Post by
MinisterofDOOM »
https://forums.nicoclub.com/ministerofdoom-u16506.html
Tue Jun 15, 2010 11:55 pm
An interesting idea. Unfortunately it misses some critical points. Server-specific chips exist because there is a NEED for them. Efficiency of processing is just one of many, many facets of a processor's effectiveness. If everyone could get by running basic low-end consumer-market chips, they'd be doing it already. The reason they have not done this is that servers have different technology requirements than consumers. Look at Xeon vs Core for example. Xeons have large caches, built-in error-correction, native support for greater amounts of memory, etc.
Atom does not. It doesn't matter how many of them you glue together, or what kind of high-tech data connect you use to interlink them; they still do not have those features. They still are not suited to server-oriented processing.
The article is also moronic in its reference to the Atom server's size. It lauds the server (ONE server) for being a quarter the size of an entire RACK. It's HUGE. You could fit MULTIPLE traditional hotswappable rack servers in the same space.
The company certainly has a point: reducing power costs for server processing would be fantastic.
But using the weakest x86 chip on the market is NOT the way to do it. What we need is a purpose-built low-power, server-oriented chip built to meet similar power specifications, but with the needs of server-oriented computing still being met.
Atom is just very ill-suited to this kind of work. Making up for its failings by daisy-chaining it isn't exactly an ideal solution in my eyes. If SeaMicro were REALLY geniuses, they'd be designing their OWN chip for this purpose.
This is sort of like using 200 Priuses to equal a Peterbilt.