flat assembler
Message board for the users of flat assembler.

Index > Heap > Intel Readies Massive Multicore Processors

Goto page 1, 2, 3, 4, 5  Next
Author
Thread Post new topic Reply to topic
HyperVista



Joined: 18 Apr 2005
Posts: 691
Location: Virginia, USA
HyperVista
Hey, check this out! Imagine, 42 x86 cores, 18 accelerator cores, and 4 embedded grapics cores!!! Time to brush up on multi-threading techniques and bus snooping for memory coherency!! Shocked
Post 14 Jun 2007, 15:17
View user's profile Send private message Visit poster's website Reply with quote
arafel



Joined: 29 Aug 2006
Posts: 131
Location: Jerusalem, Israel
arafel
CPUs.... CPUs....
When they already invent something better than rotating platters hdd. These things just keep failing.

_________________
lscr::winstuff
Post 15 Jun 2007, 22:13
View user's profile Send private message Reply with quote
quasar



Joined: 09 Oct 2005
Posts: 16
quasar
They did. Try solid state disk (if you have enough $$$$).
Post 16 Jun 2007, 08:05
View user's profile Send private message Reply with quote
arafel



Joined: 29 Aug 2006
Posts: 131
Location: Jerusalem, Israel
arafel
I meant invent something practical Smile
The moving mechanical parts got replaced by limited write/erase endurance. not sure if that's better.

_________________
lscr::winstuff
Post 16 Jun 2007, 08:46
View user's profile Send private message Reply with quote
f0dder



Joined: 19 Feb 2004
Posts: 3170
Location: Denmark
f0dder
I'd rather want a 32gig fast solid-state disk than a 42-core machine... Smile
Post 16 Jun 2007, 11:35
View user's profile Send private message Visit poster's website Reply with quote
Mac2004



Joined: 15 Dec 2003
Posts: 313
Mac2004
Quote:

Hey, check this out! Imagine, 42 x86 cores, 18 accelerator cores, and 4 embedded grapics cores!!! Time to brush up on multi-threading techniques and bus snooping for memory coherency!! Shocked


Sounds nice... I guess 'some' of today's programs need to be updated as well. Smile
Post 16 Jun 2007, 11:39
View user's profile Send private message Reply with quote
kohlrak



Joined: 21 Jul 2006
Posts: 1421
Location: Uncle Sam's Pad
kohlrak
This would be all fine and dandy if you did something about the RAM clock issue...
Post 16 Jun 2007, 17:48
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
Amusing to hear the news comes from Intel who has always had a problem with putting more processors on a single board and maintaining per-CPU efficiency.

So what will it be? 42 cores and 4 GB ram? Laughing
Post 16 Jun 2007, 23:36
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
kohlrak



Joined: 21 Jul 2006
Posts: 1421
Location: Uncle Sam's Pad
kohlrak
I'm assuming it'll be 64 bit, so it'll probably have more than 4 gigs of ram.

Heck, we'd be fine with 1 core. They also said that if we got fiber optics it'd be faster, too. With all the money we're spending, heck we could buy parts for an air conditioner or the top part of the fridge and build a better coolent system and just overclock everything the heck and back. Much cheaper, works better.
Post 17 Jun 2007, 03:46
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
kohlrak wrote:
I'm assuming it'll be 64 bit, so it'll probably have more than 4 gigs of ram.


The article says nothing about 64bit. If it was 64bit, Intel would have to play AMD's game. I doubt they'll do that.

Quote:
Heck, we'd be fine with 1 core. They also said that if we got fiber optics it'd be faster, too. With all the money we're spending, heck we could buy parts for an air conditioner or the top part of the fridge and build a better coolent system and just overclock everything the heck and back. Much cheaper, works better.


I agree that the multiple cores is just a marketing thing, just like netburst. Just that with netburst, it was the GHz that counted the most money and with this new trick it's the number of cores that counts. Considering the point that they'll still be putting the implementation on the same sized die... unless they suddenly devoloped some super duper magical approach to making cores 100 times smaller, there will really be no factual performance gain.
Post 17 Jun 2007, 12:18
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
kohlrak



Joined: 21 Jul 2006
Posts: 1421
Location: Uncle Sam's Pad
kohlrak
Quote:
I agree that the multiple cores is just a marketing thing, just like netburst. Just that with netburst, it was the GHz that counted the most money and with this new trick it's the number of cores that counts. Considering the point that they'll still be putting the implementation on the same sized die... unless they suddenly devoloped some super duper magical approach to making cores 100 times smaller, there will really be no factual performance gain.


True, i forgot about size. To me, it's all a matter of heat. The resons our stuff is clocked so low is for selling over clocked versions with new names and because of heat. Remove the heat issue and our stuff would be running pretty nicely... And i've noticed "faster than dialup" has been a big argument for marketing dialup service in my country. "High speed dialup." People are still buying into it.
Post 17 Jun 2007, 22:58
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger MSN Messenger Reply with quote
f0dder



Joined: 19 Feb 2004
Posts: 3170
Location: Denmark
f0dder
kohlrak wrote:

Heck, we'd be fine with 1 core.

Not really, things are much smoother once you go multi-core. Even if you run single-threaded applications, your overall system performance is nicer. 4 cores is a bit over the top for regular home users, and eight cores is overkill unless you know you're going to need it... but there are people who can actually utilize that much power. It's easier throwing 8 cores on a die than making a single core 8 times faster.

DustWolf wrote:

The article says nothing about 64bit. If it was 64bit, Intel would have to play AMD's game. I doubt they'll do that.

Play AMD's game? And what is that? Being smeared all over the floor by the competition? Smile
Post 18 Jun 2007, 11:28
View user's profile Send private message Visit poster's website Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
kohlrak wrote:
True, i forgot about size. To me, it's all a matter of heat. The resons our stuff is clocked so low is for selling over clocked versions with new names and because of heat. Remove the heat issue and our stuff would be running pretty nicely...


I think the heat problem is rather tough to do away with. The current materials are not flawless heat conductors. And transistors simply stop functioning at odd temperatures. Even if you cooled something with liquid helium, the point that the heat wouldn't be equally spread out could make some areas overheat while others would freeze over.

Also I'd guess that transistors have a limit to their responsivness, regardless what voltage you apply.
Post 18 Jun 2007, 15:50
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
f0dder wrote:
Not really, things are much smoother once you go multi-core. Even if you run single-threaded applications, your overall system performance is nicer. 4 cores is a bit over the top for regular home users, and eight cores is overkill unless you know you're going to need it... but there are people who can actually utilize that much power. It's easier throwing 8 cores on a die than making a single core 8 times faster.


So what'd be the point of making a general purpose multi-core machine? As I recall statistics, multicore systems are having a hard time utilizing the full potential of each core. And given the way current operating systems are put togather: shared buffers, semaphores and all, I can't say I'm suprized.

f0dder wrote:
DustWolf wrote:

The article says nothing about 64bit. If it was 64bit, Intel would have to play AMD's game. I doubt they'll do that.

Play AMD's game? And what is that? Being smeared all over the floor by the competition? Smile


I didn't get the distinct impression AMD got smeared anywhere. In places where lawyers count perhaps, but not in places where technology does. AMD set some standards when they implemented the 64bit architecture. I doubt everybody will be willing to recode all their 64bit apps just because intel decides it wants it's own implementation. Any deviation from the enstablished standard would be considered a bug.
Post 18 Jun 2007, 15:57
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
f0dder



Joined: 19 Feb 2004
Posts: 3170
Location: Denmark
f0dder
DustWolf wrote:

So what'd be the point of making a general purpose multi-core machine?

Think outside the box for a bit. Remember that intel has a lot more going on than just x86... being able to reconfigure/reroute around a dead or overheating processor could be interesting. Or having a bunch of slightly slower (less heat and power consumption) processors with really fast interconnects could be interesting for massively parallel computing tasks.

Too bad if they're going to ruin it by slapping on relatively large x86 frontends.

DustWolf wrote:

As I recall statistics, multicore systems are having a hard time utilizing the full potential of each core. And given the way current operating systems are put togather: shared buffers, semaphores and all, I can't say I'm suprized.

It's a mix of the fact that not all algorithms are easily to make parallel, that memory buses aren't fast enough (NUMA is one solution/workaround), that some work is still needed on interconnects, etc.

And there's a difference between servers, workstations, and scientific use as well. Intel's idead of shared cache is pretty decent imho, less bother syncing each core's view of physical memory, and the potential for actually sharing data instead of possibly wasting precious cache memory by having duplicate data in each core.

DustWolf wrote:

I didn't get the distinct impression AMD got smeared anywhere.

Intel introduced core2, and AMD still hasn't come up with a proper response. And unless something radical happens with their new acrhitecture during fabrication, they have big problems.

DustWolf wrote:

AMD set some standards when they implemented the 64bit architecture.

And tied us to the crappy patchwork x86 architecture for even longer. Way to go. First innovation that AMD ever did, anyway, rather than just copying intel's tech Smile

DustWolf wrote:

I doubt everybody will be willing to recode all their 64bit apps just because intel decides it wants it's own implementation. Any deviation from the enstablished standard would be considered a bug.

Ummm, say what? Intel has had x86-64 (licensed from AMD) for sevaral years now, from late model Pentium 4's across core2. Which world are you living in?

Amd screwed up with x86-64 btw, they initially didn't include CMPXCH16B. (But okay, first-gen intel implementation of x86-64 didn't have NX bit support).
Post 18 Jun 2007, 23:38
View user's profile Send private message Visit poster's website Reply with quote
kohlrak



Joined: 21 Jul 2006
Posts: 1421
Location: Uncle Sam's Pad
kohlrak
Quote:
I think the heat problem is rather tough to do away with. The current materials are not flawless heat conductors. And transistors simply stop functioning at odd temperatures. Even if you cooled something with liquid helium, the point that the heat wouldn't be equally spread out could make some areas overheat while others would freeze over.


Careful implimentation. When you set up the coolent system, have a thermometer. The hotter it gets, the more you release in that section. Though, i'm guessing that if the processor is burning hot, the whole thing is burning hot. If the ram is burning, throw the stuff onto the hottest spot (or the whole thing, but that would take some studies). And yea, it is tough to do away with. The stronger the coolent, the more careful you must be.

Quote:
Also I'd guess that transistors have a limit to their responsivness, regardless what voltage you apply.


Last time i checked, clock is the freaquency not the amplitude.
Post 19 Jun 2007, 01:37
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
kohlrak wrote:
Careful implimentation. When you set up the coolent system, have a thermometer. The hotter it gets, the more you release in that section. Though, i'm guessing that if the processor is burning hot, the whole thing is burning hot. If the ram is burning, throw the stuff onto the hottest spot (or the whole thing, but that would take some studies). And yea, it is tough to do away with. The stronger the coolent, the more careful you must be.


Seems like putting a lot of effort into something that only solves a problem indirectly... well my point is below.

Quote:
Quote:
Also I'd guess that transistors have a limit to their responsivness, regardless what voltage you apply.


Last time i checked, clock is the freaquency not the amplitude.


Increased clock does not cause more heating, increased voltage does. The point is that if you clock the processor abovie the response time of the transistors, you inherently introduce bugs (or incorrect function, as the transistors don't have the time they need to flip their states). The response time of the transistors, however, is dependant on the voltage applied. Hence in order to push into higher frequencies, you have to apply more voltage and subsequentially more cooling.

There is also the point tho, that the smaller the transistors are, the less voltage you need to get them to be more responsive. According to current devolopment I'd say we're hitting the limit (due to size of atoms) in about 16 years from now or so.
Post 19 Jun 2007, 20:06
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
kohlrak



Joined: 21 Jul 2006
Posts: 1421
Location: Uncle Sam's Pad
kohlrak
Which means our infinately powerful machines need a new way of making parts.
Post 19 Jun 2007, 20:12
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
f0dder wrote:
It's a mix of the fact that not all algorithms are easily to make parallel, that memory buses aren't fast enough (NUMA is one solution/workaround), that some work is still needed on interconnects, etc.


Just pointing out that AMD has been using that approach in their multi-CPU server systems all this time so far. AMD uses NUMA, AMD uses fast interconnects, Intel does neither. Intel has been staying well away from anything the competition uses so far by principle, what makes you think they will switch sides now?

Quote:
DustWolf wrote:

I didn't get the distinct impression AMD got smeared anywhere.

Intel introduced core2, and AMD still hasn't come up with a proper response. And unless something radical happens with their new acrhitecture during fabrication, they have big problems.


I have found dependance on old sources of information to be a bad idea when you're trying to get the best for your money. Update your data sources.

Quote:
DustWolf wrote:

AMD set some standards when they implemented the 64bit architecture.

And tied us to the crappy patchwork x86 architecture for even longer. Way to go. First innovation that AMD ever did, anyway, rather than just copying intel's tech Smile


Biased view. AMD did make their own CPU. Just because it's compatible doesn't mean it's all the same. Copying intel's tech is for example Cyrix.

Quote:
DustWolf wrote:

I doubt everybody will be willing to recode all their 64bit apps just because intel decides it wants it's own implementation. Any deviation from the enstablished standard would be considered a bug.

Ummm, say what? Intel has had x86-64 (licensed from AMD) for sevaral years now, from late model Pentium 4's across core2. Which world are you living in?


You wouldn't happen to have realized that if AMD designed the 64bit standards, that they might have designed them in such a way that they are more easily implementable on AMD's architecture?
Post 19 Jun 2007, 20:16
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
DustWolf



Joined: 26 Jan 2006
Posts: 373
Location: Ljubljana, Slovenia
DustWolf
kohlrak wrote:
Which means our infinately powerful machines need a new way of making parts.


No doubt. They have researched that too.

There is no particular way to make transistors smaller than the atoms they are made of, you need a new type of transistor (or other data processing device to do so). I believe the effort was called quantum computing.
Post 19 Jun 2007, 20:21
View user's profile Send private message AIM Address Yahoo Messenger MSN Messenger Reply with quote
Display posts from previous:
Post new topic Reply to topic

Jump to:  
Goto page 1, 2, 3, 4, 5  Next

< Last Thread | Next Thread >
Forum Rules:
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum


Copyright © 1999-2020, Tomasz Grysztar.

Powered by rwasa.