bottom right was a schematic drawing of the CELL's various units where (at least it looked like) she could ad hoc switch them on and off (changing their colour in the schematic) with consequential effect on the bottom left window, which with everything switched on ran at the full 30 fps, but basically the same render. It might very well have been the Mt. St. Helens one also show in the YouTube video, but I'm not sure.
Quote:
IBM must have had Linux on the PS3 without the Hypervisor or could the PS3 do a demo like that with all the Sony limitations? If the latter is the case I think the PS3 would be more than good enough for OS4.
Sorry, don't have a clue there :-} Hypervisor? It did run Linux, but apart from that I don't know about the details.
IBM must have had Linux on the PS3 without the Hypervisor or could the PS3 do a demo like that with all the Sony limitations? If the latter is the case I think the PS3 would be more than good enough for OS4.
I think developing 3D acceleration using the SPUs is possible and most likely it might work even better than the actual (outdated) 3D cards work with OS4. However, it would not reach the level what the full access to the PS3 hardware could allow. Not to mention that it is a tough job. (This is why Sony was not afraid to let the foreign OSes using the SPUs, you can do games, but won't get even close to the real thing.)
Is it "official" now, that Sony leaves out the PS1 backward compatibility chip (EmotionEngine) due to pricing for all consoles sold in Europe?
I don?t think it?s going to have any major effect, if you have FPE installed you know it really amazing how well it works on an AmigaOne 800Mhz, now image a multi core CPU like CELL whit all it?s SPE, PPE cores, I do not think you have any problem emulating a Playstation 1 rally.
(NutsAboutAmiga)
Basilisk II for AmigaOS4 AmigaInputAnywhere Excalibur and other tools and apps.