Hard Light Productions Forums
Off-Topic Discussion => General Discussion => Topic started by: Sandwich on September 24, 2003, 12:55:02 am
-
http://www.cnn.com/2003/TECH/ptech/09/23/sun.chip.reut/index.html
One of those "Duh!" moments. ;) Pretty cool. :yes:
-
it sounds so logical it's a wonder nobody thouhgt about it before, tho :p
-
Indeed. Man, I haven't kept up with all the CPU development of late... I'll need to soon, since I hope to upgrade by January. :nod:
-
Yeah, I want to upgrade from my old laptop soon. But my new apartment is lacking of space to put a system. Don't know where to put it and still make the place look nice :sigh:
-
Given who this is from, don't expect much.
Sun works in the server space, not the desktop. Even the desktops are server grade processors. Their chips are are all SPARC derived designs and as such do not run x86 apps, like games. I don't think there's even a major 3d app ported to them. Oh and don't forget their fastest chip is still only a 1.x gig (I think, they may have a faster one now).
More importantly, Intel (and consequently, AMD) is very unlikely to pick up on this idea. This edge to edge design would allow core speeds for the memory, but only at the cost of radically changing the addressing methodology the CPU uses. Since Intel enforces backward (and bug) compatibility back to the 8086, this is highly unlikely.
-
I just hate to think how much a chip like that would cost.....
-
I'd be more concerned about how much cooling the damn thing needs and how much this will limit upgradeability :)
Flipside :D
-
Don't forgot about the kind of power somehting like that would use.
-
Originally posted by Flipside
I'd be more concerned about how much cooling the damn thing needs and how much this will limit upgradeability :)
Flipside :D
Well if its anything like all the Sun boxes in my lab now... lots and lots of cooling, in very specific, very controlled ways. You literally cannot run some of these machines with the case open. If you do, the flow channel is not properly closed off and the CPUs crack. At $10K/cpu (1.1GHz 64bit uSparcIIIcu i think.), this is rather... undesireable.
-
Its suspiciously devoid of any details at all... sounds like a paper release *sigh*
Lightwave, PRman, Mental Ray and Houdini are available in SPARC builds, i beleive.
-
Originally posted by Flipside
I'd be more concerned about how much cooling the damn thing needs and how much this will limit upgradeability :)
Flipside :D
since it's direct connection, I suppose there should be less heat, no? drop a fan over all those processors, and it's simpler to cool down, I reckon, than lots of processors scattered on a board that can't be cooled down at all.
-
It depends, a lot of the reason that motherboards are as big as they are is because if they are too close together, the combined heat of all the components is more than can be air-cooled. That's why this chip is aimed at the server market, because I would suspect you need some very major cooling on it.
Flipside :D
-
Originally posted by wEvil
Lightwave, PRman, Mental Ray and Houdini are available in SPARC builds, i beleive.
Who in their right minds wuld USE a SPARC for 3d though? They're slower and more expensive than pretty much every other server space processor out there right now. They make for decent telco gear, and they used to be among the most reliable servers for data centers. However, their reliability has gone straight to hell of late (3/10 of my new rack boxen from them are DOA).
-
Yep, but they used to be fairly nippy in comparison about 5 years ago :p
-
Originally posted by Kosh
I just hate to think how much a chip like that would cost.....
Originally posted by Kosh
Don't forgot about the kind of power somehting like that would use.
"It's faster, cheaper, and uses less power," said John Gustafson, principal investigator for Sun's high productivity computing systems.
-
Originally posted by wEvil
Yep, but they used to be fairly nippy in comparison about 5 years ago :p
well, my PC was top of the line 5 years ago :p ( it probably didn't even exist yet, now that I think about it :doubt: )
-
This sounds like an idea a couple of stoned engineering students would have.
"So... like, dude... to make the chips talk faster, just jam them right up against each other!"
...thereby ignoring all the reasons they weren't like that in the first place. If you're going for this kind of speedup, just put everything on the same die.
-
Originally posted by ZylonBane
...thereby ignoring all the reasons they weren't like that in the first place.
Like what?
-
Cooling, routing, buffering, "glue", et al.
Benefits of this technique will likely be balanced by increased complexity accessing the chips that aren't rammed right up against it. TNSTAAFL.
-
So they've run up against the barrier of lightspeed already?
I can't believe that the speed of transmission of electrical signals between chips is already limiting processing speeds. I thought they'd hit the Uncertainty Principle first, what with the drive to miniaturise and everything.
-
At the speed recent chips run, yes, electrical transmission speed is a huge issue. Has been for years.
-
Originally posted by mikhael
Who in their right minds wuld USE a SPARC for 3d though?
True 64-bit addressing, I reckon.
-
Originally posted by aldo_14
True 64-bit addressing, I reckon.
You'd be surprised how little that actually gains you for 3d applications. Sure you get a hell of a lot of precision behind the decimal point, but its not really good for much of anything. After you get past the initial 16 bits of accuracy, we're already to a point where the resolution of your display or printer will prevent you from seeing any inaccuracies. They're that small.
Besides anything else, the average modeller, animator or renderer are all going to go for cheaper, faster chips. Even the fastest P4 or Opteron is cheaper and faster than fastest SPARC.
-
Originally posted by ZylonBane
At the speed recent chips run, yes, electrical transmission speed is a huge issue. Has been for years.
The speed of light issue doesn't come up until we near hundreds of gigs. What we're dealing with right now is how small can we really go with what we're using?
Also, the idea is a bit more sane.