9
A Little Story About GPUs

IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

  • Upload
    npinto

  • View
    2.100

  • Download
    0

Embed Size (px)

DESCRIPTION

More at http://sites.google.com/site/cudaiap2009 and http://pinto.scripts.mit.edu/Classes/CUDAIAP2009

Citation preview

Page 1: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

A Little Story About GPUs

Page 2: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

The Wheel of Reincarnation

Page 3: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)
Page 4: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)
Page 5: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)
Page 6: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

The Wheel Today

1.5 turns around the wheel?the display processor on the display processor

Page 7: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

More than one turn around the wheel?

stream processors → raster processor → display chip

Page 8: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

The conclusion in 1968

Page 9: IAP09 CUDA@MIT 6.963 - Lecture 05: Wheel Of Reincarnation (Steven Johnson, MIT)

The conclusion in 2008?Asymmetric/heterogeneous multiprocessing is hard.

(multiple memory models, different CPU characteristics)

But…serial CPUs aren’t getting fasterBut…most programs are serial and need a fast serial CPUBut…1000 Pentium M’s (Intel Core) aren’t coming to desktops

Natural tendency to few beefy CPUs + zillions of simple ones…GPGPU (Nvidia Tesla/AMD Firestream)?Cell processors?Intel Larrabee?Something else?