@Iggy
(A break-out from the X1000 dnetc benchmark thread)
@ Iggy
A Cortex-A8 (Efika MX) beats a Sam440.
A Cortex-A9 beats a G4.
You like long winded posts, but you never offer any proof for your statements.
I'd really like to see a benchmark on the second one since it seems implausible.
"A Cortex-A8 (Efika MX) beats a Sam440"PPC440EP: 2.0 DMIPS/MHz, meaning 1,334 DMIPS @ 667MHz of a Sam440ep
Cortex-A8: 2.0 DMIPS/MHz, meaning 1,600 DMIPS @ 800MHz of an Efika MX
NBench pictures the Efika MX as generally faster, even after upscaling the Sam results to 800MHz. In addition to this, the CPU used in the Efika MX has several HW accellerators that will offload and speed up many multi-media tasks from the CPU, like audio/video encoding/decoding, OpenVG, it has a SIMD (like Altivec, but not as powerful) called Neon, etc. None of this is present in the PPC440EP, which doesn't even have a L2 cache (the Efika MX CPU has 256KB). For example, the Efika MX can decode and display 720p HD video streams (it can actually decode 1080p streams, but not display it at 1080p resolution, due to some bottle necks that was removed in the i.MX53 version). The Efika is the winner.
Oh, and for the sake of comparison:
Efika MX: $129, giving you a complete, fully working system for most uses (but yes, rather limited and not expandable)
Sam440: ~$600, giving you a *motherboard only*, system components of choice *on top* of this, plus the mandatory OS4, all in all "north of" $1,000.
VAT not included.
"A Cortex-A9 beats a G4"MPC7447A: 2.3 DMIPS/MHz:
This should mean about 3,266 DMIPS @ 1,420 MHz (a *fast* G4, used for example in a Common Mac Mini)
Cortex-A9: 2.5 DMIPS/MHz (*per core*, can have 1-4 cores):
A *generic* Cortex-A9 from ARM (used in Apple A5 and Tegra 2 for instance) delivers 2,500 DMIPS per core @ 1,000 MHz, meaning 5,000 DMIPS per CPU in dual core configuration.
Then there is the upcoming (summer 2012?) Qualcomm Krait, that isn't simply just using the generic core, but optimized and tuned to improve the per-core performance to 3.3 DMIPS/MHz (built using 28nm process), resulting in a staggering 9,900 DMIPS @ 1,500 MHz.
The Tegra 3 is a *quad-core* Cortex-A9, running up to 1.300 MHz (times 2.5 DMIPS/MHz, times 4 cores). It's "
up to 3 times faster than Tegra 2 (measured with GLBenchmark 2.0 Egypt)", still being Cortex-A9, still being built in a 40nm process. Actually, the Tegra 3 is a "five core" CPU:
"The Tegra 3 is the world's first variable symmetric multiprocessing processor (vSMP). Variable SMP is the brain that makes the 4 main + 1 companion core setup tick. Basically, it smartly distributes the workload so the platform uses four cores humming at up to 1.3GHz when needed (think games), but switches to two for lower loads (think browsing with Flash), and goes down to one for casual activities (browsing, no Flash). Finally, it can also use only the companion core running from 0 to 500MHz for active stand-by, video and music." (Link:
An Interesting Article on the Tegra 3)
The scaled
NBench numbers (measuring *one* of the cores on an "old" OMAP4 based 1GHz PandaBoard) suggests that the Cortex-A9 is about on par with the G4, but again, much has happened to the Cortex-A9 CPU's since that one was made, and since most Cortex-A9 based CPU's come in dual core configuration (or quad), I actually feel my statement that Cortex-A9 beats a G4 is water proof. In addition to this, the Cortex-A9 processors (like most other ARM processors) also come with various HW accelleartors on the chip, freeing the core CPU from the most common heavy multi-media tasks, vector graphics/GUI/Flash, hardware Java/JIT acelleration, etc, things that the G4 simply don't have.
And while we are at it: Some speculations about Cortex-A15 and G5While not being a 64-bit CPU (although it will at least be able to address up to 1TB of memory thanks to the 40-bit Large Physical Address Extensions (LPAE) addressing), the Cortex-A15 will probably be giving the old G5 a match, with its
3.5 DMIPS/MHz in configurations suitable both for mobile, high-end home entertainment, or low-power servers, up to quad core running at 2.5GHz (or for servers for example, octo-core or larger configurations through multiple coherent SMP processor clusters through "AMBA 4").
For example the announced dual core:
"ST-Ericsson Nova A9600 (Dual Core @ 2.5 GHz Over 20K DMIPS)"Iggy, this is a quote from a post
you made here on amiga.org back in 2010:
e300: 1.9
PPC440: 2.0
PPC460: 2.0
Titan: 2.0 (presumably PPC450 based (http://www.morphzone.org/modules/newbb_plus/viewtopic.php?topic_id=7186&forum=11&start=20#74317))
PA6T: 2.2
PPC750: 2.3
e600: 2.3
PPC470: 2.3 or 2.5 (varying with information source)
e500mc: 2.5
PPC970: 2.9
e5500: 3.0
As a reference to
year 2012 ARM (some in quad core with clock frequencies up to 2.0-2.5GHz):
ARM Cortex-A7: 1.9
ARM Cortex-A8: 2.0
Qualcomm Scorpion: 2.1
Marvell Sheeva PJ4: 2.4
ARM Cortex-A9: 2.5
Marvell Sheeva PJ4B: 2.6
Qualcomm Krait: 3.3
ARM Cortex-A15: 3.5
This is current or immidiate future (like this year), not even mentioning the not-too-far-away 64-bit "x86-killers" from nVidia and others, based on ARMv8.
