Cdogg -- FSB and Memory Speed update

[knporter]

Hey Cdogg, If you remeber I posted a question a while back about Memory Speed and how it affects the FSB speed. After speaking with a few people and some more research, it seems slower RDRAM will not reduce the FSB speed. It just creates more bottle-neck. Have you heard anything more? Hoping to get confirmation on this.

"Great minds discuss ideas, average minds discuss events, small minds discuss people."

--Admiral Hyman G. Rickover

 

[cdogg]

You're exactly right that the memory will not cause the FSB to run slower, just bottleneck it. Take this for example:

P4 2.53GHz/533MHz FSB
Mobo Chipset that supports 533MHz FSB
PC800 RDRAM

With this configuration, PC800 would bottleneck the system because it runs an internal clock of 400MHz. One way to picture it is if the CPU sends a request to memory, it travels along at the 533MHz speed. Once there, it's instruction executes at 400MHz. Now, this might not seem like a bid deal, but if the CPU has a lot of seperate requests to make at one time, then eventually memory will get behind and force the CPU to wait.

In the past, the waiting was because of the FSB. While the FSB is still one of the biggest bottlenecks (even at 533MHz), having slower memory is just taking away what you could have.
~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein

 

[cdogg]

To go along with that, here is a link with benchmarks that include PC1066 and PC800 setups:

http://www17.tomshardware.com/cpu/02q2/020506/p4b-06.html

There are quite a few benchmarks throughout this article (just know that it's more than one page). You want to look at the difference between 2533/133/533 MHz and 2533/133/400 MHz. It appears to be very minimal...
~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein

 

[knporter]

Well it see it as 93% as fast as the 1066 RDRAM. Not a bad improvement. I would say it is worth the extra money. Also, what is the middle number? PCI bus speed? If so, it looks like the Athalon's are right there with Intel. So am to think that FSB speed just runs at a higher clock multipler on the P4 boards? Assuming all my guesses are correct, I would think improvements in the PCI bus speed would really help. "Great minds discuss ideas, average minds discuss events, small minds discuss people."

--Admiral Hyman G. Rickover

 

[cdogg]

Yeah, Intel uses a multiplier of 4 while AMD uses 2. Like you said, a multiplier isn't as efficient as the actual speed. I think current technology found on motherboards today is part of the problem. The length of the path is also another problem. Even if you widen it, it's length decreases overall stability as you bump up the speed (latency also gets worse).

There are solutions in the works to improve it, however. AMD is among the first to attack the problem. Their answer is HyperTransport. With this technology, bus speeds will initially start out at 200MHz and eventually get up to 800MHz, still using the 2x multiplier. More importantly, HyperTransport will increase the width of data bits lowering latency in the process - ranging from 2, 4, 8, 16, or 32 bits wide. 32-bit HyperTransport will actually have a data throughput of 12.8 GB/Sec which is 40 times the 266MB/Sec you see today!!

http://www.hypertransport.org/faqs.html

AMD's Hammer CPU is supposed to use this technology. Not only will it replace the FSB, but will eventually replace PCI - another important bottleneck to eliminate.

I don't know if Intel has announced official plans of their solution to the FSB limitation. They will be severely behind the competition if they don't come up with a counter-attack soon...



~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein