Dell Talking About 80-Core Chip Processor

November 20, 2008 by John Messina

Dell slide shown Tuesday at SC08 (Credit: Dell Computer)

(PhysOrg.com) -- This week Michael Dell (CEO of Dell) gave a slide presentation that included Intel´s recently developed 80-core processor. This isn't the first time that the 80-core chip was mentioned in a conference.

Two years ago CEO of Intel, Paul Otellini, made a promise of delivering an 80-core processor within the next five years. He had also noted that the chips should be able to swap data at a terabyte a second (see video).

Video: Intel tests chip design with 80-core processor

This does not mean that future designs of the x86 chip is dead. We will continue to see more and more cores with increased performance and without more power required. The trend, for Intel, is to pack more computing power into smaller machines thereby creating desktop sized supercomputers.

Intel's intention is to continue to bring out many-core processors including its upcoming Larrabee graphics chip and future server processors that may reach 32 cores. Currently, Intel´s Dunnington processor gets the prize for the most cores.

© 2008 PhysOrg.com

• ShadowRam - Nov 20, 2008
  • Rank: 2 / 5 (6)
  This is not the direction to go.
  
  Optimal solution = Main Core.. 2 or 3 cores..
  and a Massive Parallel Co-processor..
  
  (See nVidia CUDA)
  
  If Intel goes this route.. AMD and ATI's merge to make a processor as I mentioned above will destory Intel...
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• gmurphy - Nov 20, 2008
  • Rank: 3.3 / 5 (4)
  I program on cuda platforms, they're very powerful but very inflexible. Memory latency is huge so its better to do local computation where possible but the local cache for each multi processor is only 16kb. If intel can produce the goods, it would be very useful for developers, period.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• GrayMouser - Nov 20, 2008
  • Rank: 2.3 / 5 (3)
  Reminds me of the Transputer chip based systems.
  And the Butterfly Machine. Of course, with the Butterfly Machine, they found out that the configuration of the processors (grid, hypercube, etc.) didn't make a lot of difference in the time it took to solve problems.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• fleem - Nov 20, 2008
  • Rank: 2 / 5 (1)
  Right about the memory latency. The vast majority of parallel applications need at least a little main memory accesses PER CORE. 80 cores will want a LOT of memory accesses, and they won't get them in the same chip. There are some apps that crunch for a long time on the same little bit of data, but not many.
  
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• Star_Gazer - Nov 20, 2008
  • Rank: 1 / 5 (1)

  This is not the direction to go.
  
  Optimal solution = Main Core.. 2 or 3 cores..
  and a Massive Parallel Co-processor..
  
  (See nVidia CUDA)
  
  If Intel goes this route.. AMD and ATI's merge to make a processor as I mentioned above will destory Intel...

  
  
  thats why Cell CPU in my PS3 uses 1 main core and 8SPEs
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• lengould100 - Nov 21, 2008
  • Rank: 5 / 5 (1)
  An 80 core cpu needs an OS which can dedicate assign specific groups of CPU to specific operating threads continuously, with each group having direct access to RAM and I/O. Interesting problem...
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• lengould100 - Nov 21, 2008
  • Rank: not rated yet
  Processors should be grouped into 20 groups of 1 to 3 processors (average 2). Remaining processors available to be assigned as helpers to any group which is working too hard. Each group gets its own on-chip cache and stacks. OS then asigns threads to groups according to priority.
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• lengould100 - Nov 21, 2008
  • Rank: not rated yet
  If the system had available guaranteed maximum access delays to disk and etc., then at least one of the groups should be capable of critical real-time control... sweet. Robots, auto automation, ... needs a coprocessor for analog and real-world interfaces...
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• thematrix606 - Nov 21, 2008
  • Rank: not rated yet
  Does this mean I will finally be able to get 30 fps in Crysis? Hardware will finally catch up games! Yay, about time :)
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5
• jonnyboy - Jan 24, 2009
  • Rank: not rated yet

  An 80 core cpu needs an OS which can dedicate assign specific groups of CPU to specific operating threads continuously, with each group having direct access to RAM and I/O. Interesting problem...

  
  
  which, of course, brings up the problem of controlling and coordinating all the different accesses to the memory.
  
  Sounds like a slow death to me
  • report abuse
  • • Current rank
    • 1
    • 2
    • 3
    • 4
    • 5