5:4 Memory Ratio ?vs? 3:2 Memory Ratio => is there any **REAL** difference ?

[BEIF]

Hi all

=> can some please advise / explain how much of a ?noticeable-real-world-performance-difference? there would be between running the memory ratio at 5:4 ?vs? 3:2 at any given [equal] FSB in an 865PE board ?

For example, assume the (2) following scenarios:-

Option (a) is 275fsb @ 5:4 = memory running at 220 mhz @ CL2

Option (b) is 275fsb @ 3:2 = memory running at 183 mhz @ CL2

So .......... given the above options:-

=> are we talking only ?theoretical-benchmark? differences ?

*** OR ***

=> will my system & all my applications all run NOTICALBY & VISUALLY quicker at 5:4 as opposed to 3:2 ?

Ben " BEIF" Ifin

 

[o1die]

It all depends on your cpu and memory. If you have the 2.4c, running it at 250 fsb and the 5:4 ratio will put your cpu and memory in sync at 400 fsb for optimal performance. You can tell when your system is running well without those artificial benchmarks.

 

[The_Lurker]

There is a noticeable difference. System feels snappier and more repsonsive at least to me and considering your memory is that big a difference too 183 vs 220. I'd say go w/ a 220 if you can of course.

 

[Tom]

My feeling is that the difference comes from the additional speed the memory is running at, not some mystical effect caused by using a different ratio. Naturally running memory at faster speeds is going to make a system faster, at least until the memory is faster than the rest of the system can use.

 

[Jeff7181]

Originally posted by: o1die
It all depends on your cpu and memory. If you have the 2.4c, running it at 250 fsb and the 5:4 ratio will put your cpu and memory in sync at 400 fsb for optimal performance. You can tell when your system is running well without those artificial benchmarks.

Ummmmmm... if you're using a divider for the RAM... your FSB and RAM are NOT running in sync... that's why a RAM divider does.

 

[gramboh]

From some minor benchmarking I've done, there is a large increase in memory bandwidth when going from 3:2 to 5:4, large as in ~15%.

 

[mayhem316]

In my situation I had to lower the memory divider to 3/2 to squeeze the last Mhz out of my 2.4C. With the 4/5 divider I was only able to OC slithly above 3.0 with the memory timings totally relaxed in bios. Obviously my memory was holding me back as the XMS3200LL did not want to push very far past the 400Mhz spec. When I went to 3/2 I was utterly shocked that the IC7 and the 2.4C were completely comfortable at 290 FBS (3500 Mhz) and the memory was almost back to its 400 Mhz spec ( 386 to be exact....a 3.5% underclock). At this speed I can run 2225 latencies which helps bring back some of the speed lost to the underclock. Sandra reports 5620/5611 at this speed, very respectable. Untill faster memory like 500Mhz comes out I will stick to the 3/2 divider and let the cpu rip. All my benches except for the memory tests in sandra (only a 2% decrease) have gone up dramatically with the 290 FBS and the 3/2 divider. I was able to run a few bench marks at 3.6 (300 FBS/200 memory timings) but after several hours of prime95 and hotcpu tester running together a module would fail in hotcpu, so I keep the FBS at 290 and I am perfectly content with the 45% overclock. Hope this helps, good luck.

 

[gramboh]

Wow those are awesome timings and speed on a 2.4, nice job! Well I had a BIOS corruption issue earlier so I didn't get to test out my TwinX1024 much. Here is the fastest memory benchmark I got with my 2 x 512 XMS 3500 C2:

Memory speed: 434 DDR
CPU Speed: 3510 MHz
FSB: 270MHz
Divider: 5:4
Timings: 2-3-3-6
Vcore: 1.63 (reported, 1.6 BIOS)
Vdimm: 2.75
Sisoft: Int Buff: 5819 - Float Buff: 5835

Unfortunetly my 2.6C doesn't seem to run stable over 3.4GHz. It runs 100% solid at stock voltage up to 3.4 but once I got over, even at 1.65V (reported by MBM) and the side panel of the case off with a floor fan blowing several hundred or more CFM directly at the CPU (from a safe distance incase of interference) it still gets errors within a few minutes of Prime95.

For comparisons sake the fastest I got it at 1:1 with the XMS 3500 was:

Memory Speed: 450 DDR
CPU Speed: 2925MHz
FSB: 225MHz
Divider: 1:1
Timings: 2-3-3-7
Vcore: 1.5 default
Vdimm: 2.75
Sisoft: Int Buff: 5287 - Float Buff: 5208

I think the memory might have gone farther but I didn't have time to test, I was already set on exchanging for TwinX since I will be running 5:4 with memory between 200 - 210 DDR anyway.

 

[Duvie]

As mentioned above sometimes the ram divider drop to 3:2 can actually garner more performance by allowing fastr more aggressive memory timings and the bility to overclock slightly further if ram is an limiting factor...In most cases in my testing and around the net at some of the more respectable sites that added raw mhz often makes more difference then high bandwidth and theoretical sandra scores (BS as far as I am concerned)...

As for 15% for difference between 5:4 TO 3:2.....Either you were not running them with same timings or you were like in the 500mhz range I don't see that kind of memory speed making 15%...even in memory intensive apps....

I run 370mhz cas2,2,2,6 versus 450mhz cas2.5,3,3,7 and the difference in most tests was 3-5%.....The differnce was I could run another 5fsb so the ram was 380mhz still cas 2,2,2,6 and not another fsb for the 450mhz and the added 90mhz cpu made up and made test about dead heat...

 

[gramboh]

Sorry, not sure where 15% came from. I compared 2.5-4-4-8 (unrealistic, just curious) to 2-3-3-6 and it was only 3.8%, I think I accidentally compared difference FSB speeds. The difference is so marginal I'm tempted to return my Corsair and get cheap Kingston or something since it costs about half as much.