dinkster9

Distinguished
Jun 27, 2001
314
0
18,780
Okay, i have an asus a7n8x dx 2.0 board with 512 twin x corsair 256x2 llpt ram. Currently i'm running my barton 2500+ at 200x11 so the ram is running at 400mhz with 2,2,2,6 timings. My question is that the ram isn't letting me get much over 400mhz and i was thinking about going up to a gig of memory anyway. But most the ram over pc3200 has pretty bad timings. I'm running windows 2000.

Anyway, so i have 256x2 duel channel pc3200 ram (2,2,2,6)
I'm thinking of getting something like 512x2 duel channel pc3500-pc4000 from corsair, but their timings are NOT as good as the ll i have, they don't make their ll (low latency series) in any speed past pc3200.

So my question is do you think an extra few megaherts out of my fsb (say 212mhz or so) and an extra 512 of ram would be faster performing at higher timings, or the 512 i have now at Super FAST timings running at 200mhz.

I know its a strange question, but if i'm going to plunk out 300 more bucks for ram i at least don't want a performance DROP.

So more ram, higher timings, faster fsb (worse?, better?, same?) as my current setup. I'm running at 2.2ghz, i'm thinking about 2.4 is possible with my rig, anyone who has tried this, do you notice a difference at slower ram settings with pc3500 or higher ram compaired to the pc3200ll ram?

"sixth sick sheik's sixth sheep's sick"
*grabs a stick and places it into the flames; mmmm, smores*
%Think before you act would be a good motto%
 

dinkster9

Distinguished
Jun 27, 2001
314
0
18,780
512 ram running at 2,2,2,6 at 400mhz (200mhz fsb x2)

or

1024 ram running at 2.5,3,3,7 at 430 (215mhz fsb x2)

Any ideas which one would produce better ram performance and overall system performance?

"sixth sick sheik's sixth sheep's sick"
*grabs a stick and places it into the flames; mmmm, smores*
%Think before you act would be a good motto%
 

lunitic

Distinguished
Aug 6, 2003
214
0
18,680
Nope. Could be either. I doubt whether the difference is noticeable. Both somewhere near +3% in memory benchmarking, both unmeasurable small in real-life applications. There are just too many side-effects (for example, just increasing your memory from 512MB to 1024MB may *decrease* your memory performance because 1) the MMU overhead increases and 2) the cache efficiency decreases (relatively less cache to memory, therefore more chance on cache misses). You won't see that effect in benchmarking software because that is specifically written to eliminate such issues.
 

dinkster9

Distinguished
Jun 27, 2001
314
0
18,780
ya, i guess its a tough call. I've heard alot about how windows 2000 and xp only really take off after they have at least 1gig of memory to play around with. I'm not sure why that is, but i'm sure its true. I can assume alot of its due to the swap file, but with 512megs, my swap file isn't getting "overused" all that much. I've also heard some horror stores with duel channel 512x2 mem sticks on the a7n8x, has this been fixed with bios stuff or is it still around?

<font color=purple>A7n8xDX2.0|B2500+@200x11=3200+(2.2ghz)(1.775v)Corsair512twinxpc3200llpt(2.8v)R9700P(stock)160Gx2WD7200RPM 8MB HD|Enermax460W|T:43C@N,50C@FL,MB 28C|3DMark2001 17087|3DMark2003 5134</font color=purple>
 

bum_jcrules

Distinguished
May 12, 2001
2,186
0
19,780
I would think that the yields on a ~6% would be closer to the 6%. I don't think the overhead on the Memory Management Unit will really tax it to any real degree. Also, benchmarks like STREAM do take cache into account. So I don’t know why you said they didn’t.

<A HREF="http:// http://www.cs.virginia.edu/stream/ref.html#runrules" target="_new">“The general rule for STREAM is that each array must be at least 4x the size of the sum of all the last-level caches used in the run, or 1 Million elements -- whichever is larger.”</A>
STREAM Website see “Adjust the Problem Size” for that quote.


Dink,

Read <A HREF="http://www.lostcircuits.com/memory/ddr400/" target="_new">this article over at LostCircuits</A> to get some ideas on overclocking past DDR400. He didn't use your board but you will get to see the results of an overclocked dual channel system and the results of messing around with the latencies, voltage and the base clock signals.

It does show what clock speeds are possible to reach with relaxed clock latency timings.

On <A HREF="http://www.lostcircuits.com/memory/ddr400/10.shtml" target="_new">this page you can see the benchmark results</A> at a 255MHz FSB and a 255MHz memory bus. That is at a 1:1 ratio but like he stated he got better bandwidth results at 300MHz FSB and a 320MHz Memory bus. Those 255MHz numbers are at 2.5,4,4 (tCAS, tRCD, tRP) and the voltage at 2.75V.

Now there is a balance between speed and latency. All I would stress to you is please don't get hung up on any one philosophy like "lowest latency possible", "highest clock speeds possible", or "FSB and memory bus <b>must</b> be at 1:1 ratio".

There are no absolutes blanket statements that govern all systems.

With that being said, you most likely will see better results from the higher clock speeds with the relaxed latencies. The actual amount should be somewhere between 0% and 6% if you only overclock to a 212MHz based signal. I would lean to think it would be higher than 3%. Each system is unique and your mileage may vary.




<A HREF="http://www.stompfest.com" target="_new">Stompfest Sept. 13-14 - Indy. IN</A> - Should be some good gaming!!!
 

pIII_Man

Splendid
Mar 19, 2003
3,815
0
22,780
It is my opinion that anything over pc3500 ram is just factory overclocked memory. If you must purchase new mem, stick with pc3500 and increase the voltage/relax timmings to let it overclock further.

I can guarantee you that at 2,2,3,6 timmings and 2.8v you will be able to overclock your current memory a good deal further.

BTW Bum, great post!


Proud owner of DOS 3.3 :smile:
 

lunitic

Distinguished
Aug 6, 2003
214
0
18,680
6% yield on FSB would yield 6% on memory bandwidth but would yield much less in real life, because the memory bus isn't in use 100% of the time (if only because of the cache)

I'm not sure about this, but a larger memory to TLB ratio would mean a higher chance of a TLB miss and therefore a higher overhead.

That's exactly what I meant to say. In order for a benchmark to measure the memory bandwidth correctly it will have to make sure it's accessing physical memory and not the cache; it does so by making the array 4 times the cache. Therefore by eliminating the cache you won't see the effect of decreased cache efficiency.
 

bum_jcrules

Distinguished
May 12, 2001
2,186
0
19,780
For the benefit of any doubt, I think we are trying to say the same things.

Yes, Cache can increase the ratings in some benches that don't discriminate.

Yes there would be a higher load, but will it impact the MMU in some meaningful way, probably not.

And as for bandwidth, it would show up to ~6% for peak but the sustained would not be some figure less because it is not bursting all the time.

:smile:


<A HREF="http://www.stompfest.com" target="_new">Stompfest Sept. 13-14 - Indy. IN</A> - Should be some good gaming!!!