Sony's Cell chip for a New Amiga

[psxphill]

Yeah, that did come across as a bit harsh... But yeah, I'm not a fan, the actual CPU core is pretty weak and of little use for desktop work.
 
The Xbox 360's CPU is better,

You do know that it's the same CPU? The 360 has 3 of them, while the PS3 has one plus the SPE's. The book about how the CPU was designed is a very interesting read.
 
But yeah, it's a high clock speed that doesn't necessarily translate into high speed.
 
The ARM uses less power and achieves more work per clock, but the atom can be clocked higher. Software compatibility on the atom more than makes up for it's drawbacks, especially on a desktop machine where power the power is less of a concern. Even on battery my atom netbook lasts 10 hours, which is more than enough.

[Hattig]

The ARM uses less power and achieves more work per clock, but the atom can be clocked higher. Software compatibility on the atom more than makes up for it's drawbacks, especially on a desktop machine where power the power is less of a concern. Even on battery my atom netbook lasts 10 hours, which is more than enough.

Software compatibility? For AROS, sure, but not for AmigaOS! Other operating systems - Linux is on ARM and x86, Android is ARM and soon x86.

Samsung will have their 2.0GHz ARM Cortex A15 dual-core out early next year - that will outperform even the fastest Atom (event the 2.13GHz ones). The ARM ecosystem is simply far more interesting at the moment.

[bloodline]

Software compatibility? For AROS, sure, but not for AmigaOS! Other operating systems - Linux is on ARM and x86, Android is ARM and soon x86.

Samsung will have their 2.0GHz ARM Cortex A15 dual-core out early next year - that will outperform even the fastest Atom (event the 2.13GHz ones). The ARM ecosystem is simply far more interesting at the moment.

AROS is also available for ARM

[psxphill]

Software compatibility? For AROS, sure, but not for AmigaOS! Other operating systems - Linux is on ARM and x86, Android is ARM and soon x86.
 
Samsung will have their 2.0GHz ARM Cortex A15 dual-core out early next year - that will outperform even the fastest Atom (event the 2.13GHz ones). The ARM ecosystem is simply far more interesting at the moment.

Android has been on x86 for ages.
 
Next years ARM might beat this years atoms, but next years atoms will be better.
 
X86/x64 binary compatibility is more interesting to me, so ARM would have to come out with something amazingly better for me to even consider it.
 
ARM and Linux is fine for devices like NAS or TV etc, running Windows 7 on my A1200 would rock (ideally you'd have a graphics chip built into which ever cpu you had in the trapdoor).

[Hattig]

The sad thing is that the Atoms that have on-die graphics are using some pretty old tech - PowerVR SGX535s or 540s I believe - think iPhone 3 but running a bit faster.

The PowerVR 6 series is far more advanced, as is the ARM Mali T604. You might even get drivers for the latter, good luck getting open source drivers for the PowerVR on Atom chips.

As regards to next year's Atoms ... Cedar Trail is next year's Atoms now, delayed several months already because Intel can't get the graphics drivers right.

Nah, I'll stick with the proven ARM tech myself. I don't need x86 for anything. Windows 8 will come in an ARM variant if you really want to run that.

[commodorejohn]

I'm no expert on the subject, but the Cell architecture just looks unnecessarily complicated. Software-directed caching means you really need to know what you're doing and have a good compiler - and while that more or less works for a system that's fundamentally only going to be running one program at a time, trying to coordinate that extra complexity across a whole desktop OS's worth of different programs and threads sounds like a nightmare if I ever heard of it.

[Digiman]

Software compatibility? For AROS, sure, but not for AmigaOS! Other operating systems - Linux is on ARM and x86, Android is ARM and soon x86.

Samsung will have their 2.0GHz ARM Cortex A15 dual-core out early next year - that will outperform even the fastest Atom (event the 2.13GHz ones). The ARM ecosystem is simply far more interesting at the moment.

Windows 8 will also have a release on ARM.

But both are to pedestrian compared to £400 i7 PC right now let alone in the future years.

IMO just convert OS4 to x86-64bit and be done with it.

[Iggy]

Off topic alert -
 
Sin after sin, I have endured. Yet the wounds I bear, are the wounds of love.

God bless K.K. Downing!
 
Whey hey - Judas Priest fan eh?
 
One of the most underrated Brit metal bands ever!

Absolutely one of my favorites!

Just saw them in Reading PA.

I was worried that K.K. Downing's departure would really messy up things, but Ritchie Faulkner pretty good (he's not Kenny, but he's really enjoying himself).

[bbond007]

...do you think it's possible? Discuss!

A CELL trapdoor accelerator for an A1200 would be awesome.

Difficult to program or not the SPUs would make up for AGAs missing features! Maybe it could do C2P really fast

[Iggy]

The actual CPU is quite slow per clock, only made up in part by it's high clocks. It might run at 3.2GHz but it performs like a contemporary 1.6GHz CPU. It does have the ability to run two threads at the same time however, but that's not much use for AmigaOS currently.

Bingo, the Cell is significantly less powerful then most out of order PPCs on a per clock cycle basis.

I'm no expert on the subject, but the Cell  architecture just looks unnecessarily complicated. Software-directed  caching means you really need to know what you're doing and have a good  compiler - and while that more or less works for a system that's  fundamentally only going to be running one program at a time, trying to  coordinate that extra complexity across a whole desktop OS's worth of  different programs and threads sounds like a nightmare if I ever heard  of it.

I've made that argument before. Standard PPC architecture is basically  RISC. Adding SPEs and coordinating the PPE with all those coprocessors  makes things way too complicated.

Personally, I want to see Freescale's T5040 (@2.5 GHz or above).

[minator]

Cell was designed for heavy duty processing at low cost and relatively low power.
At that sort of thing it was *incredibly* fast.  On HPC benchmarks it was getting 30x gains over x86 processors.  On FFTs it was beating the P4 by 100x.

On the things it's designed for it's probably still competitive with top end x86 processors.
I have a dual core i7 MacBook Pro and I wish it had a Cell - for the audio apps I use (e.g. soft synths) it'd be far faster.


The PPE is a G4(ish) class processor, it's in-order so performance is rather erratic (it's faster at some things slower in others).

The SPEs are DSP like cores.  They require more involved programming but it's not exactly brain surgery (it's a variant of AltiVec code).  It's easier to get something up and running on the XBox 360 but to get high performance it's no easier - you use exactly the same techniques you use on the SPEs.


Cell is rather dated now but it fitted into the Amiga model very well.  If you remember the Amiga was powerful because of it's chipset.  The SPEs would work in much the same way.  The main CPU would run most stuff but hand the heavy lifting to the SPEs.

However after Sony's lies about keeping their platform open, there's no more low cost Cells. There's no point even looking at them now.

[minator]

At 6 years old, I'll bet the new quad core ARM SoCs would out perform it using much less power.

The Tegra 3 has less compute resources than a single SPE so I think that's rather unlikely.

However, it's not that far away...

The A15 is likely to be faster than the PPE on general purpose code.
On raw compute performance you'll need a GPU.  ARM's Mali-T658 has a peak compute rate higher than Cell.

[Middleman]

Windows 8 will also have a release on ARM.
 
But both are to pedestrian compared to £400 i7 PC right now let alone in the future years.
 
IMO just convert OS4 to x86-64bit and be done with it.

Maybe, but what about Amithlon?
Apparently there's an 'Amithlon Revival Project' going on right now....
 
> http://code.google.com/p/amithlon-revival/

[HenryCase]

Cell was designed for heavy duty processing at low cost and relatively low power.
At that sort of thing it was *incredibly* fast.  On HPC benchmarks it was getting 30x gains over x86 processors.  On FFTs it was beating the P4 by 100x.

Thank you for being the voice of reason minator, people have short memories it seems. Cell wasn't designed to be a general purpose desktop CPU, but rather one specialised for certain tasks. Anyone remember the supercomputer clusters being built out of PS3s (when running Linux out of the box was a possibility)? People didn't just do that because they could, they did that because it was a cost effective way of getting a lot of computing power, for certain tasks.

FWIW, I don't own a PS3, I just found most of the posts in this thread devoid of understanding, and was pleased to read that at least one person had some semblance of knowledge.

The irony is that the PS1 was successful because it was easier to program than any console that had come before.

You say this as a games developer do you? From what I've heard, PS1 was not easy to program for, easier than the Saturn, but otherwise not easier than what came before.

Thing is, my view is backed up with evidence. Compare the early PS1 games to the later ones, e.g. Tekken 1 vs. Tekken 3:





Of course, for any console, more of the power is unlocked over time as people learn how to get the best out of it, but the time it takes to unlock this power is a measure of how hard the device is to program for. If the Playstation was easy to program for, then you wouldn't see much difference in the launch titles and the later titles (look at Gamecube or Xbox 1 for an example of this). By many accounts, it took the release of a tool (the DTL-H2700 Performance Analyser, used in the development of games like Gran Turismo) before developers found out how to unlock the performance of the PS1.

Am curious to know how you got the opposite impression, so feel free to fill me in.

[Digiman]

Maybe, but what about Amithlon?
Apparently there's an 'Amithlon Revival Project' going on right now....
 
> http://code.google.com/p/amithlon-revival/

Not long ago I trawled the internet about Amithlon and found an interview with the creator. Was quite sad he felt the need to wipe the improved update.

Amithlon did more for reviving Amiga in its short life than Hyperion has done to date.

Today maybe it's impossible to genuinely revive Amiga because the software written for real Amigas, like browsers/codec support etc, is 15 years old. Does it matter even?

[smerf]

...do you think it's possible? Discuss!

Hi,

No discussion needed, next monthe go to http://www.maximumpc.com and go back into their magazine archives, and read the article "The state of GPU computing". The reason I am telling you this, is because I know that most of you are to cheap to buy the mag, so you can read it for free, because they have it in pdf format on their website, why listen to a bunch of think it know it alls from around the world, when maximumpc puts it through it's paces and tells you tested results. In other words it is not hear say from a bunch of know it alls from around the world, some that have a hard time just getting electricity.

smerf