Pretty cool, if the SeaMicro storage fabric could be added it would be even more fun, pretty awesome transactional machine in a rack. I would have loved to build a storage appliance out of a system like this when I was at NetApp.
Its a killer feature as long as the drivers are ARM-only.
Considering that Calxeda is the only one making the fabric, and Calxeda is the only one using these ARM chips in general, I'd say its a killer feature (specifically: a killer feature for Calxeda-ARM motherboards, not for any other manufacturer of ARM chips). And since AMD is partnering with Calxeda here, they get the benefit.
AMD is also the only big-player that has big-ARM chips out in any volume right now anyway. I know there are a few from China but nothing really standardized.
-------------
In any case, Xeon-D and Atom Centerton are kicking some major ass in power/performance benchmarks. I find it unlikely that the AMD chip will keep up with Intel. Still, being the only major ARM-server manufacturer is a good position to be in.
IIRC, these AMD chips are Seattle, which is 28nm (two nodes behind Intel). AMD is still hopelessly behind Intel. AMD is hoping that Zen-ARM chips will catch up in 2017.
So basically... I bet this is mostly for R&D and experimental deployments. Because if you're really deploying a multi-million dollar data-center full of computers, you want the best. And the best, for better or worse right now, is the Xeon-D.
If you want to start experimenting with ARM server development, it looks like these AMD Seattle chips are the best ones on the market, with a very intriguing implementation with Calxeda's motherboards / switching fabric as well. If Calxeda / AMD undercuts Xeon-D with cheaper chips, they probably can win with I/O heavy workloads (video hosting, cloud storage, etc. etc.)
Rumor is that these Seattle chips have 14-SATA ports (!!!!) and on-chip Gigabit-Ethernet, so AMD is definitely targeting those I/O heavy workloads.
Calxeda is no longer in existence.. SLS bought their assets to make this new product, but SemiAccurate (the linked article) seems to incorrectly suggest otherwise. SLS only mentions Calxeda in the context of them acquiring their technology, but none of the same (major) people are involved.
7 comments
[ 3.3 ms ] story [ 25.2 ms ] threadConsidering that Calxeda is the only one making the fabric, and Calxeda is the only one using these ARM chips in general, I'd say its a killer feature (specifically: a killer feature for Calxeda-ARM motherboards, not for any other manufacturer of ARM chips). And since AMD is partnering with Calxeda here, they get the benefit.
AMD is also the only big-player that has big-ARM chips out in any volume right now anyway. I know there are a few from China but nothing really standardized.
-------------
In any case, Xeon-D and Atom Centerton are kicking some major ass in power/performance benchmarks. I find it unlikely that the AMD chip will keep up with Intel. Still, being the only major ARM-server manufacturer is a good position to be in.
IIRC, these AMD chips are Seattle, which is 28nm (two nodes behind Intel). AMD is still hopelessly behind Intel. AMD is hoping that Zen-ARM chips will catch up in 2017.
So basically... I bet this is mostly for R&D and experimental deployments. Because if you're really deploying a multi-million dollar data-center full of computers, you want the best. And the best, for better or worse right now, is the Xeon-D.
If you want to start experimenting with ARM server development, it looks like these AMD Seattle chips are the best ones on the market, with a very intriguing implementation with Calxeda's motherboards / switching fabric as well. If Calxeda / AMD undercuts Xeon-D with cheaper chips, they probably can win with I/O heavy workloads (video hosting, cloud storage, etc. etc.)
Rumor is that these Seattle chips have 14-SATA ports (!!!!) and on-chip Gigabit-Ethernet, so AMD is definitely targeting those I/O heavy workloads.