Single Node Performance

It is easy to waste a lot of money on poor system design. To illustrate this, we consider the variety of AMD Athlon processors available and their costs. The same considerations apply to Intel or Alpha processors. Component prices vary a great deal during their lifetime, so we give a date for the graphs that depend upon price.

Processor price is a rapidly increasing function of speed. It can be rapidly decreasing function of time, especially for the faster processors.

Dividing by the speed of the chip, we still see that the relative expense rises rapidly for the faster chips. In this case, there was an apparent sweet spot at 600 MHz. The faster chips have a higher price-performance ratio. Depending upon the costs of the other components of the system, the entire system may have a higher (undesirable) or lower (desirable) price-performance ratio.

For our QCD codes, access to memory is quite important. In our first example, we shall compare three Celeron chips and one Pentium II chip. One of the Celerons has been overclocked, but the two with 66 MHz bus show little variation despite the difference in chip speed. The PII with a 100 MHz bus is faster despite its slower clock rate. (It also has a larger cache.)

Celeron
330 MHz
66 MHz FSB 		Celeron
OC to 416
82 MHz FSB 		Celeron
500 MHz
66 MHz FSB 		PII
350 MHz
100 MHz FSB
L
MF
MF
MF
MF
4
102
117
101
113
6
60
74
56
83
8
57
71
52
72
10
57
71
52
70
12
56
70
52
70
14
56
71
52
70

By comparing 500 MHz and 600 MHz Athlons, we demonstrate that performance does not increase in proportion to the speed of the chip. This is because memory access is fixed by the FSB speed.

Athlon
500 MHz
100 MHz FSB 		Athlon
600 MHz
100 MHz FSB 		Ratio
L
MF
MF
(1.2)
4
231
276
1.19
6
129
135
1.05
8
97
102
1.05
10
92
97
1.05
12
90
95
1.05
14
89
93
1.06

When comparing these two tables, we see that for L = 4, for which the problem fits in cache, there is a 19.5% speedup on the faster processor. But for all the larger problems, the speedup is only 5%. We expect that for even faster processors, the memory access will become an even greater issue and performance increases will be marginal.

Since memory access is so crucial, I have purchased a Pentium III 533B chip that uses PC133 memory. In theory, it should provide about 33% better performance than a similar chip with PC100 memory. I have tried three different motherboards using different support chips and the results are disappointing. The Gigabyte GA6VXE+ motherboard uses a VIA chipset, the Supermicro PIIISED uses the Intel 810e chipset and I also tried an Intel CC820 motherboard using the Intel 820 chipset. The results are not particularly better than a PII 350 or 450 MHz chip using a BX motherboard, except when the problem fits in cache. In the future, I hope to test whether the PIII prefetch instructions can be used to improve this situation. (This technique would also be applicable to a BX motherboard with PIII installed.) The next version of the Portland Group Compiler is supposed to have better facilities to tell the compiler when to prefetch data to cache.

Gigabyte
GA6VXE+
VIA 		Intel
CC820
I 820 		Supermicro
PIIISED
810e 		PII
350 MHz
100 MHz FSB 		RR PII
450 MHz
PGI compiler
L
MF
MF
MF
MF
MF
4
186
182
174
113
141
6
106
98
94
83
99
8
81
76
73
72
82
10
76
72
70
70
79
12
76
70
69
70
78
14
73
70
69
70
78

It is possible that the 815e support chip from Intel will provide better performance with the 133 MHz FSB, but I have not done any tests myself. There are benchmarks at the Intel developer's web site that look promising. However, there are support chips from ServerWorks that do support PC133 memory well. Here are results from Los Lobos, that uses Intel 733 MHz chips in IBM NetFinity servers. I don't know if the motherboards are made by IBM or a third party. Both Supermicro and Tyan are manufacturing motherboards that use ServerWorks chips. The Supermicro costs about $275. NCSA let me run benchmarks on a system with this motherboard.

LosLobos
733 MHz
133 MHz FSB 		Supermicro
370 DLE
1000 MHz
L
MF
MF
4
319
413
6
140
148
8
130
136
10
127
131
12
127
129
14
126
129

You might have noticed above that the Athlon numbers are fairly impressive especially considering that an Athlon chip will cost less than an Intel chip at comparable speed. There has been good news recently on the Athlon front. DDR memory is now available from Micron, and there are several motherboards that support DDR for sale. The Biostar M7MIA uses the AMD 760 chipset. Iwill KA266 uses the Ali MAGiK 1 chipset and the Gigabyte GA-7DXC uses the AMD 761 chipset. Dual processors have not been available for the Athlon, but there will be an AMD 760MP chip that will support dual processors. On October 10, 2000, AMD announced a demonstration dual-CPU DDR system, so it should not be too many more months before these are available to the consumer.

On January 18, Iwill announced the DVD266-R, the first DDR Dual CPU motherboard for the Pentium III processor.

Itanium Results (next slide)

Back to Outline