Menu Content/Inhalt
Home arrow Server & Workstation arrow x86 PC Architecture Evolutions

x86 PC Architecture Evolutions Print
April 2012

Computers as in PC were first mass produced in the mid eighties of the last millennium (1980).  They were based on an architecture called x86 which in turn turn came from an Intel processor called 8086 created in 1978.  Intel subsequently produced 80286, 386 and 486 processors; and then Pentium, Pentium MMX, Pentium II, Pentium III and Pentium 4; and then Core2 and Core-i processors.   These names are more for marketing purposes than representing the real capabilities of the processors.  As of today, we have Core-i3, i5, i7 and Xeon E3 and E5 processors and they are based on the same architecture with minor variations.  x86 is in the category of Complex Instruction Set (CISC) as against Reduced Instruction Set Computing (RISC).     

Presumably modern processors are faster than the initial models.  How much faster?  What else is better?  Frankly, not much else is better.  Modern processors are not cooler (as in temperature) or quieter (they need fans to take away the heat).  They are not prettier as they are similar to the heart of a human body which needs to be functioning well and not looking pretty.  However, if we extend the question from processors to the PC system level, we will see a lot more improvements such as larger and more real life visual displays, 12 channel audio, large data storage reservoirs, and wireless links to the outside world etc.  This article focuses on x86 computer processor units (CPU) only.  Let us address speed now.  There are 3 easy to understand parameters to indicate the improvement of computing speed. 

The 1st parameter is the bit length of the processor.  8086 is based on 16 bits.  80386 is based on 32 bits.  Pentium 4 is based on 64 bits and we are still on 64 bits as of 2012 and for the near future as the next jump is very big.  Based on time to the market, Opteron from AMD was the first 64b processor released ahead of Pentium 4. In comparison, Internet addressing schemes have done better in having 32 bits initially (IPv4) and progressing to 128 bits as of today (IPv6).   We did not hide IPv5 because it was never used. 

The 2nd parameter is the clock frequency of the processor which dictates how soon the next instruction should be worked on.  8086 worked on 20MHz.  Pentium 4 reached 3.8GHz (this is 190 times faster than 20MHz) in 2004 and processors stopped chasing for a faster clock since.  Why is 3.8GHz the maximum clock frequency?  The processor would get too hot and burnt out otherwise.  Running the processor at a higher clock level than officially prescribed called over-clocking is a risk to be taken by the user, as this would kill the processor.  Intel and AMD officially allowed over-clocking and even called this Turbo Clocking when processors have 4 or more cores and some cores are not being used.  Over clocking active cores is acceptable till the running temperature gets high.

The 3rd parameter is the number of cores in the processor.  All initial processors have a single core only. Intel released the first 2 core processor as Core 2 and AMD released the first 2 core processor as AthlonII.  As in 2012, Core i7 has 6 cores, Xeon E5 has 8 cores, and AMD Opteron has 12 cores.   However, the higher end Intel processors have a capability called Intel Hyper Threading.  It allows the resources of a core not currently used by an application to be used by a 2nd application concurrently.  In theory, Xeon has 16 threads and can support 16 incidences of computation.       

Processors interface with the rest of the PC such as with memory modules via links.  The number of links, which are more commonly called pins, is therefore one indication of the capabilities of the processor but this is not quite true.  Pentium 4 has 478 pins.  Core i7-3900 has 2011 pins.  Core i5 has 1155 pins.   Some earlier Core-i5 models have 1156 pins and they were not faster.  The large number of pins is needed for the processor to link to memory modules in pairs, trios and quads.

For better performance, a PC can have a companion processor or a 2nd processor.  Companion processors were common prior to the release of Pentium MMX.  Examples are 80287 and 80387 and they had longer bit length for handling floating point calculations (as against fixed point or digits for the fractional part of a number).  Higher end workstations and servers have 2 physical sockets on the motherboard.   Sure, some have 4 sockets or 8 sockets but they are not in the x86 family. 
A breakthrough has showed up and it is the advent of GPU (graphics processor units) and the system intelligence of using GPU to assist CPU for computation. The latest GPU released by Nvidia in April 2012 has over 1000 cores in the form of an add-on card with PCI Express interface.  By compiling an application on CUDA enabled compiler, the software codes will know when to use CUDA cores or CPU cores.  CUDA is an Nvidia scheme for this purpose and has attracted global attentions.  Please see a separate article on CUDA.