AMD 64 Node 0 Core Node 1 Core Node 2 Core Node 3 Core, Chapter, Analysis and Recommendations

40555 Rev. 3.00 June 2006	Performance Guidelines for AMD Athlon™ 64 and AMD Opteron™
	ccNUMA Multiprocessor Systems

Here the same two foreground threads as before were run though the cases as before—local, crossfire and no crossfire. In addition, four background threads are left running on:

•Node 0 (Core 1)

•Node 1 (Core 1)

•Node 2 (Core 0)

•Node 3 (Core 0)

Each of these background threads read a local 64 MB array and the rate of memory demand of each of these threads is varied from low to very high simultaneously. A low rate of memory demand implies that each of the background threads is demanding a memory bandwidth of 0.5 GB/s. A very high rate of memory demand implies that each of the background threads is demanding a memory bandwidth of 4 GB/s as shown in Table 1 on page 16.

Even with the background threads, there are still some free cores left in the system. We call this a highly subscribed condition.

This allows us to study the impact of the background load on the foreground threads.

As shown in Figure 7 and Figure 8 on page 28, under both low and very high loads and high subscription, we still observe that the worst performance scenario occurs when write-only threads fire at each other (crossfire).

LOW: Total Time for both threads (write-write)

2.2

2

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

0

		136%	144%
113%		1 Hop	1 Hop
	0 Hop
		1 Hop	1 Hop
	0 Hop	NO	Xfire
		Xfire

0.0.w.0 1.0.w.1 (0 Hops) (0 Hops)

0.0.w.1 1.0.w.3 (1 Hops) (1 Hops)

0.0.w.1 1.0.w.0 (1 Hops) (1 Hops)

Figure 7. Crossfire 1 Hop-1 Hop Case vs No Crossfire 1 Hop-1 Hop Case under a Low Background Load (High Subscription)

Chapter 3 Analysis and Recommendations 27