Summary
The document contains tables of CPU performance from 80486 to Intel Core 2 Duo and AMD Athlon 64,
including speed via caches and RAM derived from my benchmarks and results reports, which are available
from Compuserve PC Hardware Forum and my web site.
More Historic Data
In analysing results of the different benchmarks, in virtually all cases (except for RAM), CPU speed, including using data in cache, is shown to be proportional to CPU MHz for a particular type of processor. So it has been decided to show the MHz of available CPUs and provide simple tables with one set of numbers for each type of CPU. For results depending on memory speed, results are provided for a range of RAM types.
As the number and type of instructions used in the benchmarks is known, performance results can be easily converted to Millions of Instructions Per Second (MIPS) or Millions of Floating Point Operations Per Second (MFLOPS). Comparison numbers represent percentages of MIPS or MFLOPS divided by CPU MHz.
Contents
1. CPU Performance Tables - Introduction
In analysing results of the different benchmarks, in virtually all
cases, CPU speed, including using data in cache, is shown to be
proportional to CPU MHz for a particular type of processor. So it
has been decided to show the MHz of available CPUs and provide simple
tables with one set of numbers for each type of CPU. As the number
and type of instructions used in the benchmarks is known, performance
results can be easily converted to Millions of Instructions Per
Second (MIPS) or Millions of Floating Point Operations Per Second
(MFLOPS). Comparison numbers represent percentages of MIPS or MFLOPS
divided by CPU MHz.
CPUs can have more than one Integer or Floating Point Pipeline which
each may be able to produce one result per Hz. Also, P4 Integer
Pipelines can execute two instructions per cycle. Thus it is possible
that the percentage calculations can be greater than 100.
Overall processor performance depends on the type/version and speed
of CPU and cache sizes along with memory speed, which is dependent on
the system board used. No attempt is made to provide an overall
performance rating, the overall aim being to show some strengths and
weaknesses of each type of processor.
Note that these benchmarks reflect performance of one CPU on multi-core processors.
Back To Contents List
2. CPU Characteristics
Some of the following have been gathered from benchmark results and magazines
and may not be completely accurate. Mobile CPUs have the same speeds
using mains power but might have diferent size L2 caches and slower RAM.
The CPUID Model Code is given, indicating that different processor models have the same core CPU.
Back To Contents List
Intel
Older CPUs
CPU MHz MHz KB L1 KB L2 Type L2 FSB CPUID
From To Cache Cache Cache MHz Model
Pentium 75 200 8 256-512 External 50-66 51x-52x
Pentium Pro 180 200 8 256-512 Internal 60-66 61x
Pentium MMX 166 233 16 256-512 External 66 58x
Pentium II 233 333 16 512 Half Speed 66 63x-65x
Pentium II 350 450 16 512 Half Speed 100 65x
PII Xeon 400 450 16 512 Full Speed 100 652,653
PII Xeon 400 450 16 1024 Full Speed 100 652,653
PII Xeon 450 16 2048 Full Speed 100 653
Pentium III 450 600 16 512 Half Speed 100 67x
Pentium IIIB 533 600 16 512 Half Speed 133 67x
Pentium IIIE 550 1100 16 256 Full Speed 100 68x
Pentium IIIEB 533 1330 16 256 Full Speed 133 68x
Pentium IIIT 1000 1400 16 256 Full Speed 133 6Bx
Pentium IIIT 800 1400 16 512 Full Speed 133 6Bx
PIII Xeon 500 550 16 512-2048 Full Speed 100 672,673
PIII Xeon 600 1000 16 256 Full Speed 133 68x
PIII Xeon 700 16 1024 Full Speed 100 6A0
PIII Xeon 700 16 2048 Full Speed 100 6A0,6A1
Celeron 266 300 16 0 None 66 65x
Celeron A 300 533 16 128 Full Speed 66 66x
Celeron 2 533 766 16 128 Full Speed 66 68x
Celeron 3 800 1100 16 128 Full Speed 100 68x
Celeron 4T 1000 1400 16 256 Full Speed 100 6Bx
Celeron M 266 466 16 128 Full Speed 66 66x
Celeron M 667 16 128 Full Speed 66 68x
Celeron M 400 900 16 128 Full Speed 100 68x
Celeron M 733 933 16 128 Full Speed 133 68x
Celeron M 650 16 256 Full Speed 100 6Bx
Celeron M 733 1333 16 256 Full Speed 133 6Bx
M = Mobile
Back To Contents List
All the following have full speed L2 cache
Pentium 4 Class
CPU MHz MHz KB L1 KB L2 FSB CPUID
From To Cache Cache MHz Model
Pentium 4 1300 2000 8 256 400 F0x-F1x
Pentium 4N 1600 2800 8 512 400 F2x
Pentium 4N2 2260 3060 8 512 533 F2x
Pentium 4N3 2400 3400 8 512 800 F2x
Pentium 4EE 3200 3400 8 512+L3# 800 F2x
Pentium 4EE 3460 3730 8 512+L3# 1066 F2x-F4x
Pentium 4E 2400 2800 16 1024 533 F3x-F4x
Pentium 4E 2800 3800 16 1024 800 F3x-F4x
P4 505-519 2666 3060 16 1024 533 F3x-F4x
P4 520-571 2800 3800 16 1024 800 F3x-F4x
P4 620-672 2800 3800 16 2048 800 F4x
P4 631-661 3000 3600 16 2048 800 F6x
Xeon 1400 2000 8 256 400 F0x-F1x
Xeon MP 1 CPU 1400 2800 8 512+L3@ 400 F1x-F2x
Xeon 1800 2800 8 512+L3@ 400 F2x
Xeon 2000 3200 8 512+L3@ 533 F2x
Xeon 2800 3800 16 1024+L3$ 800 F3x-F4x
Xeon 3166 3666 16 1024+L3& 667 F4x
P4 D 805 2666 16*2 1024*2 533 F47
P4 D 820-840 2800 3200 16*2 1024*2 800 F47,F44
P4 D 920-960 2800 3600 16*2 2048*2 800 F6x
P4EE 840 3200 16*2 1024*2 800 F44
P4EE 955-965 3460 3730 16*2 2048*2 1066 F62,F64
Xeon 7020 2666 16*2 1024*2 667 F48
Xe 7110N-7140N 2500 3333 16*2 1024*2+L3* 667 F68
Xe 5020-5050 2500 3000 16*2 2048*2 667 F64
Xeon 7040 3000 16*2 2048*2 667 F48
Xeon 7030 2800 16*2 2048*2 800 F48
Xe 7110M-7140M 2600 3400 16*2 2048*2+L3* 800 F68
Xeon 7041 3000 16*2 2048*2 800 F48
Xe 5060-5080 3200 3733 16*2 2048*2 1066 F64
Pentium 4 M 1400 2600 8 512 400 F2x
Pentium 4 M 2400 3200 8 512 533 F29
Pentium 4 M 2800 3200 16 1024 533 F34
P4 M 548 3333 16 1024 533 F41
Celeron 1700 1800 8 128 400 F1x
Celeron 2000 2800 8 128 400 F2x
Cel D 310-355 2130 3333 16 256 533 F3x-F4x
Cel D 347-365 3060 3600 16 512 533 F6x
Celeron M 1400 2400 8 256 400 F29
L3# 2 MB, L3@ top models 1 or 2 MB, L3$ 0, 2 or 4 MB
L3& 0, 4 or 8 MB, L3* 4, 8 or 16 MB,
P4 D = Dual Core, Celeron D = 1 Core, M = Mobile
Back To Contents List
Pentium M Class
CPU MHz MHz KB L1 KB L2 FSB CPUID
From To Cache Cache MHz Model
PM 900 1700 32 1024 400 695
PM 733-778 1100 2100 32 2048 400 6D6,6D8
PM 730-780 1600 2267 32 2048 533 6D8
PDM T2060-2130 1600 1860 32*2 1024 533 6EC
CS T1350 1860 32 2048 533 6E8
CS U1300-1500 1060 1333 32 2048 533 6EC,6E8
CS T1300-1400 1666 1833 32 2048 667 6EC,6E8
CM 600 800 32 512 400 695
CM 353-373 900 1000 32 512 400 6D6
CM 310-340 1200 1500 32 512 400 695
CM 350-390 1300 1700 32 1024 400 6D6,6D8
CM 420-450 1060 2000 32 1024 533 6EC,6E8
CDM T2050-2450 1600 2000 32*2 2048 533 6E8
CDM U2400-2500 1060 1200 32*2 2048 533 6E8
CDM L2300-2500 1500 1830 32*2 2048 667 6EC,6E8
CDM T2300-2700 1666 2333 32*2 2048 667 6E8
PM = Pentium M
PDM = Pentium Dual Core Mobile
CM = Celeron Mobile
CS = Core Solo
CDM = Core Duo Mobile
Back To Contents List
Core 2 Class Mobile
CPU MHz MHz KB L1 KB L2 FSB CPUID
From To Cache Cache MHz Model
C2S U2100-2200 1060 1200 32 1024 533 ??
CM 520 1600 32 1024 533 6F6
CM 520-560 1600 2130 32 1024 533 10661
PDM T2310-2390 1460 1860 32*2 1024 533 6FD
C2M U7500-7600 1066 1333 32*2 2048 533 6F2,6FD
C2M T5200-5300 1600 1860 32*2 2048 533 6F2,6F6
C2M T5250-5750 1666 1500 32*2 2048 667 6F2,6F6,6FD
C2M L7200-7400 1333 1500 32*2 4096 667 6F6
C2M T7200-7600 2000 2333 32*2 4096 667 6F6
C2M T5470 1600 32*2 4096 800 6FD
C2M T7100-7250 1800 2000 32*2 2048 800 6FD
C2M T8100-8300 2100 2400 32*2 3072 800 10676
C2M L7300-7500 1400 1600 32*2 4096 800 6FA
C2M T7300-7800 2000 2600 32*2 4096 800 6FA,6FB
C2M T9300-9500 2500 2600 32*2 6144 800 10676
C2M P7350 2000 32*2 3072 1066 10676
C2M P8400-8600 2260 2400 32*2 3072 1066 10676
C2M P9500 2530 32*2 6144 1066 10676
C2M T9400-9600 2530 2800 32*2 6144 1066 10676
CXM X7800-7900 2600 2800 32*2 4096 800 6FA-6FB
CXM X9000 2800 32*2 6144 800 10676
C2S = Core 2 Solo Mobile
CM = Celeron Mobile
PDM = Pentium Dual Core Mobile
C2M = Core 2 Duo Mobile
CXM = Core 2 Duo Extreme Mobile
Back To Contents List
Core 2 Class DeskTop and Server
CPU MHz MHz KB L1 KB L2 FSB CPUID
From To Cache Cache MHz Model
Cel 420-440 1600 2000 32 512 800 10661
CeD E1200-1400 1600 2000 32*2 512 800 6FD
PDC E2140-2200 1600 2400 32*2 1024 800 6F2,6FB,6FD
C2D E4300-4700 1800 2600 32*2 2048 800 6F2,6FB,6FD
C2D E6300-6400 1860 2130 32*2 2048 1066 6F2,6F6
C2D E7200 2533 32*2 3072 1066 10676
C2D E6320-6800 1860 2930 32*2 4096 1066 6F6
C2D E6540-6850 2333 3000 32*2 4096 1333 6FB
C2D E8190-8400 2666 3000 32*2 6144 1333 10676
C2E X6800 2930 32*2 4096 1066 6F6
XEO 3040-3050 1860 2133 32*2 2048 1066 6F2,6F6
XEO 5110-5138 1600 2133 32*2 4096 1066 6F6,6FB
XEO 3060-3070 2400 2666 32*2 4096 1066 6F6
XEO E5205 1860 32*2 6144 1066 10676
XEO E7210-7220 2400 2930 32*2 8192 1066 6FB
XEO 3065-3085 2333 3000 32*2 4096 1333 6FB
XEO 5130-5160 2000 3000 32*2 4096 1333 6F6,6FB
XEO E5220-5240 2333 3000 32*2 6144 1333 10676
XEO E3110-3120 3000 3160 32*2 6144 1333 1067A
XEO X5260 3333 32*2 6144 1333 10676
XEO X5272 3400 32*2 6144 1333 10676
C2Q Q6600-6700 2400 2666 32*4 8192 1066 6F7,6FB
C2Q Q9300 2500 32*4 6144 1333 10677
C2Q Q9450-9550 2888 2830 32*4 12288 1333 10677
QXE 6700-6800 2666 2930 32*4 8192 1066 6F7,6FB
QXE 6850 3000 32*4 8192 1333 6FB
QXE 9650 3000 32*4 12288 1333 10676
QXE 9770-9775 3200 32*4 12288 1600 10676,10677
XEO E7310-7320 1600 2133 32*4 4096 1066 6FB
XEO E5310-5320 1600 1860 32*4 8192 1066 6F7,6FB
XEO E7340 2400 32*4 8192 1066 6FB
XEO L7345 1860 32*4 8192 1066 6FB
XEO X3210-3230 2133 2666 32*4 8192 1066 6F7,6FB
XEO X7350 2930 32*4 8192 1066 6FB
XEO E7330 2400 32*4 6144 1333 6FB
XEO X3320-3330 2500 2666 32*4 6144 1333 10676
XEO E5335-5345 2000 2333 32*4 8192 1333 6F7,6FB
XEO L5335 2000 32*4 8192 1333 6FB
XEO X5355-5365 2666 3000 32*4 8192 1333 6F7,6FB
XEO E5405-5450 2000 3000 32*4 12288 1333 10676
XEO L5410-5420 2333 2500 32*4 12288 1333 10676
XEO X5450-5460 3000 2160 32*4 12288 1333 10676
XEO X3320-3370 2500 3000 32*4 12288 1333 10676,1067A
XEO E5462-5472 2800 3000 32*4 12288 1600 10676
XEO X5472-5482 3000 3200 32*4 12288 1600 10676
Cel = Celeron
CeD = Celeron Dual Core
PDC = Pentium Dual Core Desktop
C2D = Core 2 Duo
C2E = Core 2 Extreme
C2Q = Core 2 Quad
QXE = Core 2 Extreme Quad
XEO = Xeon
|
Back To Contents List
AMD
Athlon MP varieties are essentially the same as XP.
CPU MHz MHz KB L1 KB L2 Type L2 FSB CPUID
From To Cache Cache Cache MHz Model
K5 75 100 8 512 External 50-66 50x-53x
K6 166 300 32 512 External 66 56x-57x
K62 300 475 32 512-1024 External 100 58x
K63 400 500 32 256+L3 Full Speed 100 59x
Duron 550 1800 64 64 Full Speed 100 63x,66x
Athlon 500 750 64 512 Half or < 100 62x
Athlon Tbird 650 1400 64 256 Full Speed 100 64x
Athlon Tbird 1000 1400 64 256 Full Speed 133 64x
AXP 1500-2100 1333 1733 64 256 Full Speed 133 66x
AXP 1600-2600 1400 2133 64 256 Full Speed 133 68x,6A0
AXP 2600-2800 2083 2250 64 256 Full Speed 167 68x,6A0
AXP 3100 2200 64 256 Full Speed 200 68x,6A0
AXP 2500-2800 1833 2133 64 512 Full Speed 133 6A0
AXP 2500-3200 1833 2333 64 512 Full Speed 167 6A0
AXP 2600-3200 1900 2200 64 512 Full Speed 200 6A0
Sem 2200-2800 1500 2000 64 256 Full Speed 167 681,6A0
Sem 3000 2000 64 512 Full Speed 167 6A0
All the following have full speed L2 cache and HyperTransport bus
CPU MHz MHz KB L1 KB L2 Maximum
From To Cache Cache RAM MB/Sec
Sem 3100 1800 64 256 3200
Sem 3000 1800 64 128 3200
Sem 2600-3300 1600 2000 64 128 3200
Sem 2500-3600 1400 2200 64 256 3200
SMo 2600-3500 1600 2000 64 128
SMo 2800-3800 1600 2200 64 256
SMo 3200-4000 1600 2200 64 512
SMo SI-40 2000 64 512
A64 FX51-57 2200 2800 64 1024 3200 2 Chan
A64 FX60-74 2600 3000 64*2 1024*2 DD2 2 Chan
A64 2800-3700 1600 2400 64 1024 DD2 1 Chan
A64 2800-3700 1800 2400 64 512 DD2 1 Chan
A64 3000-4000 1800 2600 64 512 DD2 2 Chan
A64 3700-4000 2200 2400 64 1024 DD2 2 Chan
A64 1640B 2700 64 512 DD2
A64 LE-1660 2800 64 512 DD2
AX2 3600-6000 1900 3100 64*2 512*2 DD2 2 Chan
AX2 4000-6400 2000 3200 64*2 1024*2 DD2 2 Chan
AX2 BE2300-50 1900 2100 64*2 512*2 DD2
AX2 4450B-5400B 2300 2800 64*2 512*2 DD2
AX2 4050e-4850e 2100 2500 64*2 512*2 DD2
AX2 TK-57 1900 64*2 512*2 DD2
AX2 QL-60 1900 64*2 1024*2 DD2
AMo 2700-3000 1600 2000 64 512 DD2 1 Chan
AMo 2800-4000 1600 2600 64 1024 DD2 1 Chan
TMo ML28-ML32 1600 1800 64 512 DD2 1 Chan
TMo ML30-ML34 1600 1800 64 1024 DD2 1 Chan
TMo MT32 1800 64 512 DD2 1 Chan
TMo MT34-MT37 1800 2000 64 1024 DD2 1 Chan
TMd TL50 1600 64*2 256*2 DD2 1 Chan
TMd TL52 1600 64*2 512*2 DD2 1 Chan
TMd TL56-TL66 1800 2300 64*2 1024*2 DD2
TMd ZM80-ZM86 2100 2400 64*2 2048*2 DD2
TMd RM-70 2000 64*2 1024*2 DD2
Opt 140-150 1400 2400 64 1024 DD2 2 Chan 10F50
Opt 840-850 1400 2400 64 1024 DD2 2 Chan
Opt 842-856 1600 3000 64 1024 DD2 2 Chan
Opt 242-256 1600 3000 64 1024 DD2 2 Chan
Opt 144-154 1800 2800 64 1024 DD2 2 Chan
Opt 1210-1222 1800 3000 64*2 1024*2 667 2 Chan
Opt 2210-2222 1800 3000 64*2 1024*2 533 2 Chan
Opt 8212-8222 2000 3000 64*2 1024*2 533 2 Chan
O2C 265-290 1800 2800 64*2 1024*2 DD2 2 Chan
O2C 865-890 1800 2800 64*2 1024*2 DD2 2 Chan
O2C 165-185 1800 2600 64*2 1024*2 DD2 2 Chan
CPU MHz MHz KB L1 KB L2 KB L3 RAM
From To Cache Cache Cache
PH3 8400-8750 2100 2400 64*3 512*3 2048 DD2
PH3 8600B 2300 64*3 512*3 2048 DD2
PH4 9500-9950 2200 2600 64*4 512*4 2048 DD2
PH4 9100e-9350e 1800 2000 64*4 512*4 2048 DD2
PH4 9600B 2300 64*4 512*4 2048 DD2
AXP = Athlon XP Sem = Sempron, SMo = Mobile Sempron
A64 = Athlon 64 AX2 = A64 Dual core AMo = Mobile A64
PH3 = Phenom 3 core PH4 = Phenom 4 core
Opt = Opteron, O2C = Opt Dual Core TMo = Turion 64 Mobile
TMd = Dual Core Turion 64 Mobile
DD2 - Latest DDR2 up to 800 MHz
|
Back To Contents List
Cyrix
CPU MHz MHz KB L1 KB L2 Type L2 FSB Typical
From To Cache Cache Cache MHz RAM
Cyrix 120 200 16 256-512 External 60-66 SDRAM
Cyrix MX 150 233 64 512 External 66 SDRAM
Cyrix M2 225 333 64 512-1024 External 75-83 SDRAM
|
Back To Contents List
3. Benchmark Results Used
CPUID (WhatCPU.zip) has tests using 1, 2, 3 and 4 registers, simply
adding 1. Tests are on integers, 32 bit SP and 64 bit DP floating
point, MMX 32 bit integers, SSE SPFP, 3DNow SPFP and SSE2 DPFP.
Assembly code loop has 20 adds and decrement/jump. The latter are
included in instruction count for integer speed.
BusSpd2K (.zip) is read only using streamed sequential integer data
using 64 AND from memory integer instructions and 512 MMX MOV
64 bit instructions in the assembly code timed loops. The MMX test
is intended to show maximum data transfer rate from memory.
RandMem (.zip) uses C code with of eight & xi[xi[i+0]] | xi[xi[i+2]]
statements (FP +, -, *) reading data or eight xi[xi[i+2]] =
xi[xi[i+0]] statements for read/write. The pair are used for
reading/writing 128 bytes in the loop with the indexing producing
either sequential or random accessing. Four tests cover 32 bit
integer operation and four 64 bit double precision floating point.
With indexing calculations, this benchmarks has more instructions
per word read or written than the other memory tests. Data transfer
rate may be less but the MIPS/MHz ratio higher.
BenchNT.zip provides Whetstone, Dhrystone, Linpack and Livermore
Loops Classic benchmarks, representing old code much with a small
number of instructions in loops. Benchmark codes are provided
compiled with full optimisation and no optimisation.
SSE3DNow (.zip) uses the same read and read/write functions as
MemSpeed - C code - s=s+x[m]*y[m] and x[m]=x[m]+y[m]. The tests
are run via the C code in Single Precision Floating Point with 16
unrolled statements in the inner loop (read 128 bytes, read/write
64/64 bytes). When available, the tests are also run using assembly
instructions for SSE (SP), 3DNow (SP) and SSE2 (DP) handling the
same amount of data in the inner loop.
FFTGraf (.zip) calculates Fast Fourier Transforms using single and
double precision floating point numbers. Version 1 uses optimised
C code with later versions using assembly code instructions. Memory
accessing is on a random/skipped sequential basis where performance
suffers due to burst reading from RAM. Earlier systems read in 32
byte bursts with later ones using 64 bytes with potentially slower
speed on this and RandMem benchmarks.
BusSpd2K, RandMem and SSE3DNow run tests with increasing memory
demands to measure performance via caches and RAM. Similarly,
FFTGraf uses increasing FFT sizes.
Back To Contents List
4. CPU/L1 Cache Speed %MIPS/MHz - 32 bit integers
CPUM is CPUID
BUS1 and BUS2 are BusSpd2K ANDing to 1 and 2 registers
RSRD and RSWR are RandMem serial read, read/write
DOPT and DNOP are Dhrystone 2 Optimised and Non-optimised
WHET is from Whetstone average integer MIPS
MMX1 is CPUID and MMX2 BusSpd2K
CPU CPUM BUS1 BUS2 RSRD RSRW DOPT DNOP WHET MMX1 MMX2
80486 90 45 47 53 45 53 19 44 N/A
Pentium 198 49 95 114 89 135 29 98 N/A
Pentium Pro 196 95 95 181 57 156 45 157 N/A
Pentium MMX 200 49 96 121 96 137 30 105 363 197
Celeron 196 98 96 186 89 160 45 159 292 196
Pentium II 196 97 97 186 89 160 45 159 293 196
Pentium III 197 97 98 186 89 160 45 159 289 197
Celeron M 196 97 99 193 90 176 50 156 292 199
Pentium M 196 97 99 198 91 211 52 181 286 199
Pentium 4 287 95 93 198 82 120 14 47 200 187
Pentium 4E/D 275 94 93 178 69 118 19 63 200 190
Celeron C2 M 288 98 98 189 134 264 52 176 364 196
Core 2 Duo M 293 99 99 130 155 271 53 178 365 197
Core 2 Duo 293 99 99 204 146 269 52 180 363 197
AMD K6X 197 94 96 130 109 144 22 99 197 197
Duron 244 99 187 129 93 167 61 160 399 380
Athlon 244 99 184 129 93 167 61 160 399 376
Athlon XP 244 100 187 129 93 167 61 160 389 379
Turion 64 292 99 191 137 97 197 61 169 395 395
Athlon 64 293 100 193 126 92 200 62 170 400 397
Cyrix MX 86 128 86 69 135 30 85 163
|
Note: assuming CPUs of the same MHz, Pentium 4 is faster than
AMD 64 on only one example and AMD CPUs have larger L1 cache.
Pentiums could also be faster than P4s, probably due to shorter
pipelines. Pentium 4E has even longer pipelines and performance
characteristics are different to earlier P4s, some tests producing
slower results and others faster.
Pentium M/Celeron M and Core 2 Duo CPUs show superior performance
to the P4 range on many of these tests, influenced by shorter
pipelines.
Back To Contents List
5. CPU/L1 Cache Speed %MFLOPS/MHz - 32/64 bit SP/DP floating point
CPUF is CPUID where SP and DP results are usually the same
SSRD and SSRW are SSE3DNow normal SP Read and Read/Write
RSRD is RandMem serial read, usually same as random read
WHET is Whetstone SP benchmark average
LMAX and LAVG are Livermore Loops maximum and average
FFSP and FFDP are from FFTGraf version 1 for smallest FFTs
FFS2 and FFD2 are from FFTGraf optimised version 2
CPU CPUF SSRD SSRW RSRD WHET LMAX LAVG FFSP FFDP FFS2 FFD2
80486 8 8 5 6 6 8 4 5 4 5 4
Pentium 94 33 9 22 17 33 12 18 11 17 14
Pen Pro 100 66 18 33 21 60 17 24 19 40 30
Pent MMX 95 39 11 24 17 34 13 21 17 19 16
Celeron 100 65 17 33 21 59 18 37 31 46 40
PII 100 66 17 33 21 59 18 37 28 44 33
PIII 100 66 18 33 21 59 18 37 32 46 42
Celeron M 90 90 19 43 21 87 21 46 39 56 50
Pentium M 90 90 19 43 21 88 22 45 39 55 49
P4 80 32 9 28 9 60 11 26 19 38 32
P4E/D 66 51 14 22 8 50 10 25 23 29 29
Celeron C2M 98 86 25 42 21 89 22 50 42 61 52
Core 2 DuoM 100 83 23 40 21 90 23 37 42 61 53
Core 2 Duo 100 91 23 43 21 93 22 52 43 63 54
AMD K6X 47 29 9 15 15 36 12 14 14
Duron 100 87 23 34 17 113 20 44 38 60 59
Athlon 100 86 22 34 17 113 20 44 38
Athlon XP 100 88 23 34 17 113 20 47 42 65 58
Turion 64 100 84 22 35 17 114 21 46 40 64 57
Athlon 64 100 88 23 34 17 116 21 49 44 64 59
Cyrix MX 11 10 18 8
|
AMD CPUs show superior performance to P4 of same MHz on all these
floating point tests, average 1.9 times. They also have a larger
L1 cache. Pentium M/Celeron M and Core 2 Duo CPU scores are similar
to AMD’s but L1 cache sizes are smaller.
Back To Contents List
6. CPU/L1 Cache Speed %MFLOPS/MHz - SSE SP, 3DNow SP, SSE2 DP FP
xxSP is either 32 bit SSE or 3DNow
xxDP is 64 bit SSE2
CPxx is CPUID
RDxx is SSE3DNow Read and RWxx Read/Write
FFxx is FFTGraf version 3 single and double precision
80486 -
Pentium -
Pen Pro -
Pent MMX -
Celeron -
PII -
Cyrix MX -
CPU CPSP CPDP RDSP RDDP RWSP RWDP FFSP FFDP
PIII 184 - 160 - 76 - 54 -
Celeron M 187 94 191 89 93 48 61 40
Pentium M 189 94 193 90 97 49 60 40
P4 200 100 122 61 65 33 38 32
P4E/D 200 100 146 74 86 43 34 31
Celeron C2M 399 200 244 122 186 96 83 67
Core 2 DuoM 400 200 252 128 193 96 88 69
Core 2 Duo 400 200 264 132 198 99 93 72
AMD K6X 199 - 96 - 81 - ?? -
Duron 199 - 177 - 105 - 63 -
Athlon 200 - 171 - 102 - 63 -
Athlon XP* 200 - 196 - 99 - 63 -
Turion 64+ 200 100 187 90 96 48 63 44
Athlon 64+ 197 100 194 97 99 50 71 44
- Extensions N/A
* has SSE and 3DNow, similar performance
+ has SSE, 3DNow and SSE2
|
Again AMD performance is superior to P4s but similar to
Pentium M/Celeron M. Core 2 Duo scores are by far the best
on these SSE/SSE2 tests.
Back To Contents List
7. CPU/L2 Cache Speed %MIPS/MHz - 32 bit integers
BUS2 is BusSpd2K ANDing to 2 registers
RSRD is RandMem serial read
RSWR is RandMem serial read/write
RRRD is RandMem random read
MMX2 is BusSpd2K
CPU BUS2 RSRD RSWR RRRD MMX2
80486 11 19 9 10
Pentium 29 48 24 20
Pentium Pro 57 135 49 62
Pentium MMX 29 54 25 24 33
Celeron 54 122 63 75 91
Celeron 2 61 147 81 123 106
Pentium II 19 93 26 48 30
Pentium III 19 93 26 48 30
Pentium IIIE 61 148 81 123 107
Celeron M 63 143 75 96 108
Pentium M 63 146 76 98 108
Pentium 4 69 150 42 83 106
Pentium 4E/D 63 117 36 64 111
Celeron C2 M 85 171 132 91 110
Core 2 Duo M 85 174 136 92 110
Core 2 Duo 85 179 138 95 110
AMD K62 21 62 20 13 21
AMD K62+ 63 107 79 29 82
Duron 55 89 56 54 74
Athlon 19 40 19 16 30
Athlon TB 63 93 60 55 80
Athlon XP 63 93 60 55 80
Turion 64 72 78 50 70 80
Athlon 64 78 91 57 81 121
Cyrix MX 45 22 22 38
|
Pentium 4 is faster than AMD CPUs of the same MHz up to Athlon
XP on these L2 cache tests. Athlon 64 shows improvements on
these speeds. Core 2 Duo is better than both P4 and Athlon 64
on most tests and has a much larger L2 cache.
Back To Contents List
8. CPU/L2 Cache Speed %MFLOPS/MHz - 32/64b SP/DP floating point
LINP is Lipack (Large L1 cache helps)
SSRD and SSRW are SSE3DNow normal SP Read and Read/Write
RSRD is RandMem serial read, usually same as random read
FFSP and FFDP are from FFTGraf version 1 for FFTs that use L2
FFS2 and FFD2 are from FFTGraf optimised version 2
CPU LINP SSRD SSRW RSRD FFSP FFDP FFS2 FFD2
80486 4 5 2 3 5 4 5 4
Pentium 13 17 6 7 12 9 16 13
Pentium Pro 24 60 13 26 20 19 39 30
Pentium MMX 12 23 7 9 12 11 16 13
Celeron 27 52 16 17 25 22 40 33
Celeron 2 31 56 17 17 30 27
Pentium II 15 40 11 11 16 13 26 18
Pentium III 15 40 11 11 16 13 26 18
Pentium IIIE 31 56 17 18 30 27 44 39
Celeron M 42 70 16 29 31 29 50 45
Pentium M 45 71 16 31 31 29 50 45
Pentium 4 28 31 9 30 23 18 39 32
Pentium 4E/D 21 40 12 21 22 20 21 22
Celeron C2 M 55 87 19 35 32 31 52 48
Core 2 Duo M 55 81 21 34 32 31 52 48
Core 2 Duo 59 87 21 37 33 31 52 48
AMD K62 12 8 7 5
AMD K63 14 23 7 11
Duron 38 40 12 14 16 15 26 16
Athlon 36 8
Athlon TB 38 40 12 20 16 15 28 28
Athlon XP 38 40 12 20 16 15 37 29
Turion 64 37 43 12 21 16 15 42 28
Athlon 64 38 48 12 22 19 19 45 32
Cyrix MX 11 7
|
Latest AMD CPUs have similar average performance to P4s of the same
MHz, the former benefiting from faster floating point and the latter
from more efficient L2 cache. Pentium M/Celeron M are relatively
better but top marks go to Core 2 Duo.
Back To Contents List
9. CPU/L2 Cache Speed %MFLOPS/MHz - SSE SP, 3DNow SP, SSE2 DP FP
xxSP is either 32 bit SSE or 3DNow, xxDP is 64 bit SSE2
RDxx is SSE3DNow Read and RWxx Read/Write
FFxx is FFTGraf version 3 single and double precision
CPU RDSP RDDP RWSP RWDP FFSP FFDP
Pentium III - - -
Pentium IIIE 106 - 42 - 52 -
Celeron M 110 55 42 21 51 36
Pentium M 110 55 40 20 51 36
Pentium 4 118 60 46 23 40 34
Pentium 4E/D 139 69 54 27 36 30
Celeron C2 M 171 88 85 42 78 60
Core 2 Duo M 175 86 80 42 79 60
Core 2 Duo 178 87 86 43 79 61
AMD K62 - - -
AMD K63 62 - 36 - -
Duron 64 - 27 - 27 -
Athlon - - -
Athlon TB 68 - 26 - 30 -
Athlon XP 68 - 26 - 37 -
Turion 64 73 35 21 11 43 25
Athlon 64 94 45 25 12 47 28
- Extensions N/A
|
Performance of these SIMD streaming instructions is superior on the
P4 versus AMD CPUs of the same MHz. Athlon 64 shows improvements
over earlier AMD CPUs. Again, Core 2 Duo is best.
Back To Contents List
10. RAM Speed %MIPS/MHz - 32 bit integers
Performance via data in RAM may also depend on CPU MHz besides
RAM/bus speeds. This can result in a reduction in %MIPS/MHz with
higher speed CPUs. There can also be wide variations according to
mainboard/chipset.
BUS2 is BusSpd2K ANDing to 2 registers
RSRD is RandMem serial read
RSWR is RandMem serial read/write
RRRD is RandMem random read
MMX2 is BusSpd2K MMX instructions
CPU MHz RAM BUS2 RSRD RSWR RRRD MMX2
80486 66 9 18 6 5
Pentium 100 18 31 13 6
Pentium Pro 200 30 36 14 9
Pentium MMX 200 17 34 16 6 17
Celeron A 300 66 23 42 14 11 37
Celeron A 550 66 14 26 10 8 20
Celeron 2 566 66 15 27 11 8 23
Celeron 2 800 PC100 14 25 9 6 23
Pentium II 450 PC100 18 37 12 11 27
Pentium III 450 PC100 18 37 12 11 27
Pentium III 800 PC100 15 34 9 9 24
Pentium IIIEB 666 PC133 17 32 10 8 24
Pentium IIIEB 1000 PC133 15 28 6 6 23
Celeron M 1295 44 13 4
Pentium M 1862 PC2700D 31 62 16 10 34
Pentium 4 1400 PC800 24 52 17 4 29
Pentium 4 2400 PC800 16 34 11 3 20
Pentium 4N 2533 PC1066 21 43 15 4 23
Pentium 4N 1800 PC2100 20 43 13 3 23
Pentium 4N 2533 PC2100 16 34 9 2 17
Pentium 4N 3000 PC2700 18 37 8 2 19
Pentium 4N 3300 PC3200D 23 49 19 3 26
Pentium 4E/D 3000 PC3200D 30 65 21 3 32
Celeron C2 M 2000 DD2533D 33 84 22 7 44
Core 2 Duo M 1830 DD2666D 50 104 30 11 50
Core 2 Duo XX 2400 DD2533D 45 95 9 4 51
Core 2 Duo 2400 DD2533D 45 102 29 9 54
Core 2 Duo 2400 DD2800D 51 109 35 11 60
AMD K6X 500 PC100 13 31 8 3 13
Duron 600 PC133 18 23 12 3 38
Duron 1000 PC133 10 15 7 2 26
Athlon 800 PC100 11 16 8 2 31
Athlon TB 750 PC133 17 18 9 3 31
Athlon TB 1000 PC133 11 14 7 2 26
Athlon XP 1200 PC2100 13 18 9 4 23
Athlon XP 1733 PC2100 13 19 9 5 19
Athlon XP 2167 PC2700D 12 22 13 5 18
Turion 64 1900 DD2320D 33 50 25 7 35
Athlon 64 2000 PC3200D 34 43 25 3 37
Athlon 64 2210 PC3200D 32 54 24 4 33
Cyrix MX 225 66 30 10 7
D = dual channel DD2 = DDR2
XX - slow nForce 570 chipset on some tests
|
Comparing CPUs of the same MHz and the same memory speed, P4s show
superiority performance of speed via data in RAM compared with AMD
with CPUs prior to Athlon 64. Core 2 Duo performance is outstanding
on some of these tests but poor on others, perhaps due to using the
nForce 570 chipset. Later results via Intel 965 chipset confirmed this
suspicion, giving rise to Core 2 Duo with DDR2 RAM taking the lead on performance.
Back To Contents List
11. RAM Speed %MFLOPS/MHz - 32/64b SP/DP floating point
Examples - see paragraph 10 notes, but FP variations are less
than with integers.
SSRD and SSRW are SSE3DNow normal SP Read and Read/Write
RSRD and RRRD are RandMem serial read and random read
FFSP and FFDP are from FFTGraf version 1 for largest FFTs
FFS2 and FFD2 are from FFTGraf optimised version 2
CPU MHz SSRD SSRW RSRD RRRD FFSP FFDP FFS2 FFD2
80486 66 5 2 2 1.0 3 3 4 3
Pentium 100 10 4 4 1.6 6 6 12 9
Pentium Pro 200 16 5 6 2.2 7 6 23 15
Pentium MMX 200 14 5 5 1.3 7 6 13 11
Celeron 300 19 5 4 2.2 8 6 22 12
Pentium II 450 16 5 4 2.0 5 5 15 10
Pentium III 450 16 5 4 2.0 5 5 15 10
Pentium IIIEB 800 12 4 4 1.6 5 4 22 16
Celeron M 1295 28 7 9 0.9 3 3 19 14
Pentium M DCDDR 1862 33 9 11 2.8 6 5 26 21
Pentium 4 RDRAM 2000 22 5 6 0.9 3 3 15 12
Pentium 4 DDR 2533 17 3 5 0.4 2 2 10 10
Pentium 4 DCDDR 3300 25 7 7 0.8
Pentium 4E DCDDR 3000 27 9 10 0.9 4 3 13 12
Celeron C2M DDR2 2000 39 12 12 1.8 4 4 27 23
Core 2 DuoM DDR2 1830 57 16 17 2.6 7 7 31 28
Core 2 Duo DDR2a 2400 8 3 5 1.2 3 3 26 17 XX
Core 2 Duo DDR2a 2400 50 16 16 2.5 8 7 35 31
Core 2 Duo DDR2c 2400 54 18 18 2.8 10 9 37 33
AMD K63 400 6 2 4 0.7 3 3
Duron 950 8 3 4 0.5 3 2 13 8
Athlon 550 5 0.8
Athlon TB 1000 4 0.5 2 2 13 8
Athlon XP 2167 15 6 5 1.3 3 3 16 12
Turion 64 DDR2d 1900 32 10 10 1.9 5 5 22 17
Athlon 64 2000 32 9 11 1.2
Athlon 64 DCDDR 2210 33 10 11 1.5 5 5 23 18
Cyrix MX 225 4 1.2
XX - slow nForce 570 chipset on some tests
DDR2a 533 MHz, DDR2c 800 MHz, DDR2d 3200 MHz
|
Superior floating point performance with AMD CPUs helps to overcome
inferior RAM speeds. Again, some nForce 570 chipset, Core 2 Duo scores are disappointing.
Back To Contents List
12. RAM Speed %MFLOPS/MHz - SSE SP, 3DNow SP, SSE2 DP FP
Examples - see 10. notes, but FP variations are less than with
integers.
xxSP is either 32 bit SSE or 3DNow, xxDP is 64 bit SSE2
RDxx is SSE3DNow Read and RWxx Read/Write
FFxx is FFTGraf version 3 single and double precision
CPU MHz RDSP RDDP RWSP RWDP FFSP FFDP
Pentium IIIE 800 26 - 8 - 23 -
Celeron M 1295 28 14 8 4 22 13
Pentium M DCDDR 1862 31 16 9 5 34 20
Pentium 4 DDR 2533 18 9 4 2 12 9
Pentium 4 RDRAM 2411 24 12 7 4 14 11
Pentium 4 DCDDR 2533 22 11 6 3 15 11
Pentium 4E DCDDR 3000 31 15 10 5 22 16
Celeron C2M DDR2 2000 39 19 11 6 35 25
Core 2 DuoM DDR2 1830 60 30 17 8 43 31
Core 2 Duo DDR2a 2400 7 4 3 2 26 17 XX
Core 2 Duo DDR2a 2400 51 26 16 8 46 33
Core 2 Duo DDR2c 2400 60 30 20 10 49 39
AMD K63 400 7 - 2 - -
Duron 950 15 - 5 - 13 -
Athlon Tb 1200 9 - 4 - 12 -
Athlon XP 2167 19 - 7 - 17 -
Turion 64 1900 38 19 14 7 25 17
Athlon 64 2000 34 17 10 5 25 16
Athlon 64 DCDDR 2210 39 20 12 6 26 17
- Extensions N/A XX - slow nForce 570 chipset on some tests
DDR2a 533 MHz, DDR2c 800 MHz, DDR2d 3200 MHz
|
All these could justify that a 2 GHz Athlon 64 is equivalent to
a 3 GHz P4 when running SIMD streaming instructions with data
in RAM. Yet again, some nForce 570 chipset, Core 2 Duo results are very poor.
END
Back To Contents List
To Start
More Historic Computer Speeds
Computer Speeds From Instruction Mixes pre-1960 to 1971
Whetstone Benchmark History and Results (1960's to 1990's, PCs to 2006)
Computer Speed Claims 1980 to 1996 - PCs and IBM Mainframes to 2004
Main Page Roy Longbottom's PC Benchmark Collection
To Start
Produced by Roy Longbottom December 2004
Updated August 2008
|
