|
Main Page
Introduction
Sun Microsystems T1000
The RSA Factoring Challenge
The Constructive Approach
Evaluation Summary
Future Research
Evaluation Chronology
System Configurations
Financial Disclaimer
Postscript
© Grebyn Corp. 2006
|
|
This page provides a summary of the T1000 evaluation. For the activities
that lead here, check out the Chronology page.
We will first calculate Sun's SWaP (Space Watts and
Performance) metric, discuss some issues related to the differences in
the systems and the utilization of the systems capabilities
Space
The T1000 is a nice small 1U box. In fact, when removing it from the rack
to return it (having not actually looked at it in 50 days, I was
surprised by how small (and heavy) it was. The Athlon64 and Pentium IV
machines, being deskside towers were much larger.
| T1000
| Athlon64
| Pentium IV
|
|
1U
|
4U
|
4U
|
Space Requirements
Watts
The power consumption numbers were interesting. But based on the
configurations (lots of memory and two drives on the T1000, even more drives
on the Pentium IV, older system design), it was not unexpected.
| T1000
| Athlon64
| Pentium IV
|
|
149
|
22
|
133
|
Power Consumption
Performance
In raw performance, the T1000 actually outperforms the Athlon64 and the
Pentium IV. Based upon the last set of tests run, executing the solution to
an equivalent problem required the following times.
| T1000
| Athlon64
| Pentium IV
|
|
3:24:18
|
4:12:18
|
4:56:41
|
Timings
Plugging these numbers in, we get the following SWaP values for the three
systems:
| T1000
| Athlon64
| Pentium IV
|
|
0.32
|
0.45
|
0.06
|
SWaP
Cost and Price / Performance
Clearly the T1000 configuration under test was not adequately "represented",
since the unused memory and disk drives contributed to the cost and the
power load. It appears that there are now more different configurations
from Sun and it appears that a comparable configuration, consisting of an 8
core system, single SATA drive and 2GB of memory might be available for
around $5,000 and would probably have a reduced power requirement (120 used
for these calculations).
| T1000 (as tested)
| Athlon64
| Pentium IV
| T1000 (proposed)
|
|
$14,445
|
$600
|
$1,100
|
$4,995
|
Cost
| T1000 (as tested)
| Athlon64
| Pentium IV
| T1000 (proposed)
|
|
44,000
|
1,300
|
17,000
|
12,000
|
Price / Performance Ratio
While this appears to make the new configuration competitive with the
Pentium IV under test, that system would also have to have its power
requirements adjusted for reduced disk load in order to be a fair
comparison.
Discussion
While there were many variables and issues associated with the evaluation
that could easily leave it open to criticism as not an apples-to-apples
comparison, and while it would be a really neat platform for further
research, particularly non-homogeneous multi-threaded applications, the pure
cost and price / performance of the original system, and even the proposed
replacement system (especially without similar comparisons to other
commodity multi-core offerings) for such a compute-bound application make
the selection of this system prohibitive.
Perhaps the next-generation Niagara chip (with 64 threads) and advances in
manufacturing will continue to increase the performance benefit and reduce
the cost of resulting systems. Additional algorithm and implementation
improvements to the constructive approach to more effectively utilize the
Niagara 64 bit technology and the T1000 architecture (particularly memory
utilization and potentially the networking capabilities) could also benefit
the application.
Previous
Next
|
|
 |
|
|
|