ultima wrote on 19.03.14 at 08:30:51:Only differences are ofcourse I am running dual channel DDR400 vs single channel DDR300 with the XP-M and hypertransport link 1Ghz vs 150Mhz fsb and we all know voodoo's love high bandwidth.
The comparison is incorrect.
1GHz is the path's speed between the CPUs and the AMD 8131 controller, but the clock is 200MHz (x5 = HT).
So you must compare 150MHz clock frequency of the old system with 200MHz of the new one, which is higher, so better...
But HT works completely different from the FSB. FSB is a single external bus shared across the I/O devices on the system, RAM included of course. Since the AMD64 architecture provide a memory controller inside the CPU itself, the system memory is directly linked to the CPU with his own bus (whit all the benefits of the case), while the rest of the I/O devices present in the system share a second external bus, which is the Hyper Transport. But, while the FSB has a data width of 64bit, the HT implemented around the AMD CPUs has a data width of 32bit. Speaking of bandwidth:
Previous AXP-M platform:
150MHz*64bit*2/8=2400MB/s bandwidth
Current Opteron platform:
200MHz*32bit*2/8=1600MB/s bandwidth
It must be noted that HT is a parallel path, it means that data can flow through the bus in both directions at the same time. So it can be considered splitted in two sub-links by 16bit width each (16bit_upstream + 16bit _downstream = 32bit aggregated). That said, someone could argue that the real bandwidth must be calculated as the bandwidth available in only one direction (16bit width instead of 32bit), as you would do in the FSB case.
So, speaking about pure bandwidth, you have less now. But HT is a more efficient technology and provide some clear benefits over the FSB of which the most importants are:
1) Parallelism (I/O data flowing at the same time)
2) Memory and CPU deal directly (it means that, for example, while system devices as VGA access to RAM through NB, CPU can still access to RAM at the same time with reduced latency)
Home
