New SSE2 Instructions Set

   Intel introduced the MMX (MMX for MultiMedia eXtensions) instructions set back in 1996: this was the first instruction addition to the x86 architecture since the i386 was released. Remember, the MMX instruction set was full of multimedia dedicated features that were here to accelerate applications which used them. MMX applications came a bit later on the market but most users enjoyed MMX benefits especially in games since game developers adopted them quickly. More recently Intel added SSE (Streaming SIMD Extension) to its Pentium III: this was a set of 70 supplementary extensions that used the SIMD (Single Instruction Multiple Data) technology just like the MMX or 3D Now (From AMD). SIMD principle is simple: it should treat only one pass of several data with only one instruction. Compared to a standard SISD x86 instruction (Single Instruction Single Data) where one instruction should give one result the SSE can give up to 4 results in the same clock cycle. 

The SSE instruction set principally enhances audio and video compression processes as shown by our tests: indeed compressing an audio file of 130 MB took 8.5 seconds against 10.3 seconds using a Pentium III 1GHz: for this kind of task the Pentium 4 boosts performance up to 28%. SSE 2 brings several enhancements dedicated to boost MPEG 2 encoding and file encrypting processes. First it adds 144 new instructions (oriented on memory and cache management) to the SSE & MMX existing ones but it can now handle integers of 128-bit numbers (1 per cycle), and double precision floating of 64-bit (two per cycle). Just like the MMX the SSE 2 set of instructions is no use if you don’t have compliant applications that take benefit from it: actually no applications manage it except the Direct X 8.0 API. But some compatible applications should be out very soon like the Windows Media Encoder 2, Dragon Naturally Speaking 4, etc. We also ran another test showing the undeniable power the Pentium 4 brings to high demanding multimedia applications: compressing an Indeo video of 15MB into an MPEG2 one took 1.05 minutes against 1.38 minutes for the Pentium III 1GHz, showing a difference of more than 42%. The performance enhancements shown by the tests are due both to the higher frequency of the CPU and its various MMX and SSE instructions.

Thermal Security

   The Pentium 4 CPU features a security system called Thermal Monitor. This exclusive function permanently monitors the temperature of the CPU via a Thermal Diode in order to protect the processor against long term reliability damages. If the temperature of the CPU or the chassis is higher than what it’s recommended, the rotation speed of the fans will be increased (if available) and the CPU performance may drop below the peak level to prevent critical damages. If ever the CPU reaches a high temperature of approximately 69° Celsius (in case the fan is disconnected, broken, etc.), the Thermal Monitor will instantly shut down the computer so your CPU will not burn.

Pentium 4 2.2GHz CPU Design

   The Pentium 4 2.2GHz uses the brand new µPGA Socket478 package that was inaugurated by the Pentium 4 2GHz. It is clearly the reference socket for future Pentium 4 processors. It means Socket 478 based motherboards will be upgradeable. Even if this new socket permits the Pentium 4 chip being much smaller, despite 55 new added pins, it isn’t compatible with existing Socket 423pins motherboards. However we’re expecting to see some CPU converters appear that’d give an added lifetime to existing P4 platforms. 

Intel Pentium 4 478 pins

After having changed the packaging of its Pentium 4 CPUs; making them smaller with the µPGA 478 Pin, Intel now attacks their core. The first technical advance is the use of a new Px60 process so NorthWood CPUs are now engraved in 0,13µ. The engraving fineness is not all, since the inter connexion used in the processor are now in copper instead of aluminium. Using copper is now essential to miniaturize processors, since it’s a very good conductor with very good heat dissipating capabilities. This new manufacturing process ensures the CPU can reach higher frequencies while heating less. Not to say the power consumption of the CPU has been reduced from 1.75 to 1.5Volts. But there are other interests for Intel to use the 0.13µ manufacturing process: Intel has developed new 300mm wafers that let them multiply their production volume by 2.25 (over previous 200mm wafers) implying an appreciable cost reduction of 25%. With a single Wafer Intel now carves about 180 Pentium 4 chips. The Northwood now includes 55 millions of transistors, instead of 42 millions for P4 Willamette.

Pentium 4 0.13µ Wafer (click to enlarge)
 

   Since the engraving process is finer and more accurate there’s more room available on the silica plaque so Intel engineers have increased the level 2 cache memory to 512 KB. Initially the Pentium 4 was featuring a 256 KB L2 cache memory which was clearly limiting its performance. This L2 cache size increased, and now provides 10-15% higher performance at equal frequency.

   As usual the top part of the CPU features a metal cover that was added by Intel to protect the die unit from damage if the radiator was incorrectly mounted or is too heavy as well as ensuring better thermal dissipation. The heatsink has been totally revamped: indeed Intel has released a new plastic retention mechanism, which is much handier, to attach the cooler and heatsink to the motherboard. Nonetheless you can still use old school Pentium 4 heat sink systems. 

New Pentium 4 HeatSink Retention Mechanism (click to enlarge)

   Actually the Pentium 4 2.2GHz is officially compatible with three different chipsets: the Intel i850, Intel i845 and Intel i845 B0 (DDR). Indeed Intel has at last released their DDR chipset, so you can now use a Pentium 4 along with cheap yet powerful and popular Double Data Rate memory. Teamed up with DDR memory the Pentium 4 provides an overall level of performance that is almost as good as the one you can achieve with expensive rambus memory and the i850 chipset as we will see later in the benchmarks area.