Pentium 4 1.5GHz or Athlon Thunderbird 1.2GHz (DDR)
tnaw_xtennis

Though there has been so much mixed viewpoints and feelings expressed by CPU review reporters
and readers on the just released Pentium 4, the rational way that judges a new CPU by yourself
remains the same. What is the Sweetspot of Pentium 4 and Athlon respectively? To set up a dreaming
PC, is a Pentium 4 or an Athlon your best choice?

By reviewing carefully the benchmarking results of the Pentium 4 1.5GHz and the Athlon Thunderbird
1.2 GHz (DDR), we can get the conclusion that: (1). The Integer and Floating-Point computing
performances of Athlon 1.2 are more stronger than those of P4 1.5. (2). Comparing with Athlon (DDR)
system, the FSB (Front Side Bus) and memory bandwidth of P4 system are much more powerful.
(3). SSE2 (Streaming SIMD Extensions 2) optimization of multimedia software could speed up
performances of P4 system markedly. And the optimizations are not as difficult as you may imagine.
(4). Performances of different type applications may be sensitive to one of the system components
mentioned above, P4 1.5 or Athlon 1.2 could beat away its competitor easily in the field be good at itself.

1. Performances of different type applications may be sensitive to different system components

We all know that to run software requires the coordinated interactions of numerous components
in the computer. When the instructions and data are sent to the CPU, CPU handles the incoming
instructions and data, performs necessary computations and sends instruction and data to RAM
and peripheral devices. The video card takes video data from the CPU and process it into a format
that the monitor or display device can understand and we get the final results. The CPU communicates
with the main memory and peripheral devices through FSB (Front Side Bus). The CPU can only
manipulate data that is held in the main memory (RAM), so if it needs to access data on a storage
device the CPU instructs the device to send the necessary data and instructions to RAM.

In CPU benchmarking nowadays, three main factors from the hardware side that could determine the
benchmarking performances are: (1). CPU, (2). FSB and MBW (memory bandwidth). (3). Video card.
The three parts work coordinately. If one lags behind, it becomes the bottleneck that limits the overall
performance. Let us suggest that CPU, FSB and MBW could handle over 100M units of instructions
and data/s respectively, while the video card could only process an amount of video data corresponding
to 80M units of instructions and data/s, then video card becomes the bottleneck. If CPU and video
card could process over 100M units of instructions and data/s, while FSB and MBW could handle
80M units/s only, then the bottleneck comes from FSB and MBW. And CPU would be the bottleneck
if it could only process 80M units of instructions and data/s, while the FSB, MBW and video card can
handle 100M units/s.

Most of the applications today don't request evenly the strength of the three components mentioned
above. Different types of applications may depend on different system components to bring performances
into full play. Some type of applications may depend on strength of the CPU,  i.e., CPU as bottleneck.
and don't stress FSB, MBW and video cards. While other type of applications may not stress the strength
of CPU and video cards, but depend on the FSB and MBW, i.e., FSB and MBW as bottleneck.
It is also noticeable that even for the same application the factor that determines the performance may
changes with changing of the condition applied to the application..

The figures below show an example of shifting of the bottleneck with the changing of the condition applied.

* source: www.anandtech.com

At the screen resolution of 640x480, Quake 3 performance depends mainly on the CPU, FSB and
MBW. The Quake 3 performance varies on the computer with different CPU (types, core speed), FSB
and MBW. At the screen resolution of 1024x768, the amount of video data that video card need to
process increases, and video card seems can't handle of all them and becomes the bottleneck. Since
the GeForce 2 GTS were applied equably by all testing systems, we got the same benchmarking result.  

* source: www.anandtech.com

When video card becomes the bottleneck of a system, Quake 3 performance can't be improved by
changing a more powerful CPU, FSB or MBW. Only a faster video card  can do the trick at this
moment, just as an example showed above. It is obviously for 3D gamers that a powerful video card
is a better choice than a high-end CPU, e.g., a P4 1.5 or a Athlon 1.2. Since no seriously 3D gamers
play at a screen resolution of 640x480 nowadays, and the 3D game performances of  a PC at a higher
resolution is depend mainly on the video card.

2. The sweetspots of Athlon Thunderbird 1.2GHz (DDR)

The SiSoft Sandra CPU benchmarking results in Table 1 indicates clearly that the integer and floating
-point computing performances of Athlon 1.2 are more stronger than those of P4 1.5. The SiSoft
Sandra CPU Benchmark is exclusively measures of integer and floating-point performances of a CPU.
Factors such as FSB frequency or memory bandwidth don't affect the score as you can see in the case
of Athlon.

Table 1. SiSoft Sandra CPU Benchmark

Since most of popular commercial applications and games nowadays are integer or floating-point
intensive, and don't deal with large stream data that are stressful on the FSB and memory bandwidth,
Athlon Thunderbird 1.2GHz (DDR) runs these applications and games faster than Pentium 4 1.5GHz
with taking the advantage of its stronger integer and floating-point computing ability over P4.

Below are benchmarking results of two applications that flaunt the victory of Athlon over P4 thoroughly.
RC5 and 3D Studio Max2 Rendering test respectively and exclusively the integer performance and
floating-point performance of the CPU. In both tests, FSB and memory bandwidth don't affect the results
absolutely

* source: www.anandtech.com

* souce: www.tomshardware.com

3. The sweetspots of Pentium 4

The sweetspots of Pentium 4 are on those applications dealing with large stream data that could take the
advantage of the very powerful FSB and memory bandwidth of the Pentium 4, and on those SSE2
optimized multimedia applications. These has been reviewed in the last posted, here are the key point:
(1). In the specialized memory bandwidth benchmarking, Pentium 4 1.5GHz get ahead Athlon Thunderbird
1.2GHz (DDR) by +79 ~ +161%. To the applications performing on large data streams such as image
recognition, 3D computer model review and Video 2000 MPEG2-Encoding, with taking the advantage
of the 400MHz FSB and 3.2GB/s memory bandwidth, Pentium 4 1.5GHz runs up to +41% faster than
Athlon Thunderbird 1.2GHz (DDR).
(2). In the specialized SSE2 impact benchmarking, SSE2 optimized boosts Pentium 4 1.5GHz performances
up to +104%. To applications such as 2D/3D image viewer, video and audio editing /encoding /playing,
and speech recognition with taking the advantage of SSE2 optimized, Pentium 4 1.5GHz could run up
to +243% faster than Athlon Thunderbird 1.2GHz (DDR).

The poor performance of Pentium 4 in the integer and floating-point computing exceeded all expectations
indubitably. This is certainly far way from the design goal of Pentium 4, i.e., "a micro-architecture
maintaining an average IPC (Instructions Executed Per Clock ) that was within approximately 10% to
20% of the P6 micro-architecture. In this design, although the IPC is lower, the increase in frequency
capability more than makes up (Performance = frequency x IPC) and deliver overall higher performance
capability to the end user". What is the reason for this in nature? According to analytic information at
present, one of the most possible key factors may be execution characteristics of software code at present
don't cooperate well effectively with the new micro-architecture of Pentium 4. If this is the case indeed,
software re-compiled and optimized may make Pentium 4 show better performance. The figure below show
the results of MPEG4-encoding benchmark with an original version of FlasK MPEG and a  re-compiled
version of FlasK MPEG. With the re-compiled and SSE2 optimized FlasK MPEG, Pentium 4 have up
to +266% and +395% MPEG4 encoding performance boosts respectively. The work concerning the FlasK
MPEG re-compiled and SSE2 optimized was done with effort of one night only by a guy called himself
'Alex' (not so hard as you may imagine, right?).

* souce: www.tomshardware.com

1.4GHz Pentium 4 vs. 1.2GHz Athlon Thunderbird with DDR SDRAM ( 05/11/2000 )

400MHz FSB Makes the Pentium 4 Shine ( 08/11/2000)

Intel flaunts FPU strength of Pentium 4 by SPEC CFP2000 ( 17/11/2000 )

The Sweetspots of Pentium 4 at Present ( 24/11/2000 )