Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Aklyon]

The result is a system six times more powerful than existing servers that requires eighty times less energy.

I call shenanigans. That sounds like a significant power drop for something that is increasing in power.

[Reelya]

Not really, "existing server" is quite a limited concept, and the most efficient Intel experimental processor a few years back could do a couple of teraflops on less than 60 watts.

These experimental chips aren't constrained by things like backwards compatibility with crappy instruction sets, so you can pack 100's of ultra-efficient micro-cores in there with their own working ram for less than the size and energy needs of a normal Intel 64-bit CPU.

[Aklyon]

These experimental chips aren't constrained by things like backwards compatibility with crappy instruction sets.

That makes a lot of sense, actually. ok.

[MaximumZero]

The part that I'm skeptical about is "prototype by 2015, mass production by 2018." They're going to have this up and going next year?

[Reelya]

Oh yeah, I'm heavily skeptical of seeing this ship in any practical product for a very long time. We still haven't seen any devices for sale with those Intel 2-teraflop 60 watt chips. And that was actually demonstrated working way back in 2007:

http://en.wikipedia.org/wiki/Teraflops_Research_Chip

[Descan]

From what I understood of it, the only "really new" thing was the optical-instead-of-copper "wiring"? And that's not exactly *new* new.

The rest of it is just reorganizing and adapting existing technologies in new ways. Massively parallel specialized chips rather than a few generalized powerful ones, neh?

Edit: Oh, right, the "memristors." Yeah that one's a little more new.

[miauw62]

The reason we're still using old shitty instruction sets is that all programs rely on them tho, so developing for those servers probably isn't going to be simple.

E:
I have to give HP credit for that Asimov reference though.

[Frumple]

Well, they need to hurry up and use these awesome new(ish) instruction sets/hardware setups to build programs that update all the old programs, then. Things would be able to roll in pretty easy after that, I would think. Sounds pretty simple to me. Not easy, mind, but definitely simple

[miauw62]

I think you underestimate the sheer amount of inertia that programs has.
Python 3.0 was released in 2008 but like half of all Python libraries still uses 2.7 to the point where newbies ask themselves "should I learn 2.7 or 3.4?"

[Sergarr]

Well, they need to hurry up and use these awesome new(ish) instruction sets/hardware setups to build programs that update all the old programs, then. Things would be able to roll in pretty easy after that, I would think. Sounds pretty simple to me. Not easy, mind, but definitely simple

But then Toady would have to rewrite DF...

[Reelya]

It's a little beyond just rejigging instruction sets to be more efficient, that's just RISC processor design which has been around since the 1980's. How they're leveraging more speed is by actually hard-wiring common algorithms into the silicon itself. Hence, you skip the whole "read instruction, process instruction, set up data, execute instruction" rigmarole, and just have a circuit which does that job without all the overhead or checking or memory reads needed by the instructions.

[Sergarr]

It's a little beyond just rejigging instruction sets to be more efficient, that's just RISC processor design which has been around since the 1980's. How they're leveraging more speed is by actually hard-wiring common algorithms into the silicon itself. Hence, you skip the whole "read instruction, process instruction, set up data, execute instruction" rigmarole, and just have a circuit which does that job without all the overhead or checking or memory reads needed by the instructions.

Or, in other words, instead of digital, it's analogue.

Cheers for analogue computations! Hurray!

[Putnam]

Still digital... it's about as analogue as your GPU.

[Solifuge]

Hey, know what's cool?

Fusion reactors powered by molten metal and hammers.

Seriously... this solves a crapton of problems with existing fusion reactor designs, has a built-in method for negating Neutronicity and associated radiation damage to operators and equipment, and is comparatively cheap on energy input needed. If I (DISCLAIMER: as a layperson) had to pick a candidate for the most practical first Fusion Reactor, this would be it.

[Sirus]

Hey, know what's cool?

Fusion reactors powered by molten metal and hammers.

Seriously... this solves a crapton of problems with existing fusion reactor designs, has a built-in method for negating Neutronicity and associated radiation damage to operators and equipment, and is comparatively cheap on energy input needed. If I (DISCLAIMER: as a layperson) had to pick a candidate for the most practical first Fusion Reactor, this would be it.

Unfortunately the website began having problems and I couldn't get more than 2 minutes into the video. But...holy crap.
That sounds ridiculously dwarfy.