AMD 1600Mhz FSB

Teach me what this means. I always thought the system bus on mine 200Mhz was the FSB. But not so I guess? Intel now has 1066Mhz FSB is this the same as system bus? I did a little looking b/4 I asked but could use some help.

 

http://en.wikipedia.org/wiki/HyperTransport#Front-Side_Bus_Replacement

 

Yeah, there is no FSB in AMD systems. Its been replaced by the HT bus. The 200MHz is the base clock speed based on which the Hypertransport and the processor clock speeds are derived.
The HT link for the Turions run at 800MHz and since they are dual data rate capable can achieve transfer speeds upto 1600MT/s which is equivalent to the transfer speeds by a single data rate hypertransport link running at 1600MHz. Thats where the 1600MHz comes from.

 

moon angel
Thanks for the link, very helpful but not the easiest read but I read and will re-read. To more fully understand can someone explain the way Intel work since no HT, or link but being a regular guy hard reads take me sometime input is always appreciated.

Comparisons between the two welcome, just trying to understand.

 

Like with so many other things on a computer, there are literally dozens of ways to measure the bus speed. The common theme is that "it's the bus used to transfer data between RAM and CPU".

AMD puts the memory controller on the CPU (which makes the front-side bus irrelevant, since that's only responsible for communication between CPU and memory controller). Instead, they need a new bus to transfer between RAM and memory controller. Enter HyperTransport.
Hypertransport runs at 800 or 1000MHz, and with dual-channel memory, that can be said to be doubled, so call it 1600 or 2000MHz.

Intel still uses the old front side bus, but they use a lot of trickery to make it faster. First, like AMD, they pretend that "dual-channel memory means the bus must be twice the speed", so 1066 is really only 533MHz. That is again doubled by some other excuse I can't remember offhand. (Maybe transferring data twice per clock cycle, the same way DDR RAM does), so what's technically speaking only running at 266MHz is said to run 1066 in marketing-speak.

Of course, what matters is really how much data can be transferred. If you check wikipedia, you can see how wide each bus is (how many bits transferred per cycle), and those two numbers multiplied tells you the total bus bandwidth.

Ultimately, it doesn't really matter. HyperTransport scales better, so on large servers (say, 8+ CPU's), HT is generally preferable. For smaller systems, both have plenty of bandwidth.

 

Jalf, that understand with 1 read, I think. So does the system bus on 1066Mhz run at 266Mhz? In 98-99 when I got my 1st computer was sys bus and FSB always the same and referred as such or was I wrong back then also?

 

The Intel bus is quad pumped meaning data is transferred four times that of a standard bus. So for a 1066MHz rated FSB the actual speed is 266MHz. Previously they used single data rate buses, so the actual frequency was the rated frequency.

 

so, in terms of actual data transferred in a given period of time, AMD's HT at 1600MHz beats intel's fastest FSB at 1066MHz on any given day, right?

 

Don't confuse actual vs potential, so in terms of actual data transferred RAM speed would matter, I think.
Any input?

 

In terms of how much data is actually transferred, I'd guess AMD wins. They have smaller caches, so they need to access memory more often. Hence more data is transferred... Oh, that wasn't what you meant?

Which bus is able to transfer the most data then? If HT and Intel's FSB are equally wide, then yes, HT would be faster. However, I can't remeber the width of either one, so you'll have to look it up

If Intel's bus is twice as wide, then it'll be able to transfer as much data at 800mhz as AMD does at 1600.

 

I did the searching for you in terms of bandwidth wikipedia and from AMD themselves here (Note these comparisons are for notebook processors).

 

I heard that the Cache system of AMD's and Intel's CPUs work differently, and that HyperTransport has something to do with it. Someone explain?

 

from a diagram i saw of the AMD architecture, the Hyper Transport is only used for IO operations. The memory controller is built into the Processor so the memory communicates directly with the processor.
This would mean that the FSB bottle neck of earlier architecture (still used by Intel) is solved and memory can communicate with the processor in its full speed if the Processor can handle it.
In intel if the FSB is 800MHz and the 2 Memory chips of 633MHz is running on dual channel: the combined Datarate of memory will be 633*2 = 1266MHz, but you wont get this performance because the FSB cannot transfer data at that rate.

Now even if AMD transfered data from memory through the Hyper transport it still has bandwidth of 1600-2000MHz compared to 1166 MHz highest of Intel

 

Hmm, well, apart from the hypertransport stuff, there are some differences in the cache.
Intel uses an inclusive cache hierarchy, while AMD uses exclusive.
Basically, when data is fetched from the L2 cache into the L1, on Intel CPU's, it is just copied, so everything that is in L1 is *also* in L2. That makes the cache simpler, but of course uses the space a bit less efficiently.
AMD's cache is exclusive, meaning that when data is pulled from L2 to L1, the corresponding slot in the L2 cache is marked as free. So on AMD systems, there is *never* any common data between L1 and L2 caches.
This is a bit more complex to keep track of, but it essentially frees up 128KB in the L2 cache (the L1 cache is 128KB)
Don't know if that was the difference you were thinking of

Close.
The processor communicates directly with the memory controller (which is where Intel uses the FSB)
On the other hand, AMD uses HyperTransport to connect the actual memory to the memory controller.

 

According to a sysadmin friend, AMD's Opterons are far better than Intel's Xeons for multi-cpu systems because the Hypertransport and FSB are also used for comunicating between CPUs and Intel's FSB can't handle the speeds that the Opterons can in this respect.

This is based on a practical test of upgrading Dual Core Xeons to Dual Core Opterons in a multi-cpu server designed for virtualisation.

That said, Intel's Xeons are reportedly better for scientific and research work as they are better (quicker) at raw calculations.