Design Con 2015
Breaking News
Slideshow: Intel Beats ARM Servers
9/4/2013

Image 1 of 16      Next >

HP upgraded Centerton cartridges in its Moonshot server to Avoton SoCs.
HP upgraded Centerton cartridges in its Moonshot server to Avoton SoCs.

Image 1 of 16      Next >

Return to Article

View Comments: Threaded | Newest First | Oldest First
rick merritt
User Rank
Author
Any other benchmarks out there
rick merritt   9/4/2013 1:01:44 PM
NO RATINGS
OK you dozen ARM server SoC vendors, if you have other numbers in the lab, let's hear them.

docdivakar
User Rank
CEO
Re: Any other benchmarks out there
docdivakar   9/5/2013 11:05:56 AM
NO RATINGS
@Rick: GOOD article! On the power requirements for Avoton, how does it compare to equivalent ARM processor/s at the same clock frequency?

Regarding "dense forest of passives", this is something that can be addressed by the motherboard designers. Obviously we don't have the info on routing congestion in the vicinity of Avoton chip. Otherwise many  of the passives could have been embedded which adds some cost penalty.

MP Divakar

rick merritt
User Rank
Author
Re: Any other benchmarks out there
rick merritt   9/5/2013 1:23:56 PM
NO RATINGS
@MP: I did not get as many specific power numbers as I wanted to see. Basically Intel just did perf/watt at the node level, no core to core and no deep dive across workloads yet.

resistion
User Rank
CEO
1000 mm2 Atom?
resistion   9/5/2013 1:16:14 AM
NO RATINGS
The footprint is way large. Remember when Atom was only 25 mm2; it's all grown up now, I guess.

Sheetal.Pandey
User Rank
Manager
Re: 1000 mm2 Atom?
Sheetal.Pandey   9/5/2013 7:19:26 AM
NO RATINGS
Intel and arm both are leaders in their own space. But competition comes when one try to enter in others space. I guess its good forthe whole electronics indusstry if both continue doing good business.

resistion
User Rank
CEO
Re: package sizes
resistion   9/5/2013 9:34:05 AM
NO RATINGS
I guess should be comparing package sizes, then it should be ~200 mm2 vs ~1000 mm2.

TanjB
User Rank
Rookie
Re: 1000 mm2 Atom?
TanjB   9/5/2013 12:24:23 PM
NO RATINGS
compared to the fingernail, the chip cap looks to be enclosing about 50 mm^2, not 1,000.  50 is jelly bean size for a CPU or SOC these days.

rick merritt
User Rank
Author
Re: 1000 mm2 Atom?
rick merritt   9/5/2013 1:22:27 PM
NO RATINGS
@TanjB: Right you are, the big shift from server CPUs to SoCs is something you can see in the very size of the packages

resistion
User Rank
CEO
Re: 100 mm2 Atom
resistion   9/5/2013 8:24:55 PM
NO RATINGS
The Avoton is a 34 X 28 mm2 package, and the photo seems to indicate the die is about 1/10 (~1/3 X 1/3) of that so I estimate now about 100 mm2 for Avoton, which would be about 4x the years ago die size of Atom Silverthorne. At that time the package was as small as 13 X 14 though it was often 22 X 22.

http://www.theregister.co.uk/2013/09/04/intel_avoton_rangeley_atom_c2000/

https://en.wikipedia.org/wiki/List_of_Intel_Atom_microprocessors

halherta
User Rank
Manager
ARM may still beat Intel when cost is considered
halherta   9/5/2013 1:39:35 PM
NO RATINGS
There's no doubt that Intel can beat ARM when using performance per Watt as a metric. However traditionally ARM beats Intel on cost. So in devices where best performance is not the goal (smartphones, tablets e.t.c) ARM won and will continue to win because of its lower cost.

It will be interesting to see who wins on the server front. Performance is obviously more crucial in servers....But the cost metric is also just as important.

 

LarryM99
User Rank
CEO
Re: ARM may still beat Intel when cost is considered
LarryM99   9/5/2013 3:41:34 PM
NO RATINGS
Actually, performance per watt is probably more important to a degree to the server market than cost. Keep in mind that power is an ongoing expense while you only pay once for the chip - at least until Intel and the other chip companies can figure out a way to lease them...

Also, the CPU chip cost is only one component of the cost for a server, both in terms of dollars and watts. Do we know how different a server board BOM cost is between ARM and x86? Likewise, is the only significant power usage difference between those server boards the CPU itself?

rick merritt
User Rank
Author
Re: ARM may still beat Intel when cost is considered
rick merritt   9/5/2013 4:29:56 PM
NO RATINGS
There are a lot of shoes yet to fall here but Marc Andreeessen today was very bullish on the ARM wave in a talk today

See  http://www.eetimes.com/document.asp?doc_id=1319417&page_number=2

 

Wilco1
User Rank
CEO
Re: ARM may still beat Intel when cost is considered
Wilco1   9/5/2013 4:47:01 PM
NO RATINGS
I seriously doubt Intel can beat ARM on perf/W. AnandTech showed several 6W Calxeda nodes beating low power Xeon CPUs on performance as well as perf/Watt. Next-gen 28nm Calxeda nodes based on A15 will be available in a few weeks, and should improve performance as well as perf/W further.

So Intel can only win by comparing not-yet-released 22nm chips with 40nm ones which have been shipping for a year...

And the real like for like comparison will be next year with 64-bit Cortex-A57 on 20nm. A57 is significantly faster than A15 and reaches far higher frequencies as well.

rick merritt
User Rank
Author
Re: ARM may still beat Intel when cost is considered
rick merritt   9/5/2013 6:21:41 PM
NO RATINGS
@Wilco1: Admitedly the Intel comparisons are not ideal (64 vs 32 bit CPUs eg). However, note that Avoton is shipping now.

Gondalf
User Rank
Freelancer
Re: ARM may still beat Intel when cost is considered
Gondalf   9/9/2013 4:18:12 PM
NO RATINGS
It's pretty clear that you live in your own dream.

Any attempt of Calxeda to server market is just now truncated, Avoton is clearly a winner in the perf/watt parameter and this will be good even for 20nm planar Calxeda socs. Not a chance, once Calxeda will shift to A57 the power consumption will grow like a hell.

Moreover there is a 16 core Intel Soc in Q3/2014, bye bye competitors.

Wilco1
User Rank
CEO
Re: ARM may still beat Intel when cost is considered
Wilco1   9/10/2013 6:06:35 AM
NO RATINGS
Avoton is certainly an improvement on the ridiculously slow and inefficient Centerton, but it is not as efficient as ARM chips. The 40nm Calxeda server node uses 6-7W in total including DRAM and interconnect, while 4C Avoton starts at 13W - that's only for the CPU, not the DRAM and interconnects... 

Calxeda will move to 28nm with their next generation in a few weeks and 20nm next year. Then there is AMCC with X-Gene, AMD with Hierofalcon, NVidia with Denver - all having 8 and 16-core 64-bit ARM server chips to be released in 2014 and 2015. It looks like Avoton will end up in the exact same situation as Atom is today.

Radio
LATEST ARCHIVED BROADCAST
EE Times Senior Technical Editor Martin Rowe will interview EMC engineer Kenneth Wyatt.
Top Comments of the Week
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Flash Poll