Presentation

IBM Research
!
Dec 2004
"
#
© 2004 IBM Corporation
IBM Research
$
%
&
$
%
%
&
"
'
%
–
% &
%
%
- %
– .
()
,
%
"
"
/
"
,
&
"
&
"
&
&
+
%
%
%
"
, "
&
,
"
/
"
– #
&
%
"
,
"
%
,0
,
/
&
J. Moreno / December 2004
%
%
"
($
%%
%
–
2
% *+"
%
(
– #
–
%
&
"
+
© 2004 IBM Corporation
IBM Research
'%
&, ! %
'
1/
%
%
3
J. Moreno / December 2004
%
!
© 2004 IBM Corporation
IBM Research
2
–4
100
!
–$
%
,
( %
Power Density (W/cm2)
1000
3
2
!
! +
Gate Stack
10S Tox=11A
Active
Power
10
Passive Power
1
0.1
0.01
1994
2004
0.001
1
4
J. Moreno / December 2004
0.1
Gate Length (microns)
0.01
© 2004 IBM Corporation
IBM Research
2
2
6
'
(
!
"
!
.!
"
!
$
% &
"
%
$ )% &
" #
$ )% &
+
" #
&
!
%
&
#
5
!
#
+
&
"
%
%
, 5
!
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
%
6
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
%
Power Density
(Watts / cm^2)
2
Active vs Passive Power
500
400
300
200
100
0
500
Passive Power (Device Leakage)
Active Power
350
250
180
130
100
70
50
Technology (nm)
#
&
14
IB M E S 9 0 0 0
12
B ip o la r
)2
10
CM O S
8
F u jits u V P 2 0 0 0
A ir c o o l lim it
C h ip flu x 5 0 W / c m 2
IB M 3 0 9 0 S
IB M R Y 5
N TT
6
IB M R Y 7
F u ji ts u M - 7 8 0
P u ls a r
4
Module Heat Flux (W/cm
7
J. Moreno / December 2004
IB M 3 0 9 0
2
Vacuum
0
1950
IB M 3 6 0
1960
IB M 3 7 0
1970
IB M R Y 6
C D C C yber 205
IB M 4 3 8 1
IB M 3 0 8 1
F u jit s u M 3 8 0
IB M 3 0 3 3
1980
IB M R Y 4
Apache
1990
2000
2010
Y ear of A nnouncem ent
A ft e r G h o s h a l & S c h m id t
© 2004 IBM Corporation
IBM Research
""
##
!!
$$
!
!! %%
8
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
% 5$
%
+
#1 1
$ %
<::<
*
@A
%
/
&&
!!
%
$
-
!!
!!
!!# # %%
$$
"
&
%4
6
.
<::=
&
9;
"
,
&(22 =
=:
+
'
$
4
%,
#
>
9
<::=
?<
%
<::=
<;
7 &
<::=
;<
J. Moreno / December 2004
?;
<
%
(
"
,
(22 89
:
+
3
7
% <+
!
(22 8
9:
+
© 2004 IBM Corporation
IBM Research
B &$
&
Bipolar
Frequency
CMOS
Frequency & density
CMOS SoC
Integration & density
CPU 1
L2 Cache
CPU 2
Blue Gene L
Multi-Chip Module
(MCM)
Power 5
Multi-core microprocessor,
ASIC library, design tools
1980’s
1990’s – 2000’s
2004 & Beyond
(Note: Figures not to scale)
10
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
&
2
%
2
&
%"
,
%
%
%
%
!
2
,
'
11
%
%
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
'%
&, ! %
–
,
"/
'
1/
%
%
12
J. Moreno / December 2004
!
© 2004 IBM Corporation
IBM Research
2
"
– ,
-
"
,
%
%%
%
– $
& " &
–
&
"
%
6
6
% ,
&
"
C
"
%,
&
"
%
3
D
6
"
%
% &
"%
%
,
".
$
%,
(
!"
"
%
6
&
-&
–
$
<:
:
8<
:;
:"
– +
&
+
'
,
!
-
;
:
:=
::
:2 ,&
,&
C
%,
&
– $
13
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
2
%
&
%
'
.
.
.
.
(
(
$
$
+
+
""
""
14
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
%
&
SMP
Core
Accelerator
Cache
Memory
...
Blades
%
–
–
$
15
"
%, &
,
,
$ &E$
&
,
,
% , &(
J. Moreno / December 2004
&
,% %
%,
(
/
"
&
"
% +FG
G
G
%
&
" +
© 2004 IBM Corporation
IBM Research
%4
PLB(4:1)
5
/
32k/32k L1
“DoubleFPU”
snoop
32k/32k L1
128
440CPU
I/Oproc
L2
'
'!2
%
'/
+
+
256
128 L2
440CPU
*
Multiported
Shared
SRAM
Buffer
256
256
4MB
EDRAM
Shared
L3directory
for EDRAM
.
.
L3Cache
1024+
or
144ECC
Memory
.
0
1
IncludesECC
256
“DoubleFPU”
+
"
/
/
128
16
/
Ethernet
Gbit
JTAG
Access
Gbit
Ethernet
JTAG
Torus
Tree
6out and
3out and
6in, each at 3in, each at
1.4Gbit/s link 2.8Gbit/s link
Global
Interrupt
4global
barriers or
interrupts
J. Moreno / December 2004
DDR
Control
withECC
144bit wide
DDR
256/512MB
1
*
/
*
/
1
$+
/
)
© 2004 IBM Corporation
IBM Research
%
3
%
– #
&
%
%
% "
%
&
%
– $
(
%
% ,%
"&
%
%
– C
%
%
&
%
+
(
+
1
3
&
&
6
"C 3 "
/ %
"
,"
,
%
17
&
"%
J. Moreno / December 2004
% %
© 2004 IBM Corporation
IBM Research
%
(())
&
(())
()+
*
, *
–
)
H
%
%
6
,
&
E
/
–
&
&
6 !
–1 ,
–
6
%,
F
,
&
6
F
F
&
%
1/
#
, &(
+
&
%
I 3 %
%
– (3 % &
"
%
%
–
6
– %
–#
3
%
/
,
!
18
()
()
*
,4
3
–#
(),
%
, &
I B&
%
%
/
%
&
"
%
,,&
+
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
$
%
%
&
(
()
()
()
))
-
--
(())
**
()+
--
19
**
(())
J. Moreno / December 2004
--
(),
(())
(())
© 2004 IBM Corporation
IBM Research
Off-load engines and accelerators
Multiple cores with same ISA
%( &
–
– 1 %
)$ 2
– 1 &%
+
%
&
– '
)$ 2
*
,
*
%&
"
&
/
%
–
– .
I
,
%
%( &
– C
– )3 &
%
&
"
* %
–
20
+
E
( %
" +
%
$
&
2
H%
%
%
,
%
,
&
/
%
%
%
"
&
,
, % -
!
,
&/
,
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
.
%
"
%
"
– 2
%
!
I
%
"
%
"
%%
"
%
/
!
#
%
% %
,
(
!
$
.
+
I
,
%
7 .
0
%
"
/
C&
% %
%
-
C&
(
1
%
–
%
+
.!
-
%
H
%
&
%
&
,
%
3
(
!
+
.,
&
%,
21
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
)$ $
&
%
*(
$
$
%"
JJ +
$
#
&
'
(((
!
#
"
$
!
$
#
)
$
'
*
$
!
$
%
&
$*$.
+
$
.
#
#
$ #
!
2
%
$
22
, 21K
%
&
%
&$
%
!
J. Moreno / December 2004
%
, ! %
3 '
%
!
"
'
@
:
"<
:
:
=
© 2004 IBM Corporation
IBM Research
2
%
– $
– 3
% "
%
%
-
&
,
I
2
%
%
%
%
– $
%,
J. Moreno / December 2004
&
/
– $
C
23
!&
%
%
"
/ %
© 2004 IBM Corporation
IBM Research
'%
&, ! %
–
,
"/
'
%
– 1
%
"
"
%
&,
% ,
%
1/
%
24
J. Moreno / December 2004
!
© 2004 IBM Corporation
IBM Research
2
%
#% ,
$
F
2
" H
% &
" %
&
&
%
" %
"%
/&
–2
$-
– .&
%%
.
(
"
25
,&,%
J. Moreno / December 2004
"
" %
&
"
%
+
&
&
C
.
<
&
6
,%
%
%,
"
%
%
"
"
"
,
© 2004 IBM Corporation
IBM Research
3
26
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
2
27
J. Moreno / December 2004
© 2004 IBM Corporation
IBM Research
&
'
!
5 !
1 3
!
3
1
-
2
*
(
*
)
-
1
!
'
1
(*
/
)
!
0
4 &
"
1
!
(*
!
6
!%
(*
! %
0
(
28
J. Moreno / December 2004
0
6
0
1
(
© 2004 IBM Corporation
IBM Research
'%
&, ! %
–
,
"/
'
%
– #
&
% ,
"
%
%
&,
"
%
1/
–
%
%
29
J. Moreno / December 2004
!
© 2004 IBM Corporation
IBM Research
%
!
– )
%*
– 2
–
–
%
%
&
%
%
&
"
,
%
30
%*
%
%
,
&
3 5
'
%
)
%
(
% %
" ! " +
D % "
, &
"
2
"
J. Moreno / December 2004
&
"
%
"
" F
© 2004 IBM Corporation
IBM Research
.
$
– 1,
.
%
&
&
$
6 !
# 0
%
– 1
/
31
%
J. Moreno / December 2004
&
"
&
-
&
,
,
%
© 2004 IBM Corporation
IBM Research
Dec 2004
© 2004 IBM Corporation
IBM Research
3,
,
"
%
%
"
! %
%%
&
%
%
%
%
%
,
,
%
% ,
0
"
,&
"
,
%
"
!
/
%
L
"
"
&
33
J. Moreno / December 2004
© 2004 IBM Corporation