|
|
[Continued from
Page 2]
Benchmark
Results
Sysmark 2000 is the most up to date and broad based application benchmark
available. It is suspiciously processor centric – not inappropriate when
evaluating high-end CPUs and platforms. But it is not a benchmark
that should be taken too seriously for ValuePCs.
Using the 840 run time
scores as a baseline reference, we charted the performance delta for the
1GHz and 1.2GHz Athlon DDR systems. On a clock for clock basis, Athlon
outperformed the 840 fairly consistently, averaging about 2%. On a
dollar for dollar basis, Athlon outperformed the P3 by more than 11%.
Photoshop is a glaring exception to this overall trend. SSE optimized
code could be handing the advantage to the P3, or it could be due to a difference
in cache allocation schemes. For now, we consider this real but not
adequately explained. Other
than Photoshop, we feel that the Athlon makes a great showing in SysMark.
Run times for each platform are recorded in the table below.
3D
is the application that demands the most from high end PC platforms.
We tested at low and medium resolution to allow the processor and DRAM performance
differences to show their full potential. In the charts below, DDR
makes a strong and consistent showing against the 840. On a clock
for clock basis, Athlon is 7-10% faster than the P3. On a dollar for
dollar basis Athlon is 16-24% faster than the P3.
There are no exceptions
here. 3D WinMark 2000 v1.1 also shows Athlon DDR with an incredible
performance lead. Taking an average of all of the scores produced
by 3D WinMark, the 1GHz Athlon outperforms the P3 by more than 15%, and
the 1.2GHz Athlon outperforms the P3 by more than 25%. In the
critical polygon performance tests, the Athlon+DDR performance advantage
over the P3 exceeds 40% (at 1GHz as well as at 1.2GHz).Around
the time Intel launches the P4, MadOnion will coincidentally have a new
version of 3D WinMark (2001) with endless hours of P4 performance tuning,
included for free. We are curious to see if the new benchmark might
be able to shift perceptions, and produce results that differ significantly
from what we see here.
Viewperf
is the most respected and popular MCAD style benchmark around.
It generates solid reproducible figures every time. On a clock for
clock comparison with the P3, Athlon+DDR pulls ahead by 6.7% on average.
On a dollar for dollar comparison, Athlon magnifies its lead to 15.9%
on average for all of the tests in ViewPerf.
The
most reproducible synthetic bandwidth test around is StreamD. It operates
under DOS using a 32-bit DOS extender to access large amounts of DRAM.
It issues huge volumes cache-line burst read and burst write traffic on
the processor’s external bus. It is not subject to OS variations.
Overall, Athlon delivers a performance benefit of more than 50% vs. the
P3. In the most extreme case, the Scale test shows a 2x performance
delta favoring Athlon (858MB/s vs. 419MB/s). We presume that the P4
could also do well on this test, once it shows up in late November.
Our
final benchmark is LinPack (Linear Algebra Package). It is an age-old
benchmark from the days when Digital ruled scientific computing.
This test creates memory arrays of varying size and performs double precision
floating point calculations on the data set to determine the throughput
of the platform based on the active memory size. This test stresses a
combination of the FPU unit, cache and DRAM.
One
can easily see the benefit of Athlon’s exclusive L1 and L2 caches in these
results. Both Athlon platforms stretch their peak output beyond 256K,
to something closer to about 320KB, as throughput descends to the limits
of front side bus and DRAM.
In the table above, we
have averaged the MFLOPS results for all memory array sizes under 512K and
over 512K for comparative purposes. In the cached region under
512KB, Athlon outperforms the P3 by 17 to 36%. In the DRAM bound region
over 512KB, Athlon again outperforms the P3 by 32 to 42%.
Conclusion
For as long as the P3 remains a static target at 1GHz and below, Athlon
has it completely cornered in terms of price and performance. Intel’s
Tualatin chip could only breathe new life into P3 if Intel drives clock
speeds to match Athlon in the second half of 2001. At this point
we doubt that Intel will do this, preferring instead to fight Athlon with
the P4.We expect that AMD will beat Intel to a fully integrated 512K L2
cache, by shipping a new CPU core in early Q1’00. This will be almost
6 months ahead of Intel’s transition to a 512K cache. AMD will also
debut its dual processor platform for Athlon. This will be very attractive
to the workstation market. Dual Duron platforms could also become
quite a curiosity among enthusiasts on a budget. All of AMD’s activity
to expand the high end of the market will be complemented by VIA, ALI,
SIS and others with inexpensive, high volume chip sets for Athlon and
Duron in the mainstream, adding features such as UMA. With
certainty, these same companies will also successfully deliver DDR to
the P3 platform ahead of Intel, to strengthen its position in the midrange
and the low end.There is no doubt, that with the fall of the P3, the end
of an era is upon us. 2001 will prove to be an interesting year
- to see who will control the next Dynasty, and for how long.
To
Beginning of Article
Bert
McComas
is founder of InQuest Market Research.
He has 15 years experience in technical marketing of computer chipsets,
analog and power ICs, and computer boards and systems.
Have a comment
about this article? We invite you to discuss issues or post questions on
the DigitalVideoEditing
Forum. |