Paul A. Clayton (paaronclayton@gmail.com) on 1/12/11 wrote:
---------------------------
>Wilco (Wilco.Dijkstra@ntlworld.com) on 1/12/11 wrote:
>---------------------------
>[snip]
>>Well, obviously any benchmark that shows Cortex-A9 to be faster than Atom is not
>>going to be good enough, is it? But why don't you Google for A9 benchmarks yourself? I found this in 30 seconds:
>>
>>http://www.open-chess.org/viewtopic.php?f=24&t=775&start=10
>>
>>Palm Pre 500mhz (Webos1.4.5): ~8.000 NPS
>>APAD Rockchip 600mhz (android1.5): ~10.000 NPS
>>Palm Pre oc. 800mhz (Webos1.4.5): ~13.000 NPS
>>ZT-180 10,2" Pad : ~15.345 NPS
>>Palm Pre oc. 1Ghz (Webos1.4.5): ~16.000 NPS
>>Archos 101 (android 2.1): ~78.300 NPS
>>Iphone4 800 Mhz (IOS): ~80.000 NPS
>>HTC HD2 with 1Ghz (android2.2): ~100.000 NPS
>>Samsung Galaxy Tab: ~108,500 NPS
>>Advent vega tegra2 1Ghz (android2.2): ~131.700 NPS
>>Dell Streak (Steves 1.4.5, OC 1150Mhz): ~133.000 NPS.
>>Netbook Intel Atom N270 1.6 Ghz (Windows XP): ~143.000 NPS
>>Advent vega tegra2 1Ghz 2cores (android2.2): ~220.500 NPS
>>
>>Laptop T7600 2.2Ghz 1 core, (windows vista64): 1.000.000 NPS
>
>Those numbers seem to reinforce Linus Torvalds' point.
>The highest performing (single core) ARM is noticeably
>slower than a rather ordinary Atom and an Atom is about
>a decimal order of magnitude slower than a typical desktop
>x86 processor.

The Cortex-A9 above is not a netbook CPU, however it achieves almost the same performance as a netbook CPU at a much lower frequency. That hardly supports what Linus said: "And people don't understand how big the performance gap really is. ARM is absolutely not even close. ARM right now is struggling to compete with Atom, for chissake!".

Being 8% slower is not only a tiny gap, it is Atom which is struggling to be competitive as it needs a 60% faster clock.

Anyway, the netbook/laptop version of the A9 runs at 2GHz. Ie. Atom would need to be clocked at 3GHz to give the same performance or be radically redesigned to be more efficient.

You're right today there is still a big gap with the T7600, but that gap will be narrowed by Cortex-A15.

>Here are some CoreMark scores:
>
>NVIDIA Tegra 250 (gcc 4.4.5), 2646.38 CoreMark at 1 GHz
>Intel Core i5-650 (gcc 4.4.5), 10379.90 CoreMark at 3.2 GHz
>Intel Atom N270 (gcc 4.4.1) 2 threads, 4567.25 CoreMark at 1.6 GHz
>Intel Celeron (gcc 4.1.2), 2737.85 CoreMark at 1.2 GHz
>
> I _think_ that
>Tegra uses an A8 (or derived core), so an A9 might add 25%
>to that score.

Tegra2 uses A9. The GCC score is a bit worse than the published result of 2881. Also, if you use a multi-threaded Atom then it is also fair to use a dual core A9.

>Performance per MHz or per Watt are not valid metrics
>when one is concerned with maximum performance.

Absolutely true. However the markets ARM is going after (mobile, netbook, laptops, low power servers) are not at all concerned with maximum performance.

(Even for
>a laptop--given the power consumption of current displays--
>the CPU does not gain much from being extremely power
>efficient. ARM might be able to undercut Oracle's T-series
>servers [given a larger address space] in the near future,
>but no announced ARM is close to an Intel x86 server chip
>in performance.)

Indeed, you don't need to be extremely low power in a laptop, so that's why ARMs can be clocked at far higher speeds without worrying about power consumption.

In servers, performance per Watt, performance per m^3, or performance per dollar is often far more important than maximum single threaded performance.

Wilco