By which I mean... I do not understand what technology it is based on. Not that I have no clue what an Intel Atom is. I have a single core n455 Atom.

What confuses me is that the Atom is described six different ways. I know the single core Atoms benchmark lower than equivalent Pentium III's. I know that Atoms use in order instruction execution.

But then I read....

The Atom uses the Pentium I architecture as its base since it uses in order execution.
Then, the Atom used Pentium III architecture as its base which is why it benchmarks like it.
Then the Atom uses it's own architecture as its base.
The Atom can only address up to 2GB of RAM. OK, Intel's 32 chips could only address 4GB of RAM. So it's a 32 Bit chip, right?
But it supports 64 bit instructions. So that means it uses at least the latter P4 as it's base, especially as it supports hyperthreading?
Others say that the Atom is a really, really cut back version of an i3, like the Celeron but an i3 is built around out of order execution so the Atom does not fit that.

A minimal chip like the Atom is normally a cut down version of a higher end chip but that does not seem to be the case.

I am confused. Is the Atom A P1, II or III with bits added, or is it a cut back i3 or whatever or is it truly it's own chip?

This sort of thing bothers me and it means I can't describe what an Atom is to prospective buyers – either for the chip or against. [Bear in mind low salaries in Indonesia.]

Can anyone tell me what the Atom is? It's seems a weird mixture.

Many thanks.

http://en.wikipedia.org/wiki/Intel_atom

http://en.wikipedia.org/wiki/Stealey...roprocessor%29

Also see:
http://en.wikipedia.org/wiki/Bonnell...rchitecture%29

For sure their performance is less than the i3. However, they are great for netbooks and other mobile devices. I think they are now trying to compete with ARM, with variable levels of success. I definitely would choose Atom over ARM, because Atom is x86, and because ARM tends to be locked in.

Thank you.

I had read the Wikipedia on the Atom and it was still unclear to me.
It's base is Pentium-M then?

Yes, it is derived from the Pentium-M.

Ah. Many thanks. That clears my brain!

Cheers!

Max RAM is not just CPU limited, its chipset limited. Atoms can go over 2GB easily, the NM10 chipset supports 4GB and I've heard of people getting 8GB working with some motherboards..

Its pretty much its own chip.

For the original single core 'silverthorne' atoms, which is the only atom I've ever seen compared to a pentium M head to head, yes.

Newer Atoms are faster than the old models, and may well be faster than a similar clocked P3 now.

BTW, the newer 'cedarview' atoms (N2XXX) use intel GMA video with really awful linux support, and are not really suitable for linux. IMO anyway.

That depends a bit on what you mean by equivalent. One definition would be '...of equivalent performance...', and, by that definition, an Atom would (should) perform exactly the same as equivalent pentiums (if the benchmarks reflect what you mean by performance, which they should). other definitions might be power consumption, area, transistor count or clock speed based.

Most of the above are probably at least a bit true, if you take the right (somewhat strange) point of view.

If you take the p-o-v that in order ex is the most significant determining factor in arch (it is at least a very significant factor, but the most? probably not), then you'd probably agree with this one. But, with a number of significant factors, and with the Atom not containing any significant bit of logic directly derived from the P I, then I'd argue that it is a real stretch. Or, to put it more plainly, wrong (even though I accept that you can make an argument for it).

The P III one is probably more common, as some people think that pre-fetch queue length is the most significant thing, and may even be all of arch (that is, if the queue length is 'n', the part has the same arch as all other length 'n' parts, and that's obviously rubbish, when stated that plainly, but is an implicit assumption that is quite common). This is part of the argument that the P III was wonderful because it had a relatively short queue and that the world ended when the P IV was released because it did not have a short queue. This argument sometimes goes on to say that all long queue parts are bound to be rubbish, because they have a long queue, although it is never quite explained why short queue parts aren't uniformly wonderful or why zero queue parts have died out (for this type of computation) years ago, because if it were that simple, these should be even more wonderful.

Assuming by arch you mean something more than 'instruction set(s)', then "the Atom uses it's own architecture as its base" is closer. From my understanding of how Intel did their work, it would be closer to say that they sat down and asked 'with our objectives (power, performance, chip area) can we afford (whether that be any of the perf characteristics) some particular feature (eg, out of order execution - and the answer to that has been no, at least for the earlier generations of Atoms - I think the next gen is scheduled to have oo, but I'm probably losing track of generations).

Well, I don't think that they used the P4 as its base - essentially, for the first generation, they sat down with the Core2Duo and said 'What out this can't we afford, or is inappropriate (based partly on the answer to the previous question)?' Now, this almost certainly did make the arch, at certain levels, look a lot more like something like a P III (or even a P-M), but that doesn't mean that it was actually based on either of those parts. 'Based' is a difficult word here.

In a way, that's probably the closest, in that procedurally that's probably close to what they did. The trouble is, by the time that they had made the modifications that they needed to, because of the stuff that they couldn't afford, the outcome wasn't much like the base. You could say

• that the outcome was more like some other chip, but:
• that doesn't make it based on that other chip (whichever one that is)
• even then, whichever other chip you choose, there would be differences that were very significant differences, from the p-o-v of architecture

I'd have to do some research on that, but do realise that there is no logic in that statement.

It is certainly not a strict subset of some higher end chip, which is what you might have expected. You might describe it as 'mix 'n match', but that wouldn't be completely true either - I suppose, if you kept to very 'broad brush' descriptions of arch, you'd probably be able to say it was a simple cut 'n paste job, even though that's not really true.

If you look in detail, it is its own chip, but when Intel designed it (errr, them, which is an important point) they worked with elements with which they were familiar and comfortable.