- Super-computers:
- Super-computer =
a computer that can
perform a large number of
compututations
per unit of time
compared to a
general-purpose computer
(= the CPU that we have studied so far)
Wikipedia page: click here
- Super-computer =
a computer that can
perform a large number of
compututations
per unit of time
compared to a
general-purpose computer
(= the CPU that we have studied so far)
- The first generation of
super computers were:
- SIMD computers or vector computers
- One of the
most successful super-computer in
history is
Cray-1
- The Cray-1:
- The Cray-1 was
a vector-computer
Wikipedia page: click here
- The Cray-1:
- Here is a schematic
of the Cray 1's
CPU architecture:
Notice that:
- The Cray 1 has
vector registers
(I circled the
vector registers in the
picture at the top):
- Each vector register
contains multiple (float) values
!!!
(That's why it's so big in the diagram)
- Each vector register
contains multiple (float) values
!!!
- The Cray-1 has
multiple addresses registers
(I circled the
address registers at the
bottom in the
diagram):
- Each address register can contain a different address value (= location in memory)
The Cray 1 has multiple system busses:
The Cray 1 can send out multiple memory requests with different memory addresses so that:
- The Cray 1 can
read/write multiple
(upto 64 !) operands
from/to memory
at the same time !!!!
But: only if the memory operands are stored in different memory banks
- The Cray 1 has
vector registers
(I circled the
vector registers in the
picture at the top):
- Today,
the SIMD architecture
is used in the
Graphic Processor Unit (GPU)
- A popular
programming interface (API)
for GPU is:
CUDA
(Compute Unified Device Architecture)
- For more information on CUDA, see: click here