Confused with 2Gig kits!!

Associate
OP
Joined
6 Oct 2004
Posts
677
Location
Widnes, Cheshire
Well just got my zx and it booted fine after a bios reset, running great @ 210/16 2-2-3-5 stock voltage soooo happy with it and my game load times are almost instant running X3 @1680x1050 all bells and whistles on and it runs so much better with 2Gig.
I went for the zx as i dont need the high htt..... glad i did too :D
Just did the 32M superpi 23m 19s now that means very little to me ......Is it good/bad/average???????
 
Last edited:
Associate
Joined
9 Jan 2006
Posts
1,375
It certainly tests memory to a certain extent as I took 3 mins off my Superpi 32m time by optimising the ram. Having said that I still can't break 29 minutes yet. Maybe i'll try again tonight at 2.7ghz.
 
Associate
Joined
20 Dec 2005
Posts
1,930
Bri62 said:
Well just got my zx and it booted fine after a bios reset, running great @ 210/16 2-2-3-5 stock voltage soooo happy with it and my game load times are almost instant running X3 @1680x1050 all bells and whistles on and it runs so much better with 2Gig.
I went for the zx as i dont need the high htt..... glad i did too :D
Just did the 32M superpi 23m 19s now that means very little to me ......Is it good/bad/average???????


That's kinda slow when you have a CPU running phase @ 3360Mhz. At 3000Mhz my 32M time is only 1m 08s slower @ 24m 27s. You could shave off two minutes if your RAM were running 300Mhz. Benches scale better with increased HTT as long as you can keep the timings fairly tight.
 
Soldato
Joined
18 Dec 2004
Posts
6,568
Location
London/Kent
Dirk said:
That's kinda slow when you have a CPU running phase @ 3360Mhz. At 3000Mhz my 32M time is only 1m 08s slower @ 24m 27s. You could shave off two minutes if your RAM were running 300Mhz. Benches scale better with increased HTT as long as you can keep the timings fairly tight.
I don't think that is a fair statement to make. Bandwidth has been show to make virtually no difference on an A64 seeing as the memory is accessed directly from the CPU's on-die memory controller. Also, I fail to see how bandwidth of RAM plays such a huge issue with SuperPI which is known to favour timings (as games also show).

I feel the time could be saved by upping the FSB. The FSB WILL make a difference, the RAM frequency will not, especially given the loose timings required to run @ 300mhz. You might as well run 300Mhz on a 6:5 divider (180mhz?) and this will run the RAM @ 250mhz using the stock timings. Then again, it has been shown that running 250mhz 3-4-4-8 1T is no different from 200mhz 2-3-2-5 1T so in all fairness, RAM has very little to do with any improvements to be made.
 
Associate
Joined
20 Dec 2005
Posts
1,930
The FSB WILL make a difference, the RAM frequency will not


I have run identical CPU speeds at different settings i,e 250x12 or 300x10 and RAM benches have scaled beautifully with RAM frequency. Until i could go 1:1 and raise the frequency to 300Mhz i was using 250, either 1:1 with a 12xMulti or 300x10 with a 5/6 divider. The benches were better for the lower HTT and 1:1. They improve a lot the closer you can go to 300Mhz.

I tested Superpi 1M every 5Mhz bump in HTT from 250Mhz to 300Mhz and it was a linear improvement the higher the RAM frequency got. Even after loosening the timings at 290Mhz the next run was still an improvement. Adjusting some pretty insignificant timings gained me 10Mhz, stable at 300Mhz, with benches that were better than any combination of low/high RAM Frequency, loose/tight timings.
 
Soldato
Joined
18 Dec 2004
Posts
6,568
Location
London/Kent
I can't agree with your findings then, as this has been extensively tested as well as being theoretically sound. How a lower HTT with 1:1 was faster than a higher HTT is anyone's guess unless you had to dip below 1000Mhz HTT bus speed due to multi's in which case you cannot conclusively state your point as not all the variables were kept constant.

RAM is purely about bandwidth vs commands per clock and refreshes within that given clock. High frequency on the RAM only means greater information can be obtained per unit time and this can be matched by having more commands executed for a lower amount of information (in a very basic sense).
 
Last edited:
Permabanned
Joined
16 Sep 2003
Posts
2,901
Location
Plugged into the Matrix
smids said:
I can't agree with your findings then, as this has been extensively tested as well as being theoretically sound. How a lower HTT with 1:1 was faster than a higher HTT is anyone's guess unless you had to dip below 1000Mhz HTT bus speed due to multi's in which case you cannot conclusively state your point as not all the variables were kept constant.

RAM is purely about bandwidth vs commands per clock and refreshes within that given clock. High frequency on the RAM only means greater information can be obtained per unit time and this can be matched by having more commands executed for a lower amount of information (in a very basic sense).

Im with you all the way on this Smids. ;)
 
Associate
Joined
9 Jan 2006
Posts
1,375
I think your reasoning is sound also, but then i am not entirely sure what Dirk claims in his second paragraph. If he was using the same Multi (therefore increasing his CPU speed each time) as he went from 250 to 300 then altering the timings at 285 to get to 290 might well leave him with a better overall performance as the extra 0.05ghz to CPU and 5mhz to ram probably would outway the increase in timings from 3-3-3-8 to 3-4-4-10 (i am just guessing that this is what you did roughly).

I think you might both actually be right.
 
Back
Top Bottom