Reply by stev...@adiengineering.com●August 28, 20092009-08-28
> > Intel is trying to break into the embedded industry again, so they have
> > been doing a few things differently to try to appeal embedded designers=
.
> > For instance several versions of the Atom will be guarenteed to be
> > available for at least five years.
>
> Hasn't Intel made that promise before? =A0As I recal they lost that marke=
t
> for a very good reason.
Actually, Intel commits to keeping chips on their "Embedded Roadmap"
in production for 7 years. They have an excellent history of meeting
this commitment. If you are concerned about long production life,
make sure you stick to their Embedded Roadmap devices.
BTW, the Embedded Roadmap XScale CPUs that "lost that market" as you
say are still in production as their time isn't up yet.
Reply by Boon●July 21, 20092009-07-21
dawydiuk wrote:
> I'm not familiar with the architecture of the Atom, or what tricks
> Intel has come up with recently.
>On Jul 15, 11:42=A0am, Theo Markettos <theom
>+n...@chiark.greenend.org.uk> wrote:
>> dawydiuk <dawyd...@gmail.com> wrote:
>>
>> > > It's a bit like comparing a 386 with a Pentium. Atom would indeed
be
>> > > about twice as fast at the same clockspeed, but the ARM9 would use
>> > > significantly less power even if you clocked it faster to get the
sam=
>e
>> > > performance.
>>
>> > Interesting, this is exactly what I was looking for.
>>
>> I can't find the report to cite, but there's one kicking around that
says
>> the above... but it's rather meaningless in that you can't actually
clock
>> that particular ARM that fast. =A0It's a bit like saying a Ford will
beat=
> a
>> Bugatti at fuel consumption at 250mph... only none of Ford's cars can
>> actually go that fast at the moment.
>>
>> Of course, if the ARM's performance is sufficient the above suggests
it'l=
>l
>> win over an underclocked Atom. =A0Or maybe you can find another ARM to
ru=
>n at
>> a sufficient rate to match the Atom.
>>
>> Theo
>
>Hi,
>
>I did read about an ARM9 running @ 1 GHz a while ago, not too much
>information but a 2 page product brief was there:
>http://mcu-related.com/the-news/3-newsflash/50-marvell-55nm
Thats the family of ARM9s I was referring to :)
Regards,
Eddie
Reply by An Schwob in USA●July 19, 20092009-07-19
On Jul 15, 11:42=A0am, Theo Markettos <theom
+n...@chiark.greenend.org.uk> wrote:
> dawydiuk <dawyd...@gmail.com> wrote:
>
> > > It's a bit like comparing a 386 with a Pentium. Atom would indeed be
> > > about twice as fast at the same clockspeed, but the ARM9 would use
> > > significantly less power even if you clocked it faster to get the sam=
e
> > > performance.
>
> > Interesting, this is exactly what I was looking for.
>
> I can't find the report to cite, but there's one kicking around that says
> the above... but it's rather meaningless in that you can't actually clock
> that particular ARM that fast. =A0It's a bit like saying a Ford will beat=
a
> Bugatti at fuel consumption at 250mph... only none of Ford's cars can
> actually go that fast at the moment.
>
> Of course, if the ARM's performance is sufficient the above suggests it'l=
l
> win over an underclocked Atom. =A0Or maybe you can find another ARM to ru=
n at
> a sufficient rate to match the Atom.
>
> Theo
Hi,
I did read about an ARM9 running @ 1 GHz a while ago, not too much
information but a 2 page product brief was there:
http://mcu-related.com/the-news/3-newsflash/50-marvell-55nm
If fast software development on a Windows platform is of the most
importance I would definitely go for the Atom, if not, probably not.
However, I would check out options with a Cortex-A8 to get to a
comparable performance to the Atom.
An Schwob
Reply by Theo Markettos●July 15, 20092009-07-15
dawydiuk <dawydiuk@gmail.com> wrote:
>
> > It's a bit like comparing a 386 with a Pentium. Atom would indeed be
> > about twice as fast at the same clockspeed, but the ARM9 would use
> > significantly less power even if you clocked it faster to get the same
> > performance.
>
> Interesting, this is exactly what I was looking for.
I can't find the report to cite, but there's one kicking around that says
the above... but it's rather meaningless in that you can't actually clock
that particular ARM that fast. It's a bit like saying a Ford will beat a
Bugatti at fuel consumption at 250mph... only none of Ford's cars can
actually go that fast at the moment.
Of course, if the ARM's performance is sufficient the above suggests it'll
win over an underclocked Atom. Or maybe you can find another ARM to run at
a sufficient rate to match the Atom.
Theo
Reply by Don McKenzie●July 15, 20092009-07-15
Don McKenzie wrote:
> Nobody wrote:
>> On Sat, 11 Jul 2009 10:40:08 +0200, Dombo wrote:
>>
>>>> In 1943, it was reported that the world would only ever need a
>>>> maximum of 5 personal computers.
>>> I doubt that in 1943 they were even considering the concept of
>>> *personal* computers.
>>
>> Yep; it was "5 computers", period.
>
> ohhps, correct, take that back.
>
> I didn't check the words that are coming out of my mouth (or keyboard
> for that matter) :-)
>
> Don...
>
> "Juergen Beisert" <jbeisert@netscape.net> wrote in message
> news:h3hfa7$mhk$00$1@news.t-online.com...
>> dawydiuk wrote:
>>> We are considering using an upcoming version of the Intel Atom SoC, I
>>> believe the current code name is tunnel creek. Anyway, my boss asked me
>>> the other day how we could expect the Intel Atom to perform in
>>> comparison to a comparable ARM9(same amount of cache, clock, bus to RAM,
>>> amount of RAM). I was a bit stumped, another engineer said he would
>>> expect the Atom to blow away a comparable ARM9, he said he would expect
>>> it to be two to three times faster. I'm not familiar with the
>>> architecture of the Atom, or what tricks Intel has come up with
>>> recently. It's been about five years since my computer architecture
>>> courses in school. Can anyone done any benchmarks, or could anyone
>>> comment from a theoretical point of view?
>>
>> We are working here with many different ARM CPUs and the Atom on the
>> other side.
>> Yes, the Atom is a fast CPU. But not as fast as we expected it, and needs
>> _much_ more electrical power than we expected! Due to the BIOS and legacy
>> hardware support it also fights with System Management Mode software
>> against incompatibilities. So, it does a really poor job when we try to
>> use a realtime Linux kernel (RT-Preempt) on it.
>> On the other side we are using some 400MHz ARMv6 CPUs. Due to their
>> modern SDRAM interface, second level cache and hardware floating point
>> you can compare their computing power with the Atom. Really nice CPUs,
>> and very "cool", even if they run our test suites with 100% load.
>
> That's interesting. Did you do direct benchmark comparisons with ARM11?
Various tests with scheduler load, filesystem load and memory load (based on
a 2.6.29.5-rt22 kernel)
Latencies:
ARM11 11us .. 73us
Atom 61us ... 284us (with a bad BIOS up to 58406us)
jbe
Reply by Wilco Dijkstra●July 14, 20092009-07-14
"Juergen Beisert" <jbeisert@netscape.net> wrote in message news:h3hfa7$mhk$00$1@news.t-online.com...
> dawydiuk wrote:
>> We are considering using an upcoming version of the Intel Atom SoC, I
>> believe the current code name is tunnel creek. Anyway, my boss asked me
>> the other day how we could expect the Intel Atom to perform in comparison
>> to a comparable ARM9(same amount of cache, clock, bus to RAM, amount of
>> RAM). I was a bit stumped, another engineer said he would expect the Atom
>> to blow away a comparable ARM9, he said he would expect it to be two to
>> three times faster. I'm not familiar with the architecture of the Atom, or
>> what tricks Intel has come up with recently. It's been about five years
>> since my computer architecture courses in school. Can anyone done any
>> benchmarks, or could anyone comment from a theoretical point of view?
>
> We are working here with many different ARM CPUs and the Atom on the other
> side.
> Yes, the Atom is a fast CPU. But not as fast as we expected it, and needs
> _much_ more electrical power than we expected! Due to the BIOS and legacy
> hardware support it also fights with System Management Mode software
> against incompatibilities. So, it does a really poor job when we try to use
> a realtime Linux kernel (RT-Preempt) on it.
> On the other side we are using some 400MHz ARMv6 CPUs. Due to their modern
> SDRAM interface, second level cache and hardware floating point you can
> compare their computing power with the Atom. Really nice CPUs, and
> very "cool", even if they run our test suites with 100% load.
That's interesting. Did you do direct benchmark comparisons with ARM11?
Wilco
Reply by Juergen Beisert●July 14, 20092009-07-14
dawydiuk wrote:
> We are considering using an upcoming version of the Intel Atom SoC, I
> believe the current code name is tunnel creek. Anyway, my boss asked me
> the other day how we could expect the Intel Atom to perform in comparison
> to a comparable ARM9(same amount of cache, clock, bus to RAM, amount of
> RAM). I was a bit stumped, another engineer said he would expect the Atom
> to blow away a comparable ARM9, he said he would expect it to be two to
> three times faster. I'm not familiar with the architecture of the Atom, or
> what tricks Intel has come up with recently. It's been about five years
> since my computer architecture courses in school. Can anyone done any
> benchmarks, or could anyone comment from a theoretical point of view?
We are working here with many different ARM CPUs and the Atom on the other
side.
Yes, the Atom is a fast CPU. But not as fast as we expected it, and needs
_much_ more electrical power than we expected! Due to the BIOS and legacy
hardware support it also fights with System Management Mode software
against incompatibilities. So, it does a really poor job when we try to use
a realtime Linux kernel (RT-Preempt) on it.
On the other side we are using some 400MHz ARMv6 CPUs. Due to their modern
SDRAM interface, second level cache and hardware floating point you can
compare their computing power with the Atom. Really nice CPUs, and
very "cool", even if they run our test suites with 100% load.
jbe
Reply by whygee●July 13, 20092009-07-13
Robert Adsett wrote:
> In article <p9WdnRLJO-hJGMbXnZ2dnUVZ_sydnZ2d@giganews.com>, dawydiuk
> says...
>>> Another metric for your boss: how long does Intel usually keep a chip
>>> in production?
>> Intel is trying to break into the embedded industry again, so they have
>> been doing a few things differently to try to appeal embedded designers.
>> For instance several versions of the Atom will be guarenteed to be
>> available for at least five years.
>
> Hasn't Intel made that promise before? As I recal they lost that market
> for a very good reason.
As a general rule, promises engage those who believe them.
I don't do any promise anymore, unless I am paid ...
And even then, it's hard.