The diagonal region implies that we can continue to add instructions and get higher because rate to get more data is slower than the rate to compute.
Also, the graph suggests that it is harder to use Opteron X4 machine efficiently because we need to give more data.
This comment was marked helpful 0 times.
kkz
Did this test take advantage of memory pre-fetching? (Assuming that arithmetic data was pulled from RAM).
This comment was marked helpful 0 times.
Q_Q
This graph also seems to indicate that the Opteron x4 has a faster memory bus
This comment was marked helpful 0 times.
mofarrel
@Q_Q
This actually does not show the Opteron x4 has a faster memory bus.
This is showing that they have the same one. Note that when they are both bandwidth limited (the sloping portion of the graph), they are getting the same performance. As soon as the Opteron x2 is compute limited the Opteron x4 has better performance as it has more processing power.
The diagonal region implies that we can continue to add instructions and get higher because rate to get more data is slower than the rate to compute.
Also, the graph suggests that it is harder to use Opteron X4 machine efficiently because we need to give more data.
This comment was marked helpful 0 times.
Did this test take advantage of memory pre-fetching? (Assuming that arithmetic data was pulled from RAM).
This comment was marked helpful 0 times.
This graph also seems to indicate that the Opteron x4 has a faster memory bus
This comment was marked helpful 0 times.
@Q_Q This actually does not show the Opteron x4 has a faster memory bus. This is showing that they have the same one. Note that when they are both bandwidth limited (the sloping portion of the graph), they are getting the same performance. As soon as the Opteron x2 is compute limited the Opteron x4 has better performance as it has more processing power.
This comment was marked helpful 0 times.