Intel Atom vs ARM
|
Prev: NTP: (Was: Re: performance of hardware dynamic scheduling)
Next: A race-condition in a SUN paper by Detlefs on reference counting...
From: Wilco Dijkstra on 4 Apr 2008 17:27 "Jon Harrop" <jon(a)ffconsultancy.com> wrote in message news:_YWdnaI41rM3t2vanZ2dnUVZ8sylnZ2d(a)plusnet... > Wilco Dijkstra wrote: >> "Jon Harrop" <jon(a)ffconsultancy.com> wrote in message >> news:M76dnT7jTNiCnGvanZ2dnUVZ8v6dnZ2d(a)plusnet... >>> >>> In the light of recent discussions, I thought this article would be of >>> interest: >>> >>> http://www.linuxdevices.com/news/NS5492118276.html >> >> See also: http://laptoping.com/intel-atom-benchmark.html which >> shows Atom/Silverthorne is less than half the speed of a Celeron-M... > > So the quad core ARM Cortex is 2x smaller, uses 1.5x less power and is twice > as fast according to the competition's benchmark results? Impressive. Indeed. Intel's benchmarks claim that a 1.6GHz Atom is only 4-6.5 times faster than a single-issue 400MHz ARM11. Cortex-A8 and A9 are twice as fast as ARM11, so by Intel's own admission the latest ARM cores are 1.25-2x as fast as Atom at the same frequency... (which is what I predicted a while ago). If you compare the SIMD capabilities then ARM's advantage increases even further. So with a quad core Cortex-A9 at 800MHz you would get 2.5-4x the performance of a single core Atom at 1.6GHz while still using less power... > How much cheaper is the ARM? ARM's are definitely much cheaper, but it is hard to compare like with like. Eg. OMAP is a "single-chip smart phone". It includes everything you need, a memory controller, LCD controller, graphics accelerator, sound, CMOS camera interace, on-board memory, etc. Samsung sells similar phone chips with many MBytes of flash and DRAM in the same package. The Atom is needs additional chips for all of this, which increases the cost, board size and power consumption. > I think Intel are really sticking their necks out by making statements like: > > "Thus, the average power consumption of Atom is lower" - > http://techon.nikkeibp.co.jp/english/NEWS_EN/20080403/149992/ That's pure marketing spin and then some... Atom's lowest idle power is 100mW!!! - this is in the deepest sleep mode doing absolutely nothing! That drains most batteries rather quickly... Compare this with the idle power of ARMs which is typically well below 1mW. Wilco
From: Wilco Dijkstra on 4 Apr 2008 17:40 "Elcaro Nosille" <Elcaro.Nosille(a)googlemail.com> wrote in message news:ft66um$uk0$1(a)worf.visyn.net... > Paul Gotch schrieb: > >> No because routers/network equipment want big endian data >> for efficently handling IP packets which is why a hell of >> a lot of them are MIPS or PowerPC since ARM only recently >> got a byte invariant big endian mode. Now it has one you >> are starting to see ARMs tip up in some neworking equipment. > > Do you really think that a BSWAP per word would make a > noteworthy overall performance-difference in IP-handling? > I don't think so. It's not the BSWAP that is a big issue (although it does increase power and memory bandwidth if you need to swap all traffic), it's the software that is written to work only on big-endian systems. If porting it was easy there would be no need to support efficient swapping between little and big-endian applications like ARM does. Wilco
From: Peter Matthias on 5 Apr 2008 05:51 Wilco Dijkstra wrote: > it's the software > that is written to work only on big-endian systems. There is software that is written for big-endian that is not ported to X8 (-64)? Peter
From: Nick Maclaren on 5 Apr 2008 06:15 In article <ft7i3c$pbf$1(a)news.albasani.net>, Peter Matthias <Peter.Matthias(a)ivalid.invalid> writes: |> Wilco Dijkstra wrote: |> |> > it's the software |> > that is written to work only on big-endian systems. |> |> There is software that is written for big-endian that is not ported to X8 |> (-64)? Yes. As with most such unportable software, a closer inspection almost always shows that it is either intrinsically system-specific (in which case, so what?) or so badly written as to be unmaintainable and riddled with bugs. The same applies to almost all x86-only software, of course. Regards, Nick Maclaren.
From: Wilco Dijkstra on 5 Apr 2008 07:03
"Elcaro Nosille" <Elcaro.Nosille(a)googlemail.com> wrote in message news:ft68u0$vkr$1(a)worf.visyn.net... > Wilco Dijkstra schrieb: > >> It's not the BSWAP that is a big issue (although it >> does increase power and memory bandwidth ... >> if you need to swap all traffic), ... > > There are only a few IP-fields which need to be swapped. > So the increase in power-consumption and memory-bandwith > isn't notheworthy. It depends on the kind of router, if it is a high end one the overhead might well be noticeable. In cases it isn't, I still don't think a $50 x86 will be used when a $5 MIPS/ARM does the job just as well... >> ... it's the software that is written to work only on >> big-endian systems. > > No one would implement his own IP-stack as there are > well-established and solid implementations. Exactly. And not all of this software is x86 based. Wilco |