[Papa.Smurf]

Source

Quote:

The Molecule computer can handle 20,000 threads of execution, about 40 times more than a single rack x86 cluster system and was designed around an Intel Atom N330 chip, says SGI. The computer would offer the computing power and memory bandwidth of more than 750 high-end PCs, yet it would consume less than half the power and less than 1.4 percent of the physical space.

 

[Bunnywinkles]

i want. 6 minutes to render 45min of film is too long for me! this could do it in like 6 seconds xD

 

[Licht]

Do this with 10,000 4870s or GTX280s. The power would be incomparable. Although limited. Mainly i'm getting at the potential of the Atom isn't what i would desire for a system like that. Needs that non-x86 spark of life.

 

[dieanotherday]

Folding?

 

[metala]

Imagine if dust blocks some of the air flow...

Quote:

Originally Posted by Licht Needs that non-x86 spark of life.

Yep, OpenSPARC is a good one

 

[21276]

means nothing without an unlocked multiplier.

Lol, awesome though, nice find

 

[Licht]

Quote:

Originally Posted by metala Yep, OpenSPARC is a good one

Meh, you know what i mean, standard x86 CPU aren't suited for this line of work. Well the Opterons are pretty nice for it... but i think super computing lies in the hands of the (G)GPUs of the future. Or custom made CPU/GPU/Etc. Massively parallel applications i hope to see standard within my time.

 

[GDP]

whatever, not impressed. Pails in comparison to wetare brain implants =P I cant wait for those. WE ARE THE BORG.

 

[BizzareRide]

But can it...

...RUN CRYSIS!(You thought I wasn't gonna say it huh, come on, admit it)

 

[Takendown2]

Can it be OCed? um.... and y not do this with 10000 GTX280 and create a MONSTER

 

[MasterFire]

Quote:

Originally Posted by BizzareRide But can it... ...RUN CRYSIS!(You thought I wasn't gonna say it huh, come on, admit it)

*reads too fast to think*

 

[BizzareRide]

Quote:

Originally Posted by MasterFire *reads too fast to think*

CRAP! I knew I should have spaced it out more!

 

[logo199]

The point of this wasn't 10000 of the best or fastest cores (g200 cores and stuff) it was placing 10,000 cores in a cluster and having it draw little power for the computing power it offered.

 

[r34p3rex]

Quote:

Originally Posted by Licht Do this with 10,000 4870s or GTX280s. The power would be incomparable. Although limited. Mainly i'm getting at the potential of the Atom isn't what i would desire for a system like that. Needs that non-x86 spark of life.

Quote:

Originally Posted by Takendown2 Can it be OCed? um.... and y not do this with 10000 GTX280 and create a MONSTER

You'd need an onsite nuclear reactor to power that

oh, and uhh permanent LN2 cooling

 

[seodeveloping]

Quote:

Originally Posted by r34p3rex You'd need an onsite nuclear reactor to power that oh, and uhh permanent LN2 cooling

With the right funding...

 

[Deathsnapper]

Quote:

Originally Posted by r34p3rex You'd need an onsite nuclear reactor to power that oh, and uhh permanent LN2 cooling

Exactly. What seems like hte biggest draw for this to me isn't the it's computational power. It's the fact that it can achieve it at "less than half the power". The tiny size is pretty cool too, but the power requirements are ridiculous.

A few GTX280s, or 4870s or whatever and you're already using more power than it.

I'm sure I'm not the only one who thought this would eventually happen when reading some of the articles just after the atom was first announced. Although it's got to be extremely expensive. So awesome, still.

 

[tylor3600]

If ATI and Nvidia would just shrink the cores more, the power/performance ratio would be alot better. So alot more people would adopt them because they do perform better in those kinds of environments. And yes its costs loads of money to shink a fab, but eventually they'll do it and optimize everything to get the right power/performance ratio.

 

[r34p3rex]

Quote:

Originally Posted by tylor3600 If ATI and Nvidia would just shrink the cores more, the power/performance ratio would be alot better. So alot more people would adopt them because they do perform better in those kinds of environments. And yes its costs loads of money to shink a fab, but eventually they'll do it and optimize everything to get the right power/performance ratio.

they're going to 40nm soon so let's see where that gets them

 

[Licht]

Quote:

Originally Posted by r34p3rex You'd need an onsite nuclear reactor to power that oh, and uhh permanent LN2 cooling

My point was that the performance isn't what a 10,000 core computer should be.

The normal run of the mill x86 chip isn't suited for it, especially not a Core/Core2/Netburst/<anything other then K8,K10,K10.5, or Nahalem's architecture... Core 3 wasn't it?>.

 

[Lelouch]

Quote:

Originally Posted by Licht Do this with 10,000 4870s or GTX280s. The power would be incomparable. Although limited. Mainly i'm getting at the potential of the Atom isn't what i would desire for a system like that. Needs that non-x86 spark of life.

Quote:

Originally Posted by Licht Meh, you know what i mean, standard x86 CPU aren't suited for this line of work. Well the Opterons are pretty nice for it... but i think super computing lies in the hands of the (G)GPUs of the future. Or custom made CPU/GPU/Etc. Massively parallel applications i hope to see standard within my time.

Power consumption, and heat.

Quote:

Originally Posted by Takendown2 Can it be OCed? um.... and y not do this with 10000 GTX280 and create a MONSTER

Power consumption, and heat.

The problem with these is that the atom draws VERY little, and runs very cool, along with they cost less, even if its more "economical" to buy the 280's, the power consumption between the two would make a difference very shortly after purchasing.