AMD athlon 64 Manuale Utente

40555

Rev. 3.00

June 2006

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).

 

113%

136%

144%

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)

0 Hop
0 Hop

1 Hop
1 Hop

NO 

Xfire

1 Hop
1 Hop

Xfire