Intel Shows 48-core x86 Processor
Luis Digital
Posts: 371
More and more multi-core processors come out.
Intel Shows 48-core x86 Processor as Single-chip Cloud Computer
But do not run on batteries.
Intel Shows 48-core x86 Processor as Single-chip Cloud Computer
But do not run on batteries.
said...
Intel lists the power consumption as ranging from 25w to 125w
Comments
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
For me, the past is not over yet.
We do not really want any more witch hunt threads (do we?).
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Style and grace : Nil point
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Links to other interesting threads:
· Home of the MultiBladeProps: TriBlade,·RamBlade,·SixBlade, website
· Single Board Computer:·3 Propeller ICs·and a·TriBladeProp board (ZiCog Z80 Emulator)
· Prop Tools under Development or Completed (Index)
· Emulators: CPUs Z80 etc; Micros Altair etc;· Terminals·VT100 etc; (Index) ZiCog (Z80) , MoCog (6809)
· Search the Propeller forums·(uses advanced Google search)
My cruising website is: ·www.bluemagic.biz·· MultiBladeProp is: www.bluemagic.biz/cluso.htm
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Style and grace : Nil point
This announcement puts him and other serious multi-core solutions on the street.
Get 128 of them together and some serious processing could be done. Them some fool would stick it in a cyborg which could travel back in time.
If your surname is Conner, get unlisted !!!!
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Style and grace : Nil point
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
What makes this un-serious?
I wonder at which point the overheads of the software sharing/steering and the resultant recombinations outway the advantages, even one core requires thread flushes.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Style and grace : Nil point
But, if they put Windoze on it (likely), they will require 30% for the security features, 30% for the virus scan software, say 10% for the overhead control. That leaves 30% for real work. Not bad 16 x 64MIPs = 960 MIPs useful. LOL
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Links to other interesting threads:
· Home of the MultiBladeProps: TriBlade,·RamBlade,·SixBlade, website
· Single Board Computer:·3 Propeller ICs·and a·TriBladeProp board (ZiCog Z80 Emulator)
· Prop Tools under Development or Completed (Index)
· Emulators: CPUs Z80 etc; Micros Altair etc;· Terminals·VT100 etc; (Index) ZiCog (Z80) , MoCog (6809)
· Search the Propeller forums·(uses advanced Google search)
My cruising website is: ·www.bluemagic.biz·· MultiBladeProp is: www.bluemagic.biz/cluso.htm
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
24 bit LCD Breakout Board now in. $21.99 has backlight driver and touch sensitive decoder.
From the 1984 Mac Introduction www.youtube.com/watch?v=G0FtgZNOD44
Note the computer boots and runs the entire presentation off a 400K floppy!
We used a Linux port on 16 MIPS64 core chips (6 chips per blade) for data-center
network packet "threat" sniffing (a serious application with $500M+ annual revenue).
At the time, we started getting big memory arbitration jams at about 12 cores with
SMP on some memory intense applications ... typical PC-like apps could run fine.
I suspect that "CPU Aggregation" will become a PC buzz phrase. In any event,
this multi-core x86 thingy could be a factor to drive sales growth for Intel (who
desperately needs some to make investors happy with a stock price climb).
SGI has gone down this road with it's NUMA machines, where the OS presents ONE image to it's users that embodies all the hardware. 1K CPU's, all active and available, along with a ton of RAM, Disk, etc... is typical of what an OS like Irix, and now Linux can do. Frankly, that's really killer stuff to work on.
On those kinds of computers, latency and throughput is a big deal, as the kinds of interconnects between computing units, or nodes, impact the usefulness of the nodes operating in parallel. The OS kernel that manages these things is extremely complex, and hardware dependent.
Windows isn't there AT ALL.
On machines where that latency is higher, some parallel problems are off the table.
eg: Rendering movie frames is a parallel problem, where each problem work unit can be computed as a discrete thing. This can work on a very high latency and low throughput systems. Communication between nodes is kept to a minimum. Windows can do this kind of thing easy. One stripped down OS per node, each running interconnect software.
Solving a computational fluid dynamics problem, or weather simulation, are examples of problems where the nodes must be solving simultaneously, mandating low latency and high throughput. Windows does not have the kernel for this.
Specialized hardware like this is very, very expensive and the cooling and power requirements put it off the table for all but the most wealthy. Exactly the opposite targets for windows, and this is why that kind of development has not occurred.
Intel sees the same niche here it did with workstations. At first, general purpose PC computers were kind of crappy compared to holistically designed UNIX ones.
Over time, that changed as the specialized machines lost out due to economies of scale, and incremental, if brute force general purpose software development.
Supercomputers are now in that realm as workstations were.
The Intel chip is a general purpose computer, that can begin to solve the kinds of problems a NUMA machine would, just smaller in scope. This is exactly what the PC was to the UNIX workstation in the 90's.
Economies of scale will render combination of these 40 core CPU's feasible, and they strike a happy middle! Clumps of the nodes can intercommunicate, or share RAM easily, requiring only expensive communication between distinct multi-core units!
Windows will go here, and it won't need the complexity of the older systems, and it will deliver a lot of the bang, for a fraction of the buck.
This is disruptive technology, a classic case! Highly parallel computing will be put within reach of more ordinary people.
The problem this time, unlike the workstation one, is the huge amount of software written to sequential machines, not parallel machines. That's an interesting dynamic here, I will be curious to see play out.
The solution on the Intel deal will be either:
Virtualization, where entire environments can be run together on less hardware, less power. (cloud in a box) There is a lot of business merit to this, leveraging shared resources, currently spread across application servers. I really love virtual computing with Windows. There are a lot of things about that OS that scare the heck out of me, when running live. It's a risky thing, and planning is difficult, largely due to the somewhat unpredictable nature of the updates, and OS design.
Compartmentalization, where the multi-core elements are exposed through well realized solution software and focused on specific, repeatable problems. IMHO, specific problems can be boiled down to managable bits, packaged more nicely for ordinary people, not rocket scientists, and make good use of the multi-core compute attributes, while still performing nicely enough on the larger body of linear computing centric software. Too darn bad Windows isn't a real multi-user OS.
High end work-servers, where a real multi-user OS can perform massive compute shared among many users. Windows can do this well with client server applications that target specific problems. For the open computing crowd, these machines can be used to author solution code, and perform limited scope solves before committing to the big iron.
Another niche will be multi-media. Many things can be parallelized, and the work flow is multi-user friendly now. If there is a Linux on this, the TV and Film studios will probably jump on these, perhaps sharing one among a few users, where they've got a cluster now. Keep the cluster until the tech matures, and get quicker, higher quality feedback before a commit to a long run compute job, just like the rocket scientists do.
I really want to say CAD gets a boost here. Some of the CAD programs out there have the design attributes necessary to parallelize model generation. Many of them don't. For the ones that do, it's going to be interesting to see whether or not surrendering full on parametric design to get the compute speed will be worth it. On larger assembly models, I suspect it will as many of those constraint models are best solved simultaniously. Could be big news for Aerospace / Automotive / Machine design, where these industries are sharply limited by linear compute speeds capping where they are today. Big memory has helped them, but the problem remains compute. Some trade-offs in how the models are built could open some big doors.
One sure result will be the entire low to mid market for super computing will do exactly what it did on the PC, and it will do it in many of the same ways. Kiss the little clusters good bye. It's either going to be mega huge, or general purpose.
Edited, well. Just because.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Propeller Wiki: Share the coolness!
Chat in real time with other Propellerheads on IRC #propeller @ freenode.net
Safety Tip: Life is as good as YOU think it is!
Post Edited (potatohead) : 12/5/2009 9:20:29 AM GMT
I just made a post about this over in the Sandbox....looked around and did
not see anything posted there about this... did not think to look in this forum.
Sorry all. :-(
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
"Where am I? Where am I going? Why am I in a handbasket?"
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Links to other interesting threads:
· Home of the MultiBladeProps: TriBlade,·RamBlade,·SixBlade, website
· Single Board Computer:·3 Propeller ICs·and a·TriBladeProp board (ZiCog Z80 Emulator)
· Prop Tools under Development or Completed (Index)
· Emulators: CPUs Z80 etc; Micros Altair etc;· Terminals·VT100 etc; (Index) ZiCog (Z80) , MoCog (6809)
· Search the Propeller forums·(uses advanced Google search)
My cruising website is: ·www.bluemagic.biz·· MultiBladeProp is: www.bluemagic.biz/cluso.htm
Will there be a production model when the die size is huge and when chips are subject to defects, you either have to use a majority
·of the cores or throw out chips because one of 48 cores are defective?
How does such a chip handle sharing a device?
If you have 48 cores and all of them want the disk drive at the same time, will there be a problem?
Are system programmers really waiting for such monster chips?
re the disk access, same as pc's hooked to a server. You wait in line.
Yes, server farms will use them by the truckloads. And Google (cannot remember what they call it) will love it because the power will be concentrated on the server,
and the pc will become dumb again. Hey, goodbye windoze for the masses. Office etc will reside and run on the server. No more licenses, just ads [noparse]:([/noparse]
So, the concentrated, distributed, concentrated, distributed, concentrated circle continues.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Links to other interesting threads:
· Home of the MultiBladeProps: TriBlade,·RamBlade,·SixBlade, website
· Single Board Computer:·3 Propeller ICs·and a·TriBladeProp board (ZiCog Z80 Emulator)
· Prop Tools under Development or Completed (Index)
· Emulators: CPUs Z80 etc; Micros Altair etc;· Terminals·VT100 etc; (Index) ZiCog (Z80) , MoCog (6809)
· Search the Propeller forums·(uses advanced Google search)
My cruising website is: ·www.bluemagic.biz·· MultiBladeProp is: www.bluemagic.biz/cluso.htm
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
I have to check everything I type because my windoze vista loses typed characters randomly [noparse]:([/noparse]
Now... Prop III with 1 billion transistors 35nm ...
That would be one hell of a Cray
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Links to other interesting threads:
· Home of the MultiBladeProps: TriBlade,·RamBlade,·SixBlade, website
· Single Board Computer:·3 Propeller ICs·and a·TriBladeProp board (ZiCog Z80 Emulator)
· Prop Tools under Development or Completed (Index)
· Emulators: CPUs Z80 etc; Micros Altair etc;· Terminals·VT100 etc; (Index) ZiCog (Z80) , MoCog (6809)
· Search the Propeller forums·(uses advanced Google search)
My cruising website is: ·www.bluemagic.biz·· MultiBladeProp is: www.bluemagic.biz/cluso.htm
Post Edited (Cluso99) : 12/6/2009 6:56:31 AM GMT