400MHz FSB Makes the Pentium 4 Shine
Many objections have been received since my posted the page that extrapolated 1.4GHz
Pentium 4
performance numbers from the statements made by Intel's Paul Otellini about
Pentium 4 performance.
Now, some explanations to the rationality of Intel's Paul Otellini statements and those
extrapolated
results will be given from the aspects of IPC (instructions
executed per clock) and FSB (front side bus).
1. What Determines True Processor Performance and How is the Pentium 4 Designed
The information below comes from "Inside the NetBurst Micro-Architecture of the Intel
Pentium 4
Processor", a white paper of
Intel.
What Determines True Processor Performance
The only measure of performance that really matters is the amount of time it takes to
execute a given
application. Contrary to a popular misconception, it is not clock frequency (MHz) alone
or the number
of instructions executed per clock (IPC) alone that equates to performance.
True performance is a
combination of both clock frequency (MHz) and IPC:
Performance = MHz x IPC
The Design of Pentium 4
A focused architectural definition effort was used to study the benefits of many advanced processor
technologies and determine the best approach to improve the overall performance of the processor
for
many years to come. The result of this definition effort was a micro-architecture (on the same
manufacturing process) while maintaining an average IPC that was within approximately 10% to
20% of the P6 micro-architecture (Pentium Pro, Pentium 2, Celeron, Xeon, Pentium
3 ). 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. This was done in the NetBurst micro-architecture ( Pentium 4 ) by implementing a hyper-pipelined
technology where the depth of the pipeline was doubled from that of the P6 micro-architectural generation.
Although this deeper pipeline delivers significantly higher levels of frequency, the potential performance
impacts associated with the longer pipeline were comprehended and overcome in the design, The design
effort focused on the following: …
2. FSB (front side bus) Speed and CPU Performance
FSB (front side bus) Speed has an impact on the CPU Performance. As shown in the figure below,
For
the Pentium 3 600, increasing the FSB speed from 133MHz to 150MHz, we got a performance boost
of 2.6%. While increasing the FSB speed from 133Mhz to 166MHz, we got a performance boost of 4.7%
for the Pentium 3 1000. The performance boost are also found for the AMD CPU, sure the factor of
memory bandwidth increasing should be considered as well in the cases of Athlon Thunderbird.
By extrapolation, we got performance numbers of both Pentium 3 working on 400MHz FSB speed that
is employed by Pentium 4. For both Pentium 3 600 and Pentium 3 1000, a performance boost of 46% is
obtained by increasing the FSB speed from 133MHz to 400MHz.
3. Reckoning that Performance of Pentium 4 by IPC and FSB
From the figure below, we could extrapolated the performance number of Pentium 3 1400 with 133MHz
FSB speed to be 180. As has been mentioned above, a 46% performance boost will be obtained when
increasing the FSB speed applied to the Pentium 3 from 133MHz to 400MHz, then the performance
number
of the Pentium 3 1400 should be:
Pentium 3 1400(400) = Pentium 3 1400(133) x 146%
To the worst case, we take the IPC of Pentium 4 to only 80% of the Pentium III IPC and don't considering
the memory bandwidth improved from 1.03GB/s to 3.2GB/s. Then the performance
number of Pentium 4
1400(400) should at least be:
Pentium 4 1400(400) = Pentium 3 1400(400) x 80%
Now we get the Quakes3 performance number of Pentium 4 1400
Pentium 4 1400 = Pentium 4 1400(400) = Pentium 3 1400(400) x 80%
= Pentium 3 1400(133) x 146% x 80%
= 180 x 146% x 80%
= 210
According to the suggestions of Intel's Paul
Otellini, that the P4 will beat a 1GHz Pentium 3 by 44% in "video
games like Quake III," we get the Quakes3 performance number of Pentium 4 1400:
Pentium 4 1400 = Pentium 3 1000(133) x 140% = 148 x 144% = 213
From the above results, it seems that the Intel's Paul Otellini
statements, and our extrapolated results that concering
the performance comparision of 1.4GHz Pentium 4 and 1.2GHz Athlon Thunderbird with DDR SDRAM in Quake3,
MPEG-4 Encoding and 3D Studio Max2 Rendering, are quite acceptable.