Dual Core x Dual Processor (Xeon/P4-D x Opteron/Athlon64)

[vagnermarks]

I was reading some explanations about dual processor and dual core and I found myself a little confused about this topic....

I have been a little dificult to understand what is the real difference about them.
I know that Athlon has your memory controller on die and it gives a better performance. But what about the Xeon and P4-D? The only way to increase the perfomance is just through their L2? Look at the simple example below:

Intel Xeon Dual Core: they use just one FSB. Two processor Xeons are conected in only one FSB 800 MHz.

Intel Petium D Dual Core will be share the same FSB 800/1066 Mhz and can use non-buffered and unregistered memory. May I consider it as an advantage under Xeon?

AMD Opteron Dual Processor: they use the HT (Hyper Transport) with high speedy to access the RAM and the chipset. It is an unquestionable fact that can make all the difference.

AMD Athlon 64 Dual Core: they will use the same memory controller. We can have the same performance when they will use the same FSB?

Well, it is confused.... somebody could help me to understand all this complete process?
Thanks

 

[Furen]

On the AMD side of things you will have better performance for two reasons:

First: The hypertransport is NOT a bottle neck (it runs at 800-1000MHz at double data rate, so 1600-2000 effective) and having the two cpus attached to a single memory controller drops the access latency considerably (versus a 2 socket system, which has two memory controllers--and more total bandwidth--but each cpu has to use the hypertransport to access the other's memory, which raises latency a bit).

Second: you have the system request interface and the crossbar switch on-die. This means that it operates at the same speed as the CPU cores.

In traditional FSB architectures the equivalents of the SRI and the crossbar are located on the northbridge, which adds significant amounts of latency to both cpu-to-cpu and memory-to-cpu operations. This means that on the intel side there is no difference between having two cpu cores in the same package (but different dies, a la pressler), same die (like smithfield) or different sockets (like having two xeons).

EDIT: Edited things a bit to make it clearer (hey, I know it's still kind of choppy but I dont feel like writing an essay...)

 

[Viditor]

Originally posted by: vagnermarks
I was reading some explanations about dual processor and dual core and I found myself a little confused about this topic....

I have been a little dificult to understand what is the real difference about them.
I know that Athlon has your memory controller on die and it gives a better performance. But what about the Xeon and P4-D? The only way to increase the perfomance is just through their L2? Look at the simple example below:

Intel Xeon Dual Core: they use just one FSB. Two processor Xeons are conected in only one FSB 800 MHz.

Intel Petium D Dual Core will be share the same FSB 800/1066 Mhz and can use non-buffered and unregistered memory. May I consider it as an advantage under Xeon?

AMD Opteron Dual Processor: they use the HT (Hyper Transport) with high speedy to access the RAM and the chipset. It is an unquestionable fact that can make all the difference.

AMD Athlon 64 Dual Core: they will use the same memory controller. We can have the same performance when they will use the same FSB?

Well, it is confused.... somebody could help me to understand all this complete process?
Thanks

Well, I think you have some unlearning to do first...

Hypertransport isn't a FSB, and AMD doesn't really have a FSB. What a FSB is basically is a single pipe that all communication to the CPU goes through.
With a standard system (Intel), the signal path goes
CPU -> Northbridge
Northbridge -> memory
Northbridge -> Southbridge (which contains the I/O like HDD, USB, etc...)
Northbridge -> AGP
Northbridge -> PCI bus (or in the case of the newer Intel Hub Architecture, PCI is connected to the Soutbridge, now called the I/O controller hub)

The connection between the CPU and the Northbridge is the FSB, and all communication between the CPU and the rest of the system must pass through the FSB.

For AMD it's quite different...
Because the Northbridge is actually part of the CPU itself, there is no FSB to pass through.
AMD instead has created direct point-to-point links from the CPU to many of these sections. So we have
CPU -> memory
CPU -> Southbridge (which includes the AGP, PCI, and I/O)
CPU -> other CPUs

Another difference is that HT links are in pairs and are unidirectional. This means that instead of data needing a "traffic cop" on the bus, the HT links move at full speed in one direction each.
The big advantage as you can see is that the data is distributed rather than centralized on a single bus. It also means that there are fewer "middlemen" which means a far lower latency.

Now on to dual core designs...
At present, AMD and Intel have 2 different designs for dual core.
AMD has designed the A64 chips to be dual core from the very beginning, so much of the communication for the 2 cores was already built in to the single core processors.
The main difference is the way that the 2 cores "talk" to each other...for AMD, the design of all the Hammer chips has included a crossbar (think of this as an internal router) which is connected directly to the cache of each core. For one core to "talk" to another, the path is
CPU -> CPU through the crossbar. This takes about 1 clock...

For Intel, the 2 cores are basically 2 seperate processors that have been packaged together. So for them the talk goes
CPU -> FSB (Northbridge) -> CPU As you can imagine, this takes MUCH longer (closer to 10 clocks)

Hope this helps...