Calxeda ECX-1000 Benchmarks vs. Intel Atom, TI OMAP4
Phoronix: Calxeda ECX-1000 Benchmarks vs. Intel Atom, TI OMAP4
Last week I began delivering benchmarks of the low-power yet massively scalable Calxeda EnergyCore ECX-1000 ARM Server and followed the initial tests with some ARM compiler benchmarks and other benchmarks from this 5-Watt Linux Server. In this article is what many Phoronix readers have been waiting for: comparing Calxeda's quad-core Cortex-A9 ARMv7 performance against a dual-core Texas Instruments OMAP4460 PandaBoard ES and then an Intel Atom processor.
In short, a server grade ARM processor is getting whipped upside down by a consumer grade x86 Atom processor.
The Calxeda chips exist to be WEB SERVERS
Some web benchmarks, maybe a database benchmark or two, and OpenSSL benchmarks would have been much much more informative in this review. These Calxeda boxes are really designed to be cheap web servers, so all the scientific computing stuff is pretty useless. I'm not overly impressed at what these quad-core CPUs are doing compared to a 2.5 year old consumer grade Atom either. Centerton will be more powerful and much more power efficient, so these ARM chips are not going to take over the world any time soon.
Good results, but can you test the power load?
Thanks for testing these processors. A very useful set of results.
Can you please buy a power meter and measure the amount of power consumed at the wall socket?
If that Calxeda box has a ton of cores in it, can you pull out the cards until there is a single CPU to make it fair against the atom (i.e. one socket vs 1 socket)?
My take on this is that the Atom still rules the day.
1) Lets assume that the atom draws 50 watts at the socket (like my 525 setup here does), and the Calxeda setup draws like 5 (which it wont). That savings of 45 watts turns out to be about 40 UK pounds per year on an expensive tarrif. I wouldn't care about that because:
2) I don't have any pain working with different toolchains. The toolset I use on my desktop works on the Atom. Just switching CC=arm-linux-gcc in my makefiles is headache.
3) I know that the core-i7 is more efficient than the atom, and I'll just buy those instead.
ARM64 will be the great deal
I thought it was ARM64 to be tested, there are some hardware prototypes out there.
But not all is computational power, energy budget, and good scalation and price are needed too.
How much improves the server with 2/4/8 etc processors or even better the rack with full boxes and How much does cost the hardware, and a new category how much space does it occupy.
Future minisized supercomputers ARM64 for SOHO and college departments from Caixeda and others like x-gene are saying that the space needed will be 90% less and the power budget will be more than half cut.
Also some X86 hardware emulation and mixed architecture promises are on the steak
Whipped? The 1.4 Ghz A9 quad core performed equal to or better on most tests. Performance per watt was unable to be measured but knowing that the TDP of the Atom is 13w and the total draw for the 1.4 Ghz ARM is 6w you can pretty much figure out the winner here. This is also just Cortex A9 cores on the ARM front as well which is a fair test against the older Atom seeing that they were both introduced to the market in mid 2010. Once these get updated to Cortex A15 cores running at 2 Ghz+ it will truly be a site to see. Looks like ARM is making some solid inroads and hopefully they continue this trend with ARM64. The future looks very bright!
Originally Posted by Sonadow
If you would have read the entire article or most other Phoronix articles, you would know that I certainly have USB power meters -- and that the Phoronix Test Suite can poll them automatically during testing. But why no power results were provided is mentioned in this article due to limitations. Linked to from that article are some SoC power readings. Linked to from that article was also the original article where it mentions I had to do this benchmarking at the Calxeda office so I don't have any Calxeda servers in my physical possession so I simply can't re-run more power tests or swap out hardware, etc...-
Originally Posted by bms20
What the hell do computational benchmarks tell about a platform targeted at web servers!? Correct me if I'm missing the point, but IMHO not a freakin' thing.
Yet again, content mill smell. Dude, listen to the constant complaint of your readers, will you?
Oh really? Why are financial trading firms and other companies interested in these Calxeda boxes then...? It's not for web-serving... Of the companies I heard mentioned by Calxeda folks at their offices, probably less than half of them were web-serving related...
Originally Posted by pszilard
Google Chromebook is exynos 5250, A15@1.7Ghz, sb has installed Ubuntu on it, so it's very funny to run your test software on A15 compared to atom D525.
Btw, D525 performance should be roughly equal to z2760 with tdp=6w.