SeaMicro drops another Atom bomb on the server market | Hacker News Reader

SeaMicro drops another Atom bomb on the server market

venturebeat.com

58 pointsthankuz18 comments

First of all, Atom CPU is only "efficient" power-wise when it's sitting idle most of the time.

It's not efficient per task accomplished for power used.

Secondly, "dropping an atom bomb onto *" happens to reference one of the most horrible things the United States has done to any other country, regardless if you agree if it was unnecessary or not, so implying the reference is very distasteful.

This article would be much better if they used the hilariously-clever expression "atom bomb" even more times in the first paragraph.

Oh wait, no.

Clearly, the main benefit of using those microprocessors is that you get to say that you are "dropping Atom bombs."

So now HN is trying to be PRweb?

I have to say this was one of the worst/most confusing analogies I've read in a tech article:

"But those customers got tired of using the computing equivalent of a space shuttle to go shopping at the grocery store."

Uhh, doesn't the space shuttle run 386's?

I was going to be snarky and ask,"Do you use a 386 to go shopping at the grocery store?" but then realized I use an Apple A4/Cortex-A8 to go shopping at the grocery store. (Never mind whatever is embedded in my car.)

I think I'm going to steal that analogy anyway, because I love baffling metaphors. The nice thing would be using the Canadarm to put things in your cart, especially heavy items such as cat litter and gallons of milk.

I think I'll wait for someone like Jon Stokes to talk about SeaMicro before getting too excited about how 64-bit processors mean you can address 4Gb of RAM.

Yeah especially as the Atom N550 which has been out for a while has 64 bit support.

Any article that calls a cpu a brain has no place on hacker news.

Only seems to be 2Gb RAM support too, single channel, non ECC. And Atom is very slow...

I think it may have been Google that found that random RAM errors are quite likely over the course of a year, so you really want ECC memory if you're running a server and need uptime.

I found a (the?) paper on the topic: http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf

I was going to ask you for a citation but realized that would be lazy when I could google[1] it myself. I've had a couple arguments with a hosting provider all but ignoring RAM errors, with them responding,"I rebooted and watched it for a few days, and it seems to have cured the issue."

[1] Do you ever think "bing it" will catch on? I think I've discovered their growth constraint factor.

CPU speed doesn't concern me, as memory access/front-side bus tends to be the bottleneck in the applications I'm interested in. But non-EEC is a problem. Intel intentionally crippled Atom to not cannibalize its server business. I think a similar platform but with Arm Cortex-A15 would be even more interesting.

RAM is the new disk.

And the new disk is RAM. Whoa, you just baked my brain.

There are some interesting technical details (including an answer to the ECC question) in the Stanford EE380 video: http://ee380.stanford.edu/cgi-bin/videologger.php?target=100...

My brain had a hash collision with SEAForth and thought that it would be exceedingly unlikely to get much traction if they're expecting people to program enterprise software in Forth.

Not that it wouldn't be...fun?

Reading between the lines: does this thing have the same issues with I/O that virtual machines tend to have? (That doesn't make it useless, of course...)

Here's the original article on VentureBeat: http://venturebeat.com/2010/06/13/seamicro-drops-an-atom-bom...

A few points:

* Interestingly, they use the same image at the beginning of that article and this one. They also seemed to re-use a lot of the original article text in this one.

* Most people were quick to point out that Intel's Atom processors don't support ECC RAM, and at the scale we're talking about here, RAM errors statistically will be common.

* From the article:

  Full told, SeaMicro eliminates 90 percent of the
  components from a system board. SeaMicro calls this
  CPU/IO virtualization.

The HN Discussion: http://news.ycombinator.com/item?id=1429628

What seems to be the original announcement (HN submission 2 months before original VentureBeat article): http://gigaom.com/2010/01/06/seamicros-secret-server-changes...

HN Discussion: http://news.ycombinator.com/item?id=1070705

{edit} The gigaom article was 4 months before the original VentureBeat one (8 months ago vs 1 year ago). For some reason I was doing the date math in my head in base10 (i.e. 1 year = 10 months). Maybe it's time for bed.

Here's some discussion on SeaMicro's original cluster-in-a-box offering from the ever-excellent James Hamilton: http://perspectives.mvdirona.com/2010/06/14/SeaMicroReleases...

Lack of ECC support will be the killer here, I think - it would be great to see an ARM based alternative.

I still miss SiCortex. Now those were cool low-power supercomputers.

Reply on news.ycombinator.com