The test results above are from machines used by thousands of businesses 20 years ago. How much has business changed? The fastest computer in the Byte '83 required an 18-wheel truck to move and took hours for a dozen men to set up. But some businesses must have considered it to be worth the $3,000,000. The least expensive of the machines above would have been difficult for one person to carry very far and none worked on batteries. So if modern netbooks outperform all of those old computers how can they possibly not be suitable for plenty of businesses today?

The computer landscape began shifting when IBM introduced the Personal Computer, or PC, in 1981. The PC used the Intel 8088, one of the first 16-bit microprocessors, a distant ancestor of the Atom used in most netbooks. Curiously, at 22 seconds, the original PC took exactly 10,000 times as long to run the benchmark as the netbooks using the same language. The 3033 mainframe running the PL/I language was 600 times as fast as the PC though it cost 1000 times as much. Did IBM come up with the term Personal Computer to persuade people that is was not a REAL COMPUTER? The PC was actually more powerful than more expensive machines that IBM was selling at the time, like the Systems 32 and 34. The System 34 could take a maximum of 256 KILOBYTES while the PC could use 640 kilobytes. And yet 8 terminals could be attached to the System 34 so eight people could use the machine simultaneously.

This benchmark data demonstrates a very important aspect of computer software, EFFICIENCY. Every computer ultimately runs what is called machine code, but that is not what programmers write most of the time. The vast majority of code is written in some high level language which must be translated into machine code by a program called a compiler. Occasionally code is written in something called Assembler or Assembly Language which produces the fastest code possible. But an Assembly Language programmer must know the internal structure of the CPU being coded for and this code will not work on any other type of CPU. But this is the only way to get 100% efficient code. Compiled code just isn't as good. The IBM 3033 mainframe had three different results. The 0.0078 second result was the fastest of all of the computer and software combinations tested. It was the most efficient code on the fastest machine. The code produced by the PL/I compiler took almost five times as long at 21.7% efficiency and COBOL took more than ten times as long.

This variation in efficiency does not necessarily correspond to the language used. Three different C compilers were tested on 8 megahertz 8086 processors. The C by Digital Research was exceptionally efficient at almost 68%. Microsoft's C was less than half as efficient and Computer Innovation's was slightly worse than that. The BD Systems C compiler was 45% efficient although that was on a different processor, the Z-80. But in every case the efficiency is compared to the performance of that same processor running Assembly Language. People who only use computers for running applications and don't get to select what language and which compiler their application is written in may not be in a position too choose, but they must live with the resulting efficiency of the software. A significant LACK OF EFFICIENCY can force users to upgrade their hardware.

Another cause of performance difference is the size of the data bus. The 8086 and the 8088 are the same internally and they execute the same machine code but the data bus between the memory and the processor is 16 bits wide on the 8086 but only 8 bits on the 8088. But in the tests above the 8086 also has a 60% higher clock speed at 8 MHz instead of 5. So in these test the 8088 is at a disadvantage because of both the clock speed and bus size. In the Assembly Language tests the 8086 was 2.1 times as fast as the 8088. But if that 8088 had been running at 8 MHz it would have taken only 2.5 seconds and the 8086 would only have been 31% faster. So the bus width accounted for much of the difference in this test. But the IBM PC took three times as long with the same Computer Innovations compiler as the 8086. Compilers produce a lot more code than Assembler so that code had to be read from memory over the 8-bit bus of the 8088 and consequently took even more time.

  ----------  New tests on modern CPUs  ----------  

TI OMAP5910 150 MHz GNU C  Pocket CPU   0.0522 Archos PMA400 Linux
Pent 2 300  MHz     GNU C               0.0148
AMD    433  MHz     GNU C               0.0122 OLPC-XO 1
Pent 3 500  MHz     GNU C               0.0087
Pent 3 1.3  GHz     GNU C               0.0035
Pent 4 1.8  GHz     GNU C               0.0024
Atom   1.6  GHz     GNU C   *NETBOOKS   0.0022 Asus/Lenovo
Pent 4 2.66 GHz     GNU C               0.0020 Single-Core 1 program
Pent 4 2.66 GHz     GNU C               0.0040 Single-Core 2 programs
Pent 2D 2.8 GHz     GNU C               0.0019 Dual-Core   1 program
Pent 2D 2.8 GHz     GNU C               0.0019 Dual-Core   2 programs