cel-t sdram bandwith question

Heres a question I have sort of asked before but I still don't quite understand.
In running some theoretical numbers for memory bandwith as it pertains to the Cely-t chips with sdram. Just to keep it simple lets assume that we have a fixed fsb of 133 for all chip clockings this = 1064mhz max memory bandwith if my math is correct. Now comes the question since there are many different multipliers available from the 1.0a to the new 1.4a you can have a range of cpu speeds from 1330mhz on a 1.0a to a whopping 1862mhz with a 1.4a assumming you can actually get the chip to run at those speeds. If you divide the fixed memory speed of 1064/the various cpu clock speeds you get a bandwith efficiency ranging from around 80% with a 1.0a down to about 57% with a 1.4a chip.
At what point does all of the CPU speed go to waist because of the memory bottleneck is that not why Amd and Intel p4 chips use ddr or rdram to address this memory limitation.
I really do not see the point of extreme oc'ing beyond the memory limitations but to be honest I do not know where the point of no return is. Other than bragging rights when does it stop making sense when it comes to memory vs cpu speeds?

I think this is an interesting question: when does extra MHz on the cpu become useless due to a slow bus speed. You would probably have to get a specialised benchmark to see this, and it would depend a lot on the cache of the cpu (the bigger the cache, the less the cpu has to pull from main memory)

However, it would be interesting to see a cele 1.4 @ 100bus vs a p3 700 or so on 133 bus... same cpu (more or less) but diff bus speed.

I would imagine computation heavy tasks would win on the 1.4, but data intensive tasks would win on the 700... then again IANAEE (I am not an electrical engineer)

~ShiFtY out~

davefred99 said:

Heres a question I have sort of asked before but I still don't quite understand.
In running some theoretical numbers for memory bandwith as it pertains to the Cely-t chips with sdram. Just to keep it simple lets assume that we have a fixed fsb of 133 for all chip clockings this = 1064mhz max memory bandwith if my math is correct. Now comes the question since there are many different multipliers available from the 1.0a to the new 1.4a you can have a range of cpu speeds from 1330mhz on a 1.0a to a whopping 1862mhz with a 1.4a assumming you can actually get the chip to run at those speeds. If you divide the fixed memory speed of 1064/the various cpu clock speeds you get a bandwith efficiency ranging from around 80% with a 1.0a down to about 57% with a 1.4a chip.
At what point does all of the CPU speed go to waist because of the memory bottleneck is that not why Amd and Intel p4 chips use ddr or rdram to address this memory limitation.
I really do not see the point of extreme oc'ing beyond the memory limitations but to be honest I do not know where the point of no return is. Other than bragging rights when does it stop making sense when it comes to memory vs cpu speeds?

Click to expand...

Not everything goes through the memory, just like not everything stays in theCPU. Some tasks are simply FSB dependant, others memory dependant, and some CPU/cache dependant only.

Have any of you guys ever seen any articles about this subject. I am sure I am not the only person to ask this question?

Not everything goes through the memory, just like not everything stays in theCPU. Some tasks are simply FSB dependant, others memory dependant, and some CPU/cache dependant only.

Click to expand...

Just like ol' man said, dave, if you run an FSB-dependant program, mem speed and cache aren't gonna help you much. There probably is a speed at which you would reach the limitations of your mem...but unless you are running a program specifically dependant on mem, you would not know it. Some programs run the same on my comp as they do on my g/f's rig, even though I'm 600MHz faster than her (we have the same amount of mem on the same OS)...but for the most part the apps I run on my rig would crawl on hers

well davefred99 it looks like you have 2 nice rigs to comparison test with. Run the same set of benchmarks on both rigs in your sig then underclock or overclock as necessary until you get some initial numbers then drop some results and screenshots in here and some of us will run them too.

We can even underclock to get a wide variety of benchies at specified FSB speeds.

I would be happy to run some benches, I just don't know what to run. Sisoft Sandra is not much good for that kind of comparison and i dont have any others to run that. What would be good for that kind of test.
P.S. I just installed a 1.1a in my Vh6t today and sofar looks good compared to my 1.3a. Currently running @ 1562 (142fsb) out of the gate no burnin yet. I may have some memory Stbility problems my core volts are at 1.725