Slashdot Mirror
PandaBoard ES Benchmarked
An anonymous reader writes "Phoronix has benchmarked the Texas Instruments PandaBoard ES and compared its performance against Intel Atom N270, Atom Z530, Pentium M, and Core Duo T2400 processors. The OMAP4660 dual-core 1.2GHz ARM Cortex-A9 development board generally loses out to Intel's older competition, but does manage to win in ray-tracing and other tests, and is advantageous on a per-Watt basis."
Still not as fast as it's power hogging competition, but pretty decent.
I do a lot of work with Gumstix Overo's, and at home on the original Pandaboard. I am constantly amazed at how powerful those little systems are.
To be sure my Quad Core Xeon that I cross compile on will eat them for lunch... but at 5x the cost and 50x the power consumption.
The benchmarks provided by Phoronix focus on computational power, which is a relevant criteria. Yet, ARM-based systems aren't targeted at the high performance computing field. In their domain of application, criteria such as power usage and price tends to be much more relevant than how fast it compresses files, encodes MP3s or runs synthetic benchmarks. In fact, if it is fast enough to play media then it's fast enough to do anything at all.
So, how about comparing them where they need to be compared: power output and price?
Slashdot, fix your code or at least hire someone who is competent at it to do it for you.
A lot of the tests are irrelevant when done between Atom and OMAP4.
The OMAP4 has internal accelerators for voice and video coding - stuff like VP8 and x264 can be done a lot faster on the OMAP4 if you ditch the software itself and use the OMAP4's accelerators instead.
We've been able to use OMAP4660 for encoding 720p at 30fps into H.264 while using only ~20% of the CPU. Try doing that in software on a Core i7 and see where it gets you.
When doing benchmarking on ARM Cortex chipsets, there needs to be more care taken in how you treat the accelerators of the SoC in question and if you use them at all. By the results of these benchmarking trial it seems that the accelerators were ignored altogether - not something I'd do if I were to actually use the OMAP4 for any real development.
after so many years, the constant Phoronix crap on the front page is like parody
As someone with experience doing embedded development on ARM I can tell you I found the OMAP architecture to be awful. I'll admit the only time I ever used it was on a demo board (the Beagle) vs a board with essentially identical specs from FreeScale, Renesas and a few others. TI was awful with support, their documents were awful, the hardware was flaky (overheating!?) and the sample sources and module sources they provided were absolute crap. On top of that when we did get the boards running and started comparing them the OMAP board was slow as tar on anything that involved a lot of memory operations in a small timeframe. Apparently the GLES subsystem was fantastic or something but after a few attempts we couldn't get the modules built correctly against the kernel we were using and just gave up. In the end we went with the FreeScale (not my choice) which was easily superior to the TI OMAP garbage.
Sorry TI, I'm not even touching this one.
Where are the drivers and SDK for this?
I am very small, utmostly microscopic.
A more interesting benchmark for me at the moment is performance per $, which is where Raspberry Pi is going to have a big impact soon I think.
"I bless every day that I continue to live, for every day is pure profit."
You can find an ARM vs x86 power consumption at:
http://www.brightsideofnews.com/news/2011/5/19/the-coming-war-arm-versus-x86.aspx?pageid=6
"The chart below contrasts power consumption between the Intel Atom N450 and the ARM Cortex-A8 while running miniBench. The power curves were generated from system power usage adjusted downwards so that idle system power was discarded. For the Atom, idle power was 13.7W with the Gateway netbook’s integrated panel disabled while the idle power for the Pegatron system (ARM Cortex-A8) was only 5.4W."
Reading the article I'm left with a number of what I feel are important questions.
Firstly, which distro was this run on? I don't believe the 11.12 linaro release was (www.linaro.org) used which would be the most optimized for arm choice that currently exists. I looked at the phoronix source code and it seems to have no concept of linaro at all tho it does know ubuntu and debian of which linaro is a varient.
GFX Hardware acceleration for the Pandaboard ES is a bit of a work in progress, it's hard to know if the lastest work was included.
The article does not state which version of the compilers is in use. gcc 4.4 ? 4.6? flags? It can make a bit of a difference.
There is no statement if the intel comparisons are running the same level of software or not? This is another important data point.
Others have mentioned but I will too, it does appear that software decode/encoders were uses instead of the hardware versions. This would substantially change the numbers.
All that said, for someone obtaining a board, pulling software from anywhere without direction as to how to get the best performing code for arm, this would seem to be a good sample. If the goal of the article was to see the Pandaboard ES running with the best performance as compared to intel atom running at it's best, this article misses the mark completely.
I will ENJOY seeing this absolutely DESTROYED, BEAT INTO THE ASPHALT in terms of price to performance by the Raspberry Pi very soon. Days of $100-200 ARM boards are coming to an end, now dear Pandawhatever please set the sane price of $50 for your board, or die out of existence.
I already feel badly for the people who will make the mistake of buy a computer in the near future with an ARM processor in it. The latest ARM processors performance is comparable to a 1996 Pentium class processor. In embedded device which require low wattage and high performance the natural choice is PowerPC. And sense next year being the year of GNU/Linux, and Lenovo's rights to use the THINKPAD name will expire. So if IBM can make the PowerEN a 18-core 64-bit processor all under 60-watts using a 45nm process, then surely IBM could make a dual-core 64-bit PowerPC processor using a 32nm process which would be under 5-watts for the NEW ThinkPads.
I dont.
They will have a mini computer that will go 12 hours on a charge and do everything you need for web surfing except for crap designed flash websites.
I want one, I want a laptop that will go all day long and will recharge with a solar panel.
Hell I would buy a laptop that had a e paper display and could not play video. A backpacking computer / tablet that is durable as hell, goes a week on a charge, and cna let me record my thoughts or inspect maps at camp at the end of the day would be a god send. the craptastic kindle cant do any of those, except for the last a week. I carry one for a book to read at night, just wish it had the resolution to display topo maps decently and a flip out keyboard for journal entries.
Do not look at laser with remaining good eye.
And What OS & Applications would these PowerPC based Thinkpads run? Certainly not Windows 8 This sounds like a solution looking for a problem....
Do not read this
Mac os 9 and windows NT 4.0 (nah there are plenty of nix choices with power pc support)
PPC?
1997 called and wants it's next big thing back.
I am very small, utmostly microscopic.
1996 Pentiums didnt run at 1ghz. And theyre stacking this up against Core2s, which are hardly Pentium class.
In the year of GNU/Linux you wouldn't wont to run Windows or Windows 8. Yet, Microsoft has licensed the Power architecture from IBM for the X-box360, and they have been coding PowerPC builds of windows for over 5 years now. So if your one of those who still uses windows, Microsoft could easily build "windows 8" for the NEW Thinkpads which would also be easer than building for ARM.
The panasonic toughpad with ics android http://www.panasonic.com/business/toughpad/us/best-android-rugged-tablet-overview.asp
Troll much?
These SoCs can devote a dozen or so mm^2 for the CPU cores, yet they still achieve performance comparable to a 1GHz Atom but at far far lower power consumption levels. In addition the rest of the SoC contains application specific accelerators (graphics, video, security are the most common) for the difficult tasks. The Phoronix benchmarks sadly didn't test the SoC as a whole, just the CPU cores. It's also up in the air whether ARM Neon was actually utilised in all the tests.
Performance != Clock Speed.
Ok, let me add one more requirement.
Must not cost as much as a car.
Panasonic Toughpad $2,995.00
If I cant use it as a shovel in gravel or as a oar in the water taped to a stick and it will take ZERO damage.... It's not worth it.
and the battery life of the toughpad is not at 12 hours with fast recharge like E-paper devices.
Do not look at laser with remaining good eye.
You do know that all 3 current gen game consoles are driven by PPC CPUs, right?
Yeah sort of makes the ipad look frugal LOL.
No, they're stacking this up against Pentium Ms and Cores. The Core architecture was a short-lived 32-bit mobile architecture in 2006, derived from the older P6 family. They are the final generation of actual Pentium (discounting the rebranding of new low end chips). The Core2 architecture was a completely new design, based off the same paradigms as the P6 family. Both mainstream systems detailed in this review were over five years old.
Performance is, however, strongly affected by clock speed. Cortex-A8 isn't too far off microarchitecturally from the Pentium (two-issue, in-order), so they're probably not too far off in performance-per-cycle. Current-gen A9 cores are (I would say) around a fast Pentium II or a low-clocked Pentium III.
I think the limitations of boards like the Panda force programmers and designers to be more efficient instead of brute-forcing like the competition.
A lot of the tests that were done would have benefited from having Hard Float. Ubuntu ARM port does not have Hard Float. They should have used the Debian HardFloat port to get more accurate performance metrics of what the hardware can do.
I'm not arguing over semantics or fractions of percentages - Hard Float would have given an easy 20% increase in performance for some tests! For example here's an engineer from Genesi showing off the Debain Hard-Float work a few months back... 300% increase in some places?
Would you benchmark cars giving all the others high-octane fuel except one?
Please let it Soft-float fucking die already. It's horrible.
that's totally silly, a non-x86 cpu has no benefit for running windows. You can't run x86 software on it, and that's why anyone uses windows in the first place. Windows without backwards compatibility will not sell.
I would like to see a 12 hour benchmark that reported normalized results
in terms of Kg of battery. All these processors are in the ball park for
operations per second but many can NOT do it all day long.
Twelve and 24 hour results are needed to be sure. But a smart phone with a three hour
battery life is not a smart design. Simply from the safety point of view this is important.
Truth is stranger than fiction, but it is because Fiction is obliged to stick to possibilities; Truth isn't. Mark Twain.
That argument it so old, yet it keep people from migrating in the 80-90's from x84 to the superior architectures of Mipps, Motorola's 68000, Digital's Alpha, IBM's PowerPC, or MOS 6502. Microsoft NT was initially designed to operate on a broad set of architectures, so your implying that windows NT for Itanium doesn't support most of Microsoft tools. Even Apple with OSX is minimizing cpu architecture dependencies as much a possible. Furthermore, we no longer care about windows backwards compatibility just us windows in a virtual machine until you find or port the application to Unix. We use Unix and Unix-like systems. Have you seen the road maps for Intel's future processors at 22nm?( http://www.tested.com/news/intels-ivy-bridge-mobile-processor-roadmap-leaked/3255/ ) The TDP are all above 16 watts that abysmal for a 22nm process, and a two-core design.
The compilers, libs and more for x86 and friends
are so much more mature this is hardly a fair game.
The best high school players paired against the
winner of the superbowl....
And it is not just the processor the comparison
seemed to depend a lot on graphics drivers that are
just now using graphics hardware.
Truth is stranger than fiction, but it is because Fiction is obliged to stick to possibilities; Truth isn't. Mark Twain.
Make that two orders, and it better have a a metal chassis.
I know tobacco is bad for you, so I smoke weed with crack.
Ugh... It's not MY argument, it's the reality of history. A software translation layer (VM) will be less efficient than microcode, which is what intel processors use anyway to decode x86 into a RISC ISA. You really don't know what you're talking about here.