Raspberry Pi 1 and 2 compared

A NEON SIMD Extension

The single-instruction multiple data [5] (SIMD) instruction set extension means that the logic in the chip uses one command to processes several pieces of data simultaneously (see the "Extended Instruction Set" box). This technology is especially important for use with multimedia applications responsible for such things as MP3 file decoding, facial recognition in photographs, and greater playback rates for high-definition videos.

The ARM11 of the RPi1 series already had a SIMD extension, but the new Neon instruction set offers significantly more possibilities. The ARM11 operates with 32-bit registers, allowing the simultaneous addition of four 8-bit values.

Neon also introduces 64- and 128-bit registers, paving the way for the addition of up to 16 8-bit values in one computing step. On top of this, the computing unit was given a series of new instructions. Neon handles SIMD instructions with this series depending on application and code quality between 1.5 and 12 times faster than the ARM11 on the RPi1s.

The RPi2B requires an average 3.5 to 4 watts of power. The RPi1B+ was satisfied with 3W. The BCM2835 is still fabricated with 65nm technology, whereas the BCM2836 is manufactured with 28-nm technology. The availability of more transistors allows four modern Cortex-A7 cores instead of only one ARM11 core to be placed on the same area.

Comparing Generations

Comparing the computing power of two processors is not easy, because you must consider the intended purpose, the performance capability of factors such as RAM and cache, and the processor-specific optimizations of the software. The Dhrystone million instructions per second (DMIPS) values determined via the Dhrystone test provide a good approximation [6] when comparing the power of different architectures.

Instead of the 875DMIPS found in the RPi1, the RPi2B has a 6,840DMIPS rating, which indicates a sevenfold increase in computing performance. In practice, the amount of increased performance depends on whether the software allows for parallel processing on the four cores. In many situations, in fact, the performance increase is observed to be even greater, such as when supplementary and faster working RAM accesses the SD card – a slow process – fewer times.

Table 1 provides an overview of the results achieved by the Raspberry Pi compared with results from other currently available single-board computers (SBCs) and a series of computers ranging from those available in earlier days to those found in today's market. The data for devices containing an ARM core are taken from ARM. As a rule, the performance data claimed is usually higher than that observed during actual use. The data of the x86 systems come from real-life measurements, mostly from Sandra Dhrystone [7].

Table 1

Comparisons of Computing Performance

                 

Compared with

Platform

RAM

Chip

Technology

Architecture

DMIPS/MHz

Cores

MHz

DMIPS

RPi1

RPi2

Raspberry Pi

Raspberry Pi 1

512MB

Broadcom BCM2835

65nm

ARM1176JZF

1.25

1

700

875

100%

13%

Raspberry Pi 2

1GB LPDDR2

Broadcom BCM2836

28nm

Cortex-A7

1.90

4

900

6,840

782%

100%

Banana Pi

Banana Pi

1GB

Allwinner A20

40nm

Cortex-A7

1.90

2

1,000

3,800

434%

56%

Banana Pi M2

1GB

Allwinner A31S

40nm

Cortex-A7

1.90

4

1,000

7,600

869%

111%

Other SBCs

Beaglebone Black

512MB

TI Sitara AM3358/9

45nm

Cortex-A8

2.00

1

1,000

2,000

229%

29%

Hummingboard i2

1GB

Freescale i.MX6 DualLite

40nm

Cortex-A9

2.50

2

1,000

5,000

571%

73%

Cubox-i4Pro

2GB

Freescale i.MX6 Quad

40nm

Cortex-A9

2.50

4

1,000

10,000

1,143%

146%

Odroid C1

1GB DDR3

Amlogic S805

28nm

Cortex-A5

1.57

4

1,500

9,420

1,077%

138%

Smartphones

Galaxy S3 Mini (GT-I8190)

1GB

ST-Ericsson NovaThor U8500

45nm

Cortex-A9

2.50

2

1,000

5,000

571%

73%

iPhone 5

1GB

Apple A6

32nm high-K metal gate

ARMv7s Swift [Apple]

3.50

2

1,300

9,100

1,040%

133%

Game Consoles

PlayStation 2

36MB

EmotionEngine

250nm

RISC, based on MIPS R5900

20.34

1

295

6,000

686%

88%

Apple Computers

Apple ][e

64KB

MOS Technology 6502

8,000nm

MOS Technology

0.43

1

1

0.43

0.05%

0.01%

Apple Macintosh 128K

128KB

Motorola 68000

3,500nm

CISC

0.23

1

6

1.4

0.16%

0.02%

iMac G3

32MB

PowerPC 750 G3

260nm

PowerPC G3

2.25

1

233

525

60%

8%

Intel and AMD PCs

No Name PC 1

64MB

Pentium II, 300MHz

350nm

x86 Intel

0.91

1

300

273.6

31%

4%

No Name PC 2

384MB

AMD Duron, 800MHz

180nm

AMD Spitfire

2.81

1

800

2,250

257%

33%

Dell Inspiron 7520

8GB

Intel Core i7-3632QM

22nm

Intel Core i7

14.19

4

2,200

99,750

11,400%

1,458%

Hetzner EQ4 server

32GB

Intel Core i7-3770

22nm

Intel Core i7

14.19

4

3,400

106,530

12,175%

1,557%

Upton compares the subjectively observed performance of a RPi2B with that of an AMD Duron CPU running at 800MHz. Equipped with a slim system, a desktop built with this chip would still work quite nicely. Compared with the RPi1 that has something like a Pentium II 300MHz level of performance, this represents significant progress.

The Raspberry Pi Foundation has always mentioned the so-called factor 6, which refers to the speed increase over the old version of the Rasp Pi. This sixfold increase results from the average of several measurements in different areas. For instance SysBench, a multithreaded CPU benchmark, shows that the RPi2B is six times faster. However, the PC may exceed or fall below this rate depending on the application.

Buy this article as PDF

Express-Checkout as PDF

Pages: 8

Price $2.95
(incl. VAT)

Buy Raspberry Pi Geek

SINGLE ISSUES
 
SUBSCRIPTIONS
 
TABLET & SMARTPHONE APPS
Get it on Google Play

US / Canada

Get it on Google Play

UK / Australia

Related content