Intel Designs vs the PowerPC Family


NOTE: A seldom-quantified chip characteristic is a design's IPC number, which together with clock rate, enables one to make cross-family comparisons. For this chart we measure IPC by SPEC2000 number/clock rate. It should be clear that
performance = clock_rate * efficiency
          that is,
SPEC2000 = clock_MHz * IPC
Obviously, focusing on the clock rate alone, to the exclusion of the IPC -- as the naive computer user from his ignorance is wont to do -- is misleading. The fact that the word "speed" is used as a synonym for both "clock rate" and "performance", further compounds the error. Phrases like "faster CPU" suffer from the same ambiguity.

Tests for dual-core chips use only one core. SIMD and AltiVec units are not used in the tests.


Last update: 08 Dec 2009
CLOCK
PERFORMANCE
CPU chip clock
(MHz)
mill.
trans.
process
(µm)
size
(mm2)
Dhry-
stone
MIPS
SPEC 95 SPEC 2000 IPC
(SPEC 2000/MHz)
GFlops
per core in
clusters,
dbl. prec.
PS7
Bench
power
dissipation
watts
pricing &
availability
pipeline front-side bus cache,
on-chip (off-chip)
ALU units
units/ins-per-clock/bits
recent
Apple, etc.
product
int int fp int fp int fp real
(theor.)
real/
theor
norm.
score
typ. max. US $ at length simul
instr.
lines MHz thru put/sec L1 L2 L3 int fp SIMD*
O    l    d         I    n    t    e    l                    

Pentium MMX (Intel)

 

 

 

 

 

200

4.5

0.35

141

-

6.4

---

- - -

- - -

?

?

- - -

7.3

15.7

95

6/98

200 mob.

0.25

95

3.4

5.0

230

4/98

233

0.35

141

-

7.0

4.5

- - -

- - -

?

?

7.9

17.0

106

6/98

233 mob.

0.25

95

3.9

5.5

359

4/98

266 mob.

-

---

---

- - -

- - -

?

?

5.3

7.6

466

Celeron (Intel)

 

266

7.5

0.25

131

-

---

---

- - -

- - -

?

?

- - -

?

16.9

106

6/98

300

?

?

159

Pentium II (Intel)

 

233

7.5

0.35

203

-

9.4

6.7

- - -

- - -

?

?

- - -

?

34.8

161

6/98

233 mob.

0.25

131

7.5

10.6

466

4/98

266

-

10.7

7.5

- - -

- - -

?

?

9.8

19.5

198

6/98

266 mob.

8.6

12.1

696

4/98

300

-

11.9

8.1

- - -

- - -

?

?

13.0

?

305

6/98

333

-

13.0

8.8

- - -

- - -

?

?

15.3

23.7

412

350

-

13.9

10.2

- - -

- - -

?

?

?

24.5

519

400

-

15.8

11.4

- - -

- - -

?

?

?

27.9

722

Pentium II Xeon (Intel)

 

400

7.5

0.25

131

-

16.3

12.1

- - -

- - -

?

?

- - -

?

38.1

2800

7/98

CLOCK
PERFORMANCE
CPU chip clock
(MHz)
mill.
trans.
process
(µm)
size
(mm2)
Dhry-
stone
MIPS
SPEC 95 SPEC 2000 IPC
(SPEC 2000/MHz)
GFlops
per core in
clusters,
dbl. prec.
PS7
Bench
power
dissipation
watts
pricing &
availability
pipeline front-side bus cache,
on-chip (off-chip)
ALU units
units/ins-per-clock/bits
recent
Apple, etc.
product
int int fp int fp int fp real
(theor.)
real/
theor
norm.
score
typ. max. US $ at length simul
instr.
lines MHz thru put/sec L1 L2 L3 int fp SIMD*
G    3         a    n    d        G    4            

PPC 603e (IBM, Motorola)

 

 

 

 

 

200

2.6

0.35

80

283

5.6

4.9

- - -

- - -

?

? - - -

4.0

5.0

-

4/98

0.29

43

2.5

4.0

98

300

424

7.4

6.1

- - -

- - -

?

?

4.0

6.0

160

PPC 604e (IBM, Motorola)

 

250

5.1

0.25

47

??

11.1

7.8

- - -

- - -

?

?

- - -

6.0

10.6

295

4/98

375

??

15.6

9.7

- - -

- - -

?

?

8.0

14.5

645

9/98

PPC G3 (IBM, Motorola)

 

750

233

6.4

0.25

67

427

11.0

8.1

- - -

- - -

?

?

- - -

5.6

8.8

$495

4/98

4 ? 64 66? - 32K (1M) - 1/1/32 1/1/32 none

500

0.22

40

1160

23.8

14.5

- - -

- - -

?

?

6.0

-

?

9/98

PPC G3 (IBM)

 

 

 

 

 

750CX

550

21.5

0.18

42

??

---

---

- - -

- - -

?

?

- - -

5.5

5.5

?

11/00

4 ? ? 100 - 64K 256K ? 1/1/32 1/1/32 none

750CXe

600

20

43

1392

26

16

- - -

- - -

?

?

70

6.0

-

?

7/01

133 - ?

iBook
iMac

750FX
SOI, Lo-K

900

38

0.13

37

2088

40

22

- - -

- - -

?

?

95

6.1

-

?

5/02

5 ? 64 200 - 64K 512K - 2/2/32 1/1/32

iBook,5/02

750GX
SOI, Lo-K

1000

?

52

2320

52

30

469

- - -

?

?

-

8.3

-

?

3/04

- 1024K

750VX
SOI, Lo-K, AltiVec

1800

?

0.09

?

??

---

---

- - -

- - -

?

?

?

?

?

Q2/04
cancelled ?

5+ 400 (= 200x2) - (4 MB) 4/2/128

( iBook?? )

PPC G4 (Motorola) AltiVec, SMP, non-NUMA

 

 

 

 

 

7400

450

10.5

0.20

83

825

---

---

- - -

- - -

?

?

?? GF
(0.45 GF)
- 101

8.0

8.0

?

7/99

4 ? 64 100 - 64K (2M) - 2/3/32 1/1/64 2/1/128

Cube

7410, Nitro

667

10.5

0.18

83

1223

---

---

- - -

- - -

?

?

- - -

6.3

6.3

?

1/01

old TiBook

7440

700

33?

83?

1264

---

---

- - -

- - -

?

?

8.0

8.0

?

10/01

7 16 64 133 1.1 GB
(dual: 1.1)
64K 256K - 4/4/32 1/1/64 4/2/128

TiBook 667
iMac2

7450, G4e

867

33

106

1566

---

---

- - -

- - -

?

?

14+

17+

$435

7/01

(2M),
1/4-CPU

PM 733,
800, 867

7455, G4
Apollo 6, SOI

1000

33

106

2280

---

---

306

187

0.31

0.19

- - 267
(dual)

21

30

$295

1/02

167 1.3 GB
(dual: 1.3)
(2M)DDR
1/4-CPU

PM 800,
933, 1000

7455, G4
Apollo 6, SOI, Lo-K

1420

3192

---

---

418
(or 560?)

248

0.30 - 0.40

0.19

338
(dual)

30

42

$475

1/03

PM 1400

7457, G4
Apollo 7, SOI, Lo-K

1333

58

0.13

98

??

---

---

- - -

- - -

?

?

?? GF
(1.33 GF)
- 101

14

?

$189

9/03

133 1.1 GB
(dual: 1.1)
512K (2M)DDR
1/4-CPU

AlBook
iBook(7447)

7470

1500

?

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

For 7/2002,
cancelled

7 ? 64 266 (= 133x2) - ? 512K (4M)DDR
1/4-CPU

7457-RM, G4
Apollo, SOI

2000

?

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

For 2004,
cancelled

7 ? 16 (RapidIO) 333 (= 167x2)
RapidIO@ 500 MHz
- 64K 512K -

7448, G4
SOI, Lo-K

1700

?

0.09

?

3910

---

---

- - -

- - -

?

?

- - -

21

30

?

03/06

7 ? 64 200 - 64K 1024K -

8641, G4
1 core
SOI, Lo-K

1500

?

?

3450

---

---

- - -

- - -

?

?

- - -

?

?

?

H2/05

7 ? 128 (mem)
16 (RapidIO)
667 (= 333x2)
RapidIO@ 500 MHz
mem controller
on chip, ECC
- 64K 1024K -

8641D, G4
2 cores
SOI, Lo-K

1500
2000

?

?

3450
?

---

---

- - -

- - -

?

?

- - -

15

25

?

H2/05

64K
* 2
1024K
* 2

PPC 7500 (Motorola) - [ "Motorola source" reveals 7470, and 7500, The Register Feb 11, 2002 ]

 

 

 

 

 

7500

?

?

0.13

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

For 2/2003,
cancelled

14 ? 128 (mem)
16 (RapidIO)
266 (= 133x2)
RapidIO@ 500 MHz
- ? 512K - 4/4/32 1/1/64 4/2/128

PPC G5 (Motorola) SOI, SMP, NUMA - [ 8500 data from MacOSRumors & The Register, so use with utmost caution. Anyway, these optimistic rumors of late 2001 never materialized. 8540 data from Motorola ]

 

 

G5 - 8500

1600

58

0.13

192

??

---

---

1342

1364

0.84

0.85

- - -

30

30

$700

For 2/2002,
cancelled

10 ? 128 400 (= 100x4)
mem controller
on chip
- 128K 512K (2-8M) ?/?/64 ?/?/64 8/4/128

(future)
PowerMac
G5,
cancelled

2000

??

---

---

1678

1705

?

?

?

Cancelled

2400

??

---

---

2013

2046

?

?

?

8540
integrated

800

65?

?

1852

---

---

- - -

- - -

?

?

- - -

6.5

6.5

$200

"soon"

? ? 64 333 (= 167x2)
mem controller
on chip
- 32K 256K - ?/?/32 none none
CLOCK
PERFORMANCE
CPU chip clock
(MHz)
mill.
trans.
process
(µm)
size
(mm2)
Dhry-
stone
MIPS
SPEC 95 SPEC 2000 IPC
(SPEC 2000/MHz)
GFlops
per core in
clusters,
dbl. prec.
PS7
Bench
power
dissipation
watts
pricing &
availability
pipeline front-side bus cache,
on-chip (off-chip)
ALU units
units/ins-per-clock/bits
recent
Apple, etc.
product
int int fp int fp int fp real
(theor.)
real/
theor
norm.
score
typ. max. US $ at length simul
instr.
lines MHz thru put/sec L1 L2 L3 int fp SIMD*
P    o    w    e    r    4         a    n    d        P    o    w    e    r    5            

Power3 (IBM) SMP, NUMA

 

 

 

 

 

Power3

375

?

?

?

-

---

---

- - -

- - -

?

?

1.10 GF
(1.50 GF)
0.73 -

?

?

?

ca. 1998

12 ? ? ? - ? ? ? ?/?/64
2/2/64,
2/4/64
none

Power4 (IBM) SOI, Lo-K, SMP, NUMA

 

 

 

 

 

Power4
2 cores

1300

174

0.18

414

??

---

---

839

1266

0.65

0.97

2.7 GF
(5.2 GF)
0.51 -

150

150

$5,000

12/01

14 200 similar similar similar 96K 1440K
/2
(32M) 2/?/64
2/2/64
or
2/4/64
(FMA)
none

Power4+
2 cores

1700

184

0.13

267

??

---

---

1113

1699

0.65

1.00

3.7 GF
(6.8 GF)
0.55 -

60

60

?

5/03

mem: 256
FSB: 64
next CPU: 128
 
 
mem: 567 MHz
FSB: 567 MHz
next CPU: 850 MHz
 
mem: 18.1 GB
FSB: 4.5 GB
next CPU: 13.6 GB
TOTAL=36.2 GB / 2
(dual,4 cores:72.4 GB)
(128M)

PowerPC 970 (IBM) [Power4 derivative], non-NUMA (GPUL & GPUL2)

 

 

 

 

 

G5
IBM-970

2000

56

0.13

118

6444

---

---

1041

1168

0.52

0.58

5.0 GF
(8.0 GF)
0.62 562
(dual)

66

97

$180

8/03

16 (int)
21 (fp)
19 (Alti-Vec)
215 32 in +
32 out
1000 (= 500x2) 7.1 GB
(dual: 14.2)
96K 512K none 2/4/64
2/2/64
or
2/4/64
(FMA)
4/4/128

PowerMac
dual G5

G5
IBM-970FX
SSOI

1600

0.09

66

-

---

---

- - -

- - -

0.43

0.58

- - -

17

29

?

-/05

?? ??

2500

7540

---

---

1082

1361

?? GF
(10.0 GF)
- 714
(dual)

50

93

$180

7/04

1250 (= 625x2) 8.9 GB
(dual: 17.8)

PowerMac,
iMac

3000

8700

---

---

- - -

- - -

?? GF
(12.0 GF)
- -

60

83

?

9/04
delayed

1500 (= 750x2) 10.7 GB
(dual: 21.4)

IBM-970MP
Antares
2 cores

2500

?

154

-

---

---

1438

2076

0.58

0.83

?? GF
(12.0 GF)
- 860
(4cores)

40

100

?

10/05

1250 (= 625x2) 8.9 GB
(dual, 4 cores: 17.8)
96K
* 2
1024K
* 2
4/4/128
each core

PowerMac
quad G5

Power5 (IBM) SOI, Lo-K, SMP, NUMA, SMT

 

 

 

 

 

Power5
2 cores

1900

276

0.13

389

??

---

---

1452

2702

0.91

1.42

4.3 GF
(7.6 GF)
0.56 -

160

?

?

6/04

14 ? ? mem controller
on chip
similar 96K 1920K
/2
(36M
/2)
1/4-CPU
2/4/64
2/2/64
or
2/4/64
(FMA)
none

2000

1820

2844

- - -

?

?

?

??

mem: 25 GB
FSB: 6 GB
next CPU: 64 GB
TOTAL = 95 GB / 2
(dual, 4 cores: 190 GB)

Power5+
2 cores

1900

0.09

251

??

---

---

?

3007

?

1.58

- - -

?

?

?

7/06

similar

2200

?

?

- - -

?

?

?

7/06

PowerPC 975-976 (IBM) [Power5 derivative], SMT (GRUL)

 

 

 

 

 

G6
IBM-975

3000

98

0.09

?

??

---

---

1650

1750

0.55

0.58

- - -

63

?

?

9/04
delayed

? ? ? mem controller
& HyperTransport
on chip
- 96K 1024K none 2/4/64
2/2/64
or
2/4/64
(FMA)
4/4/128

3400

??

---

---

1870

1972

?? GF
(13.6 GF)
- -

86

86

?

IBM-976 (or 980)
2 cores

4000

?

0.065

?

??

---

---

2200

2320

0.55

0.58

- - -

?

?

?

8/05
plan

? ? ? - 96K ? AltiVec 2

Power6 (IBM) SOI, Lo-K, AltiVec 2, SMP, NUMA, SMT, BCD

 

 

 

 

 

Power6
2 cores

5000

700

0.065

341

??

---

---

3000

5200

0.75

1.30

12 GF
(16.0 GF)
0.75 -

130

130

?

5/07

? ? ? mem controller
on chip
300 GB ? 4096K
* 2
(32M
/2)
? ? AltiVec (VMX)

PowerPC 980 (IBM) [Power6 derivative], (P6UL)

 

 

 

 

 

IBM-980 (or 990)

6000

?

0.065

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

2007

? ? ? ? - ? ? ? - ? AltiVec 2 ?

(6 GHz seems unlikely at 0.065)

Cell Series (Sony/Toshiba/IBM) SOI, Lo-K

 

 

 

 

 

Cell variant
3 PPE cores
SMT, water-cooled

3200

165

0.09

168

??

---

---

- - -

- - -

?

?

?? GF
(12.8 x 3 = 38 GF)
- -

?

?

?

11/05

? ? ? mem
GPU
I/O
 
mem: 22 GB
GPU & I/O: 22 GB
TOTAL = 44 GB
32K
* 3
1024K
/ 3
none 2/2/64
2/4/64
(FMA)
AltiVec 2

SP: 25.6 GF
DP:  12.8 GF
x 3 = 115 GF

Microsoft
Xbox 360

Cell
4 cores

4600

?

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

12/05

? ? ? ? 6.4 GB (per core?) ? ?
1/2/64
(FMA)
? Sony/IBM
workstation

Cell
Single Power core
w. SMT & AltiVec (PPE)
+ 8 SIMD cores (SPE)

"Broadband Processor Architecture"

3200

234

235

??

---

---

- - -

- - -

?

?

?? GF
(12.8 GF)
- -

?

?

?

7/06

21
(PPE int)
? mem
GPU
I/O
 
mem controller
on chip


mem
GPU
I/O
 
mem: 25 GB
GPU & I/O: 77 GB
TOTAL = 102 GB
64K

256K x 8
512K 4/4/128

4/4/128 x 8
Sony
PlayStation3
---
home server
& HDTV
(2007)

4600

?

??

---

---

- - -

- - -

?

?

?? GF
(18.4 GF)
- -

70

80

?

?

?
   Cell total
SP: 256 GF
DP:   18 GF

Cell variant ?
"Broadway"

?

?

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

9/06

? ? ? ? ? ? ? ? ? ?

Nintendo
Wii

Cell

?

?

0.065

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

3/07

? ? ? ? ? ? ? ? ? ?

Cell

?

?

0.045

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

? 2008

? ? ? ? ? ? ? ? ? ?

Cell

?

?

0.032

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

? 2011

? ? ? ? ? ? ? ? ? ?

PPC 400 Series (IBM)

 

 

 

 

 

PPC-440

700

?

?

?

??

---

---

- - -

- - -

?

?

?? GF
(5.6 GF)
- -

5

8

?

2004

7 ? ? ? - 64K 2048K - 2/2/32 2/4/64
(FMA)
?

PPC-440
2 cores for Blue Gene

700

?

?

?

??

---

---

- - -

- - -

?

?

- - -

10

15

?

2004

4M
/2
Blue Gene/L

PPC-450

1000

?

?

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

2006

7 ? ? ? - 64K 2048K - 2/2/32 2/4/64
(FMA)
? Blue Gene/P

Power7 (IBM) SOI, Lo-K, AltiVec 2, SMP, NUMA, SMT

 

 

 

 

 

Power7
2 cores

9000

?

0.045

?

??

---

---

- - -

- - -

?

?

- - -

?

?

?

2008

? ? ? ? - ? ? ? ? ? ?

TRIPS chip - DARPA-funded (IBM, U. Texas), sources: IBM (Aug 03). IEEE (Jan 05).

 

 

 

 

 

Prototype
2 cores

500

250

0.13

?

16 ops 
*MHz*cores
= 16,000 
MIPS / chip

---

---

ca 1.3K
per core

ca 1.3K
per core

ca 2.66

ca 2.66

?? GF
(6.3 GF)
- -

?

?

?

12/05

? ? ? ? - ? ? ? -/16/64 -/16/64 ?

Full scale
8 cores

10000

?

??

?

1,000,000 
MIPS / chip

---

---

ca 20K
per core

ca 20K
per core

ca. 2.00

ca. 2.00

?? GF
(125 GF)
- -

?

?

?

2012

? ? ? ? - ? ? ? ? ? ?
CLOCK
PERFORMANCE
CPU chip clock
(MHz)
mill.
trans.
process
(µm)
size
(mm2)
Dhry-
stone
MIPS
SPEC 95 SPEC 2000 IPC
(SPEC 2000/MHz)
GFlops
per core in
clusters,
dbl. prec.
PS7
Bench
power
dissipation
watts
pricing &
availability
pipeline front-side bus cache,
on-chip (off-chip)
ALU units
units/ins-per-clock/bits
recent
Apple, etc.
product
int int fp int fp int fp real
(theor.)
real/
theor
norm.
score
typ. max. US $ at length simul
instr.
lines MHz thru put/sec L1 L2 L3 int fp SIMD*
I    n    t    e    l         a    n    d        A    M    D            

Pentium 3 (Intel) SMP, non-NUMA

 

 

 

 

 

 

1000

?

?

?

-

---

---

420/236

- - -

0.42

?

? - 132

?

?

?

?/98

12 ? ? ? - ? ? ? ?/?/32 ?/?/32 none

Pentium 4, 5, 6 (Intel) no SMP

 

 

 

 

 

2000

42

0.18

217

-

---

---

656

714

0.33

0.36

- - -

70?

70?

$401

9/01

20 126 64 400 (= 200x2) 3.2 GB 20K 256K none 3/5/32 2/2/64 MMX,SSE,SSE2
Pentium 4C
Northwood,
SMT

2200

55

0.13

132

-

---

---

811

802

0.37

0.36

- - -

?

?

$562

1/02

20 126 400 (= 200x2) 3.2 GB 20K 512K none 3/5/32 2/2/64

2800

-

---

---

1010

947

0.36

0.34

2.3 GF
(5.6 GF)
0.41 -

?

68

$218

10/02

3200

-

---

---

1261

1285

0.39

0.38

- - 427

82

?

$417

6/03

800 (= 200x4) 6.4 GB
Pentium 4EE

3200

178

237

-

---

---

1387

1414

0.43

0.44

- - -

92

124

$957

2/04

2M
Pentium 4E
Prescott-32
SS, Lo-K, 1-way

3400

125

0.09

112

-

---

---

1404

1447

0.41

0.43

?? GF
(6.8 GF)
- -

109

?

$417

5/04

31

128 28K 1024K ? SSE3
Pentium CT
Prescott-64

3600

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

-

8/04

31

128 800 (= 200x4) 6.4 GB 44K 1024K 2M 3/5/64 2/2/64 Apple's "Developer Transition Kit"

3800

-

---

---

- - -

- - -

?

?

- - -

?

?

$851

1/05

Dothan
(Pentium-M), SS

1000
2000

140

84

-

---

---

630
1261

642
1285

0.63

0.64

- - -

?
21

?
?

-

5/04

14

? 350 (= 175x2)
400 (= 200x2)
- ? 2048K none 3/5/32 2/2/64 1000GHz: AppleTV (2007)
Tejas
Pentium 5

5-7 GHz

?

130

-

---

---

- - -

- - -

?

?

- - -

125

?

-

For Q2/05,
cancelled

?

128 1066 (= 266x4) - 40K 2048K yes 3/5/64 2/2/64
Nehalem
Pentium 6, SS, Lo-K

10-20 GHz

?

0.065

?

-

---

---

- - -

- - -

?

?

- - -

?

?

-

For 2006,
cancelled

?

? 1200 (= 300x4) - ? ? ? ? ?

Pentium 4 (Intel) Dual-core chips "Pentium D" (Netburst architecture)

 

 

 

 

 

Smithfield EE 840
2 Prescott cores
SMT

3200

230

0.09

206

-

---

---

- - -

- - -

?

?

?? GF
(6.4 GF)
* 2
- -

95

125

$1,000

7/05

31 128 64 800 (= 200x4) 6.4 GB
/ 2
20K 1024K
* 2
none 3/5/64 2/2/64 MMX,SSE,SSE2
Smithfield 840
2 Prescott cores
no SMT

230

206

-

---

---

- - -

- - -

?

?

95

125

$530

7/05

Paxville 7000MP
2 Prescott cores
(for servers), SMT
SMP 4-way

3000

?

?

-

---

---

- - -

- - -

?

?

- - -

135

165

?

10/05

2 mem controllers
on chip
2048K
* 2
Dempsey
(for servers), SMT, VT
[Bensley platform]

3460
3730

?

0.065

?

-

---

---

- - -
1800

- - -
- - -

?
0.48

?

?? GF
(6.9 GF)
- -

130

130

$3,700

5/06

64
* 2
666 (= 333x2) 8.6 GB
* 2
1024K
* 2
8M
Cedar Mill 672
2 cores on 2 dies
(or just 1 core?)
in 1 package, no SMT

3800

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

$600

11/05

64 800 (= 200x4) - 1024K
* 2
none
Presler 945
(Smithfield dieshrink)
2 cores on 2 dies
in 1 package, SMT

3400

376

162

-

---

---

- - -

- - -

?

?

- - -

?

?

$163

1/06

- 2048K
* 2

Server - Pentium 4 (Intel) SMP, non-NUMA, SMT

 

 

 

 

 

Xeon

2400

55+

0.13

131+

-

---

---

- - -

- - -

0.36

0.34

3.4 GF
(4.8 GF)
0.71 -

?

?

?

??

20 126 64 ? - 20K 512K none 3/5/32 2/2/64 MMX,SSE,SSE2

3060

-

---

---

1138

1103

0.37

0.36

?? GF
(6.1 GF)
- 488
(dual)

60

?

$455

3/03

533 (= 133x4) 4.3 GB
(dual: 4.3)

?

?

-

---

---

1294

1186

0.42

0.39

- - -

?

103

$690

7/03

1M
Xeon-64 Nocona
(ca. Prescott)
SMP 2-way
3600 ?

0.09

?

-

---

---

- - -

- - -

?

?

?? GF
(7.2 GF)
- -

?

?

$850

8/04 31 126 800 (= 200x4) 6.4 GB
(dual: 6.4)
40K 1024K ? 3/5/64 SSE3
Xeon-64 Potomac
SMP 4-way
3330 ? ?

-

---

---

- - -

- - -

?

?

- - -

?

?

$3,692

3/05

31 126 666 (= 333x2) - 1024K 8M
Tulsa
7140M
2 Prescot cores
(for 4-socket servers)
Pellston virtualization
3400 1300 0.065 ?

-

---

---

- - -

- - -

?

?

- - -

?

150

$1,940

8/06

31 126 64 800 (= 200x4) 8.5 GB / 2
(dual: 17.0 / 4)
40K 1024K
* 2
16M/2 3/5/64 2/2/64 SSE3

Desktop & Server  Dual-core chips (Dothan & Dothan+ architectures), no SMP (except for servers), no NUMA (except as noted), no SMT (except as noted)

 

 

 

 

Yonah
Core Duo, T2500
2 Dothan cores
(for mobile)
[Napa platform]

2160 152

0.065

?

-

---

---

1754

1580

0.82

0.73

- - -

13

31

$637

1/06 12 ? 64 666 (= 166x4) 5.3 GB 40K 2M/2 none 3/5/32 2/2/64 SSE3 Mac mini,
MacBook,
MacBook Pro,
iMac

Merom
Core 2 Duo,
T---/T7400/T7600
2 Dothan+ cores, 4 issue
(for mobile)
[Napa64 platform]

??
2160
2333

?

0.065

?

-

---

---

- - -

- - -

?

?

- - -

??
20
20

5
34
34

-
$423
$637

1/07
8/06
8/06

14 (int) ? 64 666 (= 166x4) 5.3 GB 64K 4M/2 none ?/?/64 2/2/64 SSE3
(4/6/128)
MacBook,
MacBook Pro,
iMac (sep06)

Conroe
Core 2 Duo, E6600/E6700
2 cores
(for desktop)
[BridgeCreek/Averill platforms]

2400
2666

298

143

44,329

---

---

- - -

- - -

?

?

- - -

?

-
65

$316
$530

7/06
7/06

1066 (= 266x4) ?

Conroe
Core 2 Extreme, X6800
2 cores
(for desktop games)
[BridgeCreek/Averill platforms]

2930
3200

?

?

-

---

---

- - -

- - -

?

?

- - -

?

75
95

$999
-

7/06
Q3/06

1066 (= 266x4)
1333 (= 333x4)
8.5 GB
10.7 GB

Woodcrest
Xeon 5148/5150/5160
2 cores
(for 2-socket servers), SMP
[Glidewell/Bensley platforms]

2333
2666
3000

?

?

-

-

-

- - -
2800
3057

- - -
2500
2783

?
1.05
1.02

?
0.94
0.93

9.6 GF
(12.0 GF)
0.80 -

?

40
65
80

$455
$690
$851

Q3/06
6/06
6/06

64
* 2
1333 (= 333x4)
* 2
10.7 GB
* 2
4M/2 Xserve,
4 core Mac Pro
Kentsfield
Core 2 Quad, Q6600
Core 2 Extreme, QX6700
4 cores, 2 Conroe dies
(for desktop games)

2400
2666

582

?

-

---

---

- - -

- - -

?

?

- - -

?
?

110
130

$851
$999

1/07
11/06

128 64 1066 (= 266x4) 8.5 GB
8M/4 future 4 core iMac
Clovertown
Core 2 Quad, E5335/E5345/X5355/X5365
4 cores, 2 Woodcrest dies
(for 2-socket servers), SMP
[Glidewell/Bensley platforms]
2000
2333
2666
3000
? ?

-

---

---

- - -

- - -

?

?

- - -

?
?
?
?

80
80
120
--

$690
$851
$1,172
--

01/07
12/06
12/06
04/07
128 64
* 2
1333 (= 333x4)
* 2
10.7 GB
* 2
8M/4 8 core Mac Pro
Tigertown
Core 2 Quad, E7340
4 cores, 2 dies
(for 4-socket servers), SMP
[Clarksboro platform]
2400 ? ?

-

---

---

- - -

- - -

?

?

- - -

?

80

$1,980

Q3/07 ? QuickPath interconnect QuickPath interconnect 10.7 GB
* 4
8M/4 ?
Dunnington
8 cores
(for servers)
3000 ? ?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2008 ? QuickPath interconnect QuickPath interconnect ? ? ?
Bloomfield
4 cores, 1 die
Glio (mobile),
Bloomfield (desktop),
Gainestown (server)
4000 ?

0.045

?

-

---

---

- - -

- - -

?

?

- - -

?

130

?

4/08

14 (int) ? QuickPath interconnect 1600 (= 800x2)
on chip mem controller
for DDR3
? 40K ? ? / 4 -/-/64 2/2/64 SSE4
(4/6/128)

Penryn
(Merom-series dieshrink,
but Hafnium dioxide, Hi-K)
- - - - - - - - - - - - -

E5260 2-core package

E5260 2-core package
(Wolfdale, for 1-2 socket servers)

4-core package
(for mobile, Montevina platform)

4-core package
(Yorkfield, for 1-socket servers)

X5460 4-core package
(Harpertown, for 2-socket servers)

6-core package
(Dunnington, for servers)

Diagram
of new
process
technology

2600
3330
xxx
3000
3200
?


410
847
820
820
?


?
?
?
?
?


-


---


---


- - -


- - -


?


?


-

-

-


?
?
?
?
?


65
45
80
120
130


$851
?
?
$1,172
?


2/08
11/07
1/08
1/08
1/08
H2/08

14 (int)
?

?

?? ??
1333 (= 333x4)
1066 (= 266x4)
1066 (= 266x4)
1600 (= 400x4)
1066 (= 266x4)

?
40K
6M/2
?
12M/4
12M/4
3M*3

none
none
none
none
16M/6
?/?/64 2/2/64
MacBook &Pro
-
-
8 core Xserve
8 core Mac Pro
-
???
32 cores
(for servers)
? ? ?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

? ? QuickPath interconnect QuickPath interconnect ? ? ?
Atom, SMT
1 core
2 cores


?
1600
47
?

0.045

?

-

---

---

- - -

- - -

?

?

- - -

0.030
?

2.5
16

?

6/08
7/08

14 (int) ? 64 ? ? 40K ? ? -/-/64 2/2/64 future MacTablet?
Nehalem
Xeon 3500, Xeon 5500
Nehalem microarchitecture
4 cores
off-die GPU in package
SMT, NUMA
[Tylersburg platform]
? 781
?

0.045
0.032

?

-

---

---

- - -

- - -

?

?

- - -

90

130

?

Q3/08
2009

14 (int) ? QuickPath interconnect on chip mem controller ? 256K
512K
1M
2M
8M
16M
-/-/64 2/2/64
•• Nehalem
Xeon 5000 series
Westmere microarchitecture
4 cores (for 2 & 4 socket servers)
AES encryption
SMT, NUMA
? ?

0.032

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

Q1/10

14 (int) ? 64 on chip mem controller ? yes yes 24M -/-/64 2/2/64
Nehalem-EX
Nehalem microarchitecture
8 cores
SMT, NUMA
? 2300

0.045

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

H2/09

14 (int) ? 64 6.4 GB/s QPI ? yes yes 24M -/-/64 2/2/64
Gesher microarchitecture
on-die GPU
? ?

0.032
0.022

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2010
2011

14 (int) ? 64 ? ? 40K ? ? -/-/64 2/2/64
Yorkfield
8 cores
? ?

0.045

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

Q4/08

14 (int) ? 64 ? ? 40K 12M/2 ? -/-/64 2/2/64
Gainstown
16 cores
? ?

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2009

14 (int) ? 64 ? ? 40K ? ? -/-/64 2/2/64
Gulftown
Kiefer design
32 cores
(8 nodes x 4 cores ea.)
? ?

0.032

25.6

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2010

14 (int) ? 64 ? ? 40K 0.5M
* 8
3M
* 8
-/-/64 2/2/64

IA-64 (Intel) SMP, non-NUMA??

 

 

 

 

 

Itanium

800

25

0.18

?

-

---

---

380

701

0.48

0.88

- - -

130

130

$2,000

5/01

10 ? 64 266 (= 133x2) 2.1 GB 48K 96K (2-4M) 4/?/64 2/2/64 MMX,SSE,SSE2

McKinley
(Itanium 2 3M)

1000

221

464

-

---

---

807

1356

0.80

1.36

3.2 GF
(4.0 GF)
0.79 -

100

130

$4,227

6/02

8 ? 128 400 (= 200x2) 6.4 GB 32K 256K 3M 6/?/64 2/4/64

Madison
(Itanium 2 6M)

1500

410

0.13

374

-

---

---

1322

2119

0.80

1.36

4.5 GF
(6.0 GF)
0.74 -

-

130

$4,227

7/03

8 ? ? 400 (= 200x2) 6.4 GB
(dual: 6.4)
32K 256K 6M ?/?/64 2/4/64

LV Itanium

1000

?

?

-

---

---

- - -

- - -

?

?

- - -

55

62

$744

9/03

? ? ? ? - 32K 256K 1.5M ?/?/64 ?/?/64

Deerfield, LC

1400

?

?

-

---

---

- - -

- - -

?

?

- - -

55

62

$1,172

9/03

? ? ? ? - 32K 256K 1.5M ?/?/64 ?/?/64

Madison 2
(Itanium 2 9M)

1600

500

480

-

---

---

1590

2712

?

?

?? GF
(6.4 GF)
- -

-

150

$4,200

11/04

? ? ? 533 (= 133x4) 8.5 GB
(dual: 8.5)
32K 256K 9M ?/?/64 2/4/64

Montecito 9000
SMT, VT, 2 cores

1600

1720

0.09

580

-

---

---

- - -

- - -

?

?

- - -

-

104

$3,700

7/06

? ? 128 800 (= 200x4) 12.8 GB
(dual: 25.6)
? ? 24M/2 ?/?/64 ?/?/64

Montvale
improved Montecito
SMT, 2 cores

2400

?

0.065

?

-

---

---

2400

4300

?

?

8.8 GF
(9.6 GF)
0.90 -

-

130

?

Q4/07

? ? 128 1066 (= 266x4) 16.7 GB
(dual: 33.4)
? ? 24M/2

Tukwila
SMT, 4 cores

2400

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2008

? ? QuickPath interconnect QuickPath interconnect

mem controller
on chip
- ? ? ?

Tukwila
SMT, 8 cores

?

?

?

-

---

---

1780 1780

?

?

- - -

?

?

?

For 2007,
cancelled

? ? - ? ? ?

Tukwila
SMT, 16 cores

?

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

2008

? ? - ? ? ?

Athlon XP (AMD) SMP, non-NUMA

 

 

 

 

 

Athlon XP

1600

46

0.18

150

-

---

---

684

611

0.43

0.38

1.6 GF
(3.2 GF)
0.50 -

80

80

$270

11/01

10 ? 64 266 - 128K 256K - 3/3/32 3/2/64 yes

Thunderbird

1666

0.13

80

-

---

---

655

- - -

0.39

?

- - 211

?

?

?

5/02

Barton 3200+

2200

>46

>80

-

---

---

1044

873

0.47

0.40

- - 332

60

77

$464

5/03

512K

Opteron (AMD) SOI, NUMA

 

 

 

 

 

Athlon64, 3200+

2000

106

0.13

193

-

---

---

1202

1170

0.60

0.58

- - 356

?

?

$399

9/03

12 (int)
17 (fp)
? FSB=
HyperTransport
mem=64lines
800 (= 400x2)
mem controller
on chip
HT: 6.4 GB
mem: 5.3 GB
TOTAL 11.7 (no duals)
128K 1024K - 3/6/64 3/2/64 yes

Athlon64, FX-51

2200

-

---

---

1322

1287

0.60

0.58

- - 416

?

?

$637

10/03

FSB=
HyperTransport
mem=128lines
3 CPU-BUSES:
mem: 5.3 GB
FSB(HT): 3.2 GB
nextCPU(HT): 6.4 GB

TOTAL 14.9
(dual 29.8) NUMA

Athlon64, FX-57

2800

-

---

---

1680

1624

0.60

0.58

- -

?

?

$1,031

6/05

Athlon64, FX-62
2 cores

2800

20,405

---

---

1837

2256

0.67

0.81

- -

?

125

$1,000

6/06

Opteron
64-bit
SMP, NUMA

1800
[Mod244]

120

-

---

---

1170

1219

0.60

0.58

?? GF
(3.6 GF)
- 332

-

80

?

4/03

2000
[Mod246]

-

---

---

1317

1293

0.66

0.65

2.9 GF
(4.0 GF)
0.71 -

-

89

$794

8/03

2000
[Mod??]

0.09

?

45

45

?

8/04

Athlon 64, X2
5000+
SMT, 2 cores

2600

154

183

20,630

---

---

- - -

- - -

?

?

?? GF
(4.8 GF)
- -
(sing.)

?

95

$403

6/05

FSB=
HyperTransport
mem=128lines
? ? 1024K
/ 2

Opteron 875
2 cores

2200

?

?

-

---

---

- - -

- - -

?

?

?? GF
(4.4 GF)
- -
(sing.)

?

95

$2,650

4/05

Athlon64, FX-70
used for dual socket
Quad FX-74

2 cores each

3000

?

?

-

---

---

- - -

- - -

?

?

- - -

?

125

$999/2

Q3/07

? ? ? ? ? ? ? - ? ? ?

Opteron 2218
2 cores

2600

?

220

-

---

---

- - -

- - -

?

?

- - -

?

?

$2,649

8/06

? ? ? ? ? ? ? - ? ? ?

Opteron Pacifica
2 cores
OS virtualization

?

?

0.065

?

-

---

---

- - -

- - -

?

?

- - -

?

95

?

10/06

? ? ? ? ? ? 512K
* 2
- ? ? ?

Barcelona
(or Deerhound)
Opteron K8L arch.
4 cores

2500

?

0.065

283

-

---

---

- - -

- - -

?

?

- - -

?

120

?

Q3/07

? ? ? ? ? ? 512K
* 4
2 MB
/ 4
? ? ?

Wolfhound
Opteron K8L arch.
4 cores
(for 2-socket servers), SMP

?

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

Q3/08

? ? ? ? ? ? - 2 MB
/ 4
? ? ?

Cerberus
Opteron K8L arch.
4 cores
(for 4-socket servers), SMP

?

?

?

-

---

---

- - -

- - -

?

?

- - -

?

?

?

Q3/08

? ? FSB=
HyperTransport3
mem=? lines
? 3 CPU-BUSES:
mem: ? GB
FSB(HT3): 19.2 GB
next CPU(HT): ? GB

TOTAL ??
(dual ??) NUMA
? - 6 MB
/ 4
? ? ?
CLOCK
PERFORMANCE
CPU chip clock
(MHz)
mill.
trans.
process
(µm)
size
(mm2)
Dhry-
stone
MIPS
SPEC 95 SPEC 2000 IPC
(SPEC 2000/MHz)
GFlops
per core in
clusters,
dbl. prec.
PS7
Bench
power
dissipation
watts
pricing &
availability
pipeline front-side bus cache,
on-chip (off-chip)
ALU units
units/ins-per-clock/bits
recent
Apple, etc.
product
int int fp int fp int fp real
(theor.)
real/
theor
norm.
score
typ. max. US $ at length simul
instr.
lines MHz thru put/sec L1 L2 L3 int fp SIMD*
*SIMD: Vector- and multimedia-instruction units are characterized here by <num computation-units> / <num concur. instructions> / <computation-unit bit-width>

CAUTION IN USING THIS CHART [written Dec. 2001 but applies to whole]:
Since many of the entries in the chart above deal with unreleased products, and in some cases unannounced products, the detailed chip characteristics presented often rely on hearsay and rumor. Therefore consider the data of any given entry more as the starting point for your own investigations rather than authoritative truth. The best source of what a future chip will be, is naturally the manufacturer's own site. Next in line come the professional engineering news sites like eTimes. Lately I have relied heavily on The Inquirer.

SPEC2000 numbers for the G4 are based on Motorola's Dhrystone-figures, plus a bunch of my, perhaps dubious, assumptions, namely: (1.) Dhrystone-21-MIPS and SPEC2000-int measure approximately the same thing, and (2.) the scale factor between the two measures is 5:1, so 1,566 MIPS is the same integer-performance as SPEC2000int 313. (3.) Another assumption (better than nothing) is that if a given chip model has a SPEC2000 score of 700 @ 1-GHz, the chip will have a score of 1400 at 2-GHz, i.e., linear clock-scaling. We all know things don't work this perfectly.

So you will see that the only value of the chart I have posted is to put together in one place all this data (from questionable sources), to make a few (questionable) extrapolations from it, and TO SERVE AS A REFERENCE POINT FOR DISCUSSION. I intend to update the chart as errors and unacceptable values are established, and as the future unfolds.

KINS COLLINS, 2001, 2006


Hilbert Tu - Resume
 home
 the macintosh