This stuff is awesome for mass computation and clustering of intense CPU applications, but with only 4GB of RAM per blade there seems to be some applications limitations.
Regardless, SeaMicro is a great pickup for AMD. They've been getting their asses absolutely handed to them in the server space. Hopefully this gets them competitive again.
Actually, CPU intensive applications are a bad fit for seamicro products. Basically, the cpus in their blades suck, and the entire architecture is designed to favor io-heavy, jumpy (so not fit for GPU), but easily partitionable loads. Basically, web servers.
with AMD, they will be able to better integrate the cpu cores with their IO systems -- Intel was unwilling to let them fiddle around with Atom IP. I expect that there will be some kind of custom-fit Bobcat-derived cpu products for the web server market. Very low price tag, but also very low cost to build.
This is a very smart acquisition. From what I saw SeaMicro has a great innovation in the form of their 'virtualize IO.' It is a great idea to share IO of machines in a cluster. They can make big headway by integrating tuned CPU's into those clusters. This is going after the entire stack of an IO bound application.
The custom ASIC work is interesting, but I tend to think Caldexa's approach of just putting 4x10Gbit ethernet links on each CPU instead of building specialized virtualization HW would be a more worthwhile advancement. That said I think it's awesome and there's a huge opportunity for AMD to do great things with some cool super small super dense computing here.
6 comments
[ 3.2 ms ] story [ 33.4 ms ] threadRegardless, SeaMicro is a great pickup for AMD. They've been getting their asses absolutely handed to them in the server space. Hopefully this gets them competitive again.
with AMD, they will be able to better integrate the cpu cores with their IO systems -- Intel was unwilling to let them fiddle around with Atom IP. I expect that there will be some kind of custom-fit Bobcat-derived cpu products for the web server market. Very low price tag, but also very low cost to build.
exit the x86 market, enter the vertically integrated server hardware space, this is insane.
why aren`t they spending that over a quarter of a billion dollars on building better cpus/gpus. not trying to become some sort of ibm.