1:1 memory speed or Lower CAS result in faster overall system?

[bluesaint7]

I have a dilemma and hope someone can help me decide which is the best path with the least amount of gambling.


I just purchased a 2.4C + IC7 which i know does 3.3ghz.

My question is which memory setup should i get to get the most performance for the buck. I'm not crunching pennies here, but i also don't believe in spending big $ for little gain.

Would the system be considerably faster if i spend $430 on OCZ 4200EL to run at 1:1 and CAS 3-3-4-7 Or would it be better to run a $300 Kingston HyperX PC3500/equivilent and run at 5:4 and CAS 2-2-2-6

Hope someone can shed the light for me.

Thanks

 

[Boonesmi]

its always better to run the fsb and memory in sync with each other

 

[oldfart]

Would the system be considerably faster if i spend $430 on OCZ 4200EL to run at 1:1 and CAS 3-3-4-7 Or would it be better to run a $300 Kingston HyperX PC3500/equivilent and run at 5:4 and CAS 2-2-2-

Your system would be slower 1:1 3-3-4-7 than 5:4 2-2-2. Its been tested and proven many times. The high latency kills performance. More $$ for less performance!

 

[bluesaint7]

so which Ram combo would be best bang per buck then?

 

[Shimmishim]

well 3.3 ghz is like 275 something mhz on the fsb

5:4 is 220 mhz on the ram

i'd say go with some nice mushkin lvl 2 black pc3500

it can do 2-2-3 on intel chipsets

 

[bluesaint7]

what about running 1:1 at 2.5-3-4-6 compared to 5:4 at 2-2-3-6 running same FSB which would be faster?

 

[Sunny129]

Originally posted by: bluesaint7
so which Ram combo would be best bang per buck then?

i'd also recommend mushkin, but it is hardly the best bang for your buck. mushkin is the best, and it comes at a hefty price. i'd say kingston hyperX is a good bang for the buck. buffalo brand is even cheaper, but it doesn't have the reputation companies like mushkin, corsair, and kingston do.

 

[Nebor]

Well, you'll never hit 3.3ghz w/ a 2.4 @ 1:1. You might be better off going for a 3.2ghz overclock w/ PC4200. Running the processor and memory in synch does a lot for you. And if you've read Anandtech quest for the holy grail of memory, you'd see that timings are really insignificant when compared to any kind of major (10mhz or more) DDR clock increase.

You should get the highest frequency you can, then tighten down the timings as much as you can. But really, timings will only give you a few points in most benchmarks.

Just thinking through the numbers just now, I realized, a 2.8C and PC4000 would be a sweet combo. Running 1:1 all the way up to 3.5ghz.

 

[MajorSlaughter]

Excellent reading from Zroc, sit down with a couple of beers and read through it.

everything you want to know about memory bandwith

Lots of charts for those that don't have the patience to read it ! lol

 

[Nebor]

Originally posted by: MajorSlaughter
Excellent reading from Zroc, sit down with a couple of beers and read through it.

everything you want to know about memory bandwith

Lots of charts for those that don't have the patience to read it ! lol

Which shows us that timings doesn't make NEARLY as big a difference as clock speed. DDR400 w/ CL2.5 > DDR320 w/ CL2.

 

[jiffylube1024]

Originally posted by: MajorSlaughter
Excellent reading from Zroc, sit down with a couple of beers and read through it.

everything you want to know about memory bandwith

Lots of charts for those that don't have the patience to read it ! lol

That was an excellent writeup. It also shows that timings don't rule all on the newer P4 Dual Channel setups.

 

[oldfart]

Timings matter more than DDR speed. High latency is great if all you care about is SiSoft scores. If you care about how your PC performs in actual applications, low latency is more important. You will see that 1:1 high latency and 5:4 low latency scores are very close. You can save some $$ by going the latter route.

Here are some benches pulled from users @ ABXzone: and other sites:

Quake3 800x600 with DDR440 2,5/4/4/8 379,6 fps
Quake3 800x600 with DDR352 2/2/2/5 394,1 fps
-------------------------------------------------------------

Low Latency: Kingston 5:4 @ 2-2-2-5, 250FSB, Turbo & MAM Enabled

XMS4000 DDR500 1:1 3-4-4-8 340.8 FPS
XMS3200 DDR400 5:4 2-3-2-6 338.9 FPS

----------------------------------------------------

XMS4000 DDR500 1:1 3-8-4-4 320 FPS
XMS4000 DDR500 1:1 2.5-7-4-4 338 FPS
XMS3200 DDR400 5:4 2-5-2-2 340.5 FPS

-----------------------------------------------------

3DMark2K1
250 FSB | 1:1 | DDR500 | 3-4-4-8
17979

250 FSB | 5:4 | DDR400 | 2-3-3-5
17964

---------------------------------------------------------------

250/250 (2.5-4-3-7)
SuperPi 1M / 16M seconds 45/1102
Prime95 Bench seconds 66.435
3Dmark 2001 / 2003 12197/3467

250/200 (2-3-3-6)
SuperPi 1M / 16M seconds 46/1096
Prime95 Bench seconds 66.836
3Dmark 2001 / 2003 12304/3461
--------------------------------------------------------------

High Bandwidth: Samsung 1:1 @ 3-4-4-8, 250FSB Turbo & MAM Enabled
SuperPi 2M: 110s

Low Latency: Kingston 5:4 @ 2-2-2-5, 250FSB, Turbo & MAM Enabled
SuperPi 2M: 108s

------------------------------------------------------------
200.5 FSB/200.5 Mem (DDR401) 2.5/7/3/3
QIII- 363.3
Comanche- 49.35
3DMark- 16031

200.5 FSB/160.4 Mem (DDR320.8) 2/6/2/2
QIII- 363.9
Comanche- 49.43
3DMark- 16123

----------------------------------------------

By using 3DMark03 to judge what is my maximum FSB/memory/video setting for my P4 2.8c cpu on my Asus P4C800, I obtained the following results:

1) no name PC3200 (1G); FSB=235; 5:4 ratio; 2-3-3-6 SPD; result = 4375, CPU=730

2) Corsair XMS3700 (1G); FSB=235; 1:1 ratio; 3-4-4-8 SPD; result = 4316, CPU=759

3) Corsair XMS3700 (1G); FSB=250: 1:1 ratio; 3-4-4-8 SPD; result = 4328, CPU=810


---------------------------------------------------------------------------------

 

[Sunny129]

Originally posted by: Nebor
Well, you'll never hit 3.3ghz w/ a 2.4 @ 1:1. You might be better off going for a 3.2ghz overclock w/ PC4200. Running the processor and memory in synch does a lot for you. And if you've read Anandtech quest for the holy grail of memory, you'd see that timings are really insignificant when compared to any kind of major (10mhz or more) DDR clock increase.

You should get the highest frequency you can, then tighten down the timings as much as you can. But really, timings will only give you a few points in most benchmarks.

Just thinking through the numbers just now, I realized, a 2.8C and PC4000 would be a sweet combo. Running 1:1 all the way up to 3.5ghz.

so that PC4200 can't be OCed an extra 25mhz to PC4400 speeds?

 

[Soulkeeper]

i got my P4 system running 5:4 with the fastest timings
this seems to be the best since i would have to run my fsb slower if i did 1:1
and seti definately likes the lower latencies

 

[Sunny129]

Originally posted by: Nebor
Which shows us that timings doesn't make NEARLY as big a difference as clock speed. DDR400 w/ CL2.5 > DDR320 w/ CL2.

Originally posted by: oldfart
Timings matter more than DDR speed. High latency is great if all you care about is SiSoft scores. If you care about how your PC performs in actual applications, low latency is more important. You will see that 1:1 high latency and 5:4 low latency scores are very close. You can save some $$ by going the latter route.

seems like there's a bit of debate over which results in better performance...higher clock and loose timings, or lower clock and tight timings. obviously a stick of PC3200 @ cas3 will be faster than a sitck of PC2100 @ cas2, which would suggest that clock is more important. however, nobody compares PC3200 to PC2100 anymore due to the extreme difference in clocks here. now memory clock may still give an edge over timings when comparing PC3500 to PC4000, but any difference you see in Sandra or any other benchmark won't be felt by the user in any real world situation. and if the user can save some money by getting lower clocked memory (due to the undesired nature or memory prices) and run it with a 5:4 divider w/ lower timings and perform only minimally slower than some higher clocked, more expensive memory, then the benchmark differences will be minimal as well, just as oldfart said. that being said, a minimal difference in benchmarks translates into a negligible difference in real world performance.

the moral of the story...SAVE SOME MONEY WITHOUT SACRIFICING A SIGNIFICANT AMOUNT OF PERFORMANCE!!!:beer:

 

[Nebor]

While oldfarts numbers DO support his case, I question the system from which they're derived. Seems like an AMD setup, and I'm not arguing about low latencies for AMDs, as they clearly are more important there. But when you reach the world of the P4 C, those SMALL clock differences are all multiplied by FOUR. That's what makes the difference...

 

[oldfart]

These were ALL P4 865/875 systems.

Real world testing by users prove that latency is just as important or even more important than DDR speed.

As Sunny129 pointed out, the actual differences are very small. CPU speed is vastly more important to system performance than ram speed or latency. I cringe when I see people spending more money for the ram than CPU and mobo combined! The bang for the $$ ratio is horrible.

 

[jose]

oldfart,

Are you running a 2.8 ?

Right now I'm running 3.34ghz (145fsb, 2.5,3,3,7) 1:1 w/ khx3500 on a p4b533e. w/ win2ksp3.

I'm trying to figure out which cpu to get. I would like to upgrade to a diff system, consisting of :
p4c800e w/ 1 gig (KHX3500) & a 9800pro (currently ti4400).

Would a 2.6 better than a 3.0 ?

Also one of the reasons for upgrading would be to move from win2ksp3 to winxp . Not really sure about this, because win2k runs very well but doesn't support hyperthreading. My friend claims that XP multi-tasks better than 2000. But he went from a 2.8/512/GF3 w/ 120 hd to a 2.8/1gig/9800pro w/ 2 raptors in raid 0.

Appreciate your input.

Regards,
Jose

 

[oldfart]

I run a 2.6C. Look at my rigs for details

 

[joe2004]

There are number of things connected with memory performance, some you don't even see but they are programmed with spd or motherboard adjusts those timings as it goes higher with FSB (sort of divider thing, if you remember that). Some power mobos are going to give you chance to adjust certain things on your own, some won't. That is the reason for memory incompatibility many times.
Timings have huge impact on the performance. I think in the future we shall see more and more memory timings in the Bios and there will be less surprises. We might even see motherboards booting without memory (at least dynamic one) and enabling users to set up memory timings the way they see it fit.

Well, you'll never hit 3.3ghz w/ a 2.4 @ 1:1.

Not true. That is achievable with both OCZ or Geil PC 4300, I've seen it, and my Geil PC 4300 (newegg says 4200)comes awfully close to it as well. It depends how much voltage you can shoot into it. I am at 272 MHz FSB and that on Turbo and Pat enabled on Asus. With just 0.1V I believe I could hit 3.3 GHz but I am not complaining. This is all on air cooling, both memory and CPU.