Help! Memory/FSB experts needed.

[Tempered81]

just want to get the max out of my 2x1gb supertalent ddr2 800mhz ram

right now its 5 5 5 15 at 448.8mhz with a 2T command rate, can someone offer me a faster performing setup assuming that the mem is capable of it?

example. 3 3 3 8 at 350mhz with 1T command rate. would that be faster?

 

[nefariouscaine]

good sir - if you get those sticks to run at 350 at 3338 T1 you'd IMO be faster - question what voltage are you running them on as you might be able to push them further

I honestly can't speak to how the P965 behaves nor the memory timings settings it has so anything I say would be opinion on speeds - my next question is what is the fsb you want to run at?

 

[Tempered81]

well in sig, my fsb is 359mhz with a 9x multiplier 1.38 on the cpu vcore. and i upped the voltage on the memory by .3 i believe in bios, because my cpu guide reccommended that. Pci-e, northbridge, were upped by .1 and the cpu by like .08.

the memory runs at 448 right now. but i just would like to know in regards to memory what factors create the biggest performance gain?

should i be running a 1:1 ratio with the FSB, as i hear a lot of talk about this

should i go for tight timings sacraficing overall memory mhz?

should i go for a higher fsb, with a lower multiplier, say 400 x 8? - this for one, i have read is difficult with a 4300 to achieve fsb's of over 380 is this true?

how high should my cpu vcore be?

so as you can see, im basically a newb who has followed an overclocking guide, but would like some expert advice.

 

[nefariouscaine]

Honestly I'm a little unsure on e4300 settings as its been months since I toyed with one at all - I had issues on getting mine above 3ghz and then got a e6400 then my e6600 quickly to follow - you vcore don't seem high but your cooling isn't mentioned at all and IIRC your memory is running at 2.1v

I'm not sure what dividers you have or how easy it is to reset the bios on that board of yours but I would play around with it to within a minor area - you aren't likely to hurt much by attempting the 400x8 as you suggested and IMO would only need a bios reset if it didn't like it - as it stands you should worry about cpu clocks more than ram frequency - 200mhz on your cpu will yield higher performance than 50mhz on ram

I've also seen areas where 1:1 is mentioned to be over rated - i say pull out the calculator, pen & paper and then sit down to play with your cpu speed and memory ratio's. Look for the highest stable cpu speed with a divider that gives you DDR2 900mhz or less as you know the ram shouldn't be an issue at that point

 

[BonzaiDuck]

Nefarious gives good observations here.

I can only note what I've put in another thread.

There may be a tighter set of latencies that are stable for the SuperTalents at their rated DDR2-800 speed. I think you understand (from your OP), that running them at a higher speed (such as 930 Mhz) means loosening the timings and increasing the voltage.

IF (capital-I-F) you can tighten the latencies and maintain stability at 350 Mhz x 2 or (DDR)= 700 Mhz, you might be able to tweak further to get a 1T command-rate. You can tighten tRC to the limit of tRP+tRAS.

So if synthetic benchmarks are a valid basis of comparison, both I and Nefarious have shown that you can get at least the same bandwidth (or close to it) at lower FSB speeds and tighter latencies, compared to the rated (or higher) speed and looser timings.

Here's the screenie I posted in the thread about "1:1" CPU/DDR ratio:

Crucial PC28000's 1:1, 352 Mhz with 3,3,3,6,1T,tRC=9

What it "MEANS" is subject to debate. If Tom's Hardware Guide compares several modules using Everest memory benchmarks as well as SiSoft Sandra's, it's a basis for comparison. Personally, I think the performance is stellar, and I've begun testing just to see how it compares to speeds of DDR=800 Mhz, etc. I don't get any noticeably stunning results so far at dividers other than 1:1. To me, it seems to be a wash.

Nefarious is RIGHT that you can drop your multiplier and reach 1:1 @ 400 Mhz -- or near that, depending on hardware. But the latencies will need to be looser than what you can achieve at lower FSB settings. He's also right about the possibility of using a different divider ratio, but again -- looser latencies will be called for as well.

 

[Tempered81]

my clear cmos is connected to the case reset switch The memory divider is 4:5 & yes, mem runs at 2.1 volts. Thanks coke and duck, it seems that loose timings at a higher speed, and tight timings at a lower speed return similar performance results. Am i right to assume that? maybe ill shoot for the 1:1. i cant boot into 8x400 though, damnit. I read that a lot of 4300s cant boot above 380. my board will supposedly do 505mhz FSB.

so, suppose that you guys had these components, how would YOU overclock them?:

1 Core2duo e4300 1.8ghz 800mhz L2 (capable of 3232mhz @ 1.380 vcore, currently at 359 x 9)
(i have posted at 376mhz at around 3.5 ghz, but would always freeze up booting into windows. I keep speedstep on)

2 x 1GB ddr2 Ram supertalent pc6400 800mhz @ 2.1 volts (spd:200 3 3 3 8 - 11 2T, 333 4 4 4 10 - 14 2T, 400 5 5 5 15 - 21 2T)

1 Gigabyte GA 965-ds3 revision 3.3 bios f12 motherboard (capable of 500mhz+ fsb)

1 8800GT 512mb BFG oc edition 70mm fan Pci-e x16

 

[BonzaiDuck]

I haven't tried any of the newer models of C2D other than an E6600.

But both my E6600 and Q6600 (C2Q) are spec'd at 1,066 Mhz FSB -- or 266 CPU_FSB (or what we used to call "external frequency").

Your E4300 is spec'd at 800 Mhz FSB. As much as forum members may want to be cavalier about processor-model specs or what they "mean," I'd think that taking an 800FSB processor to 1,333FSB -- going from 200 Mhz to 333 Mhz -- is a leap by itself.

But I suppose you'd also be looking at what other members have posted as stable results.

In other posts -- another thread, for instance -- I cite Everest benchmarks in a Tom's Hardware article from early this year on memory modules, juxtaposed with my results with Everest. I think these benchmarks reflect a combination of memory and processor-cache performance. That might bring in a consideration of "processor efficiency" benches using Sisoft Sandra -- an indicator of matching cache and RAM performance.

Without working with the E4300, I'd suggest a strategy more in line with its stock spec of 800 Mhz FSB (200 Mhz), so that you over-clock it mildly to (maybe) 333, and with good memory modules at 1:1, tighten the latency timings. With DDR2-800s, you still have an option of running a different divider.

With little experience per that processor and given the specs, it seems to me that 1.8 Ghz to 3.2+ is quite a leap. But if you're doing it with a VCORE of only 1.38V, even given the price of the processor itself, why push your luck? Even with my E6600, I had to go up as high as 1.44 to 1.46V just to get from 2.4Ghz to 3.3 or 3.4.

For the stepping/revisions of my E6600 and Q6600, I've decided not to push VCORE much beyond 1.41V. That's just my own decision . . .

 

[Tempered81]

yeah i think my chip is good too, i've never put the vcore over 1.4 cuz i heard that was bad. but also it runs hot as hell, if i do intel TAT or prime it overheats and crashes
but for games & everyday use, its fine. So given that its at 3232mhz at 1.38, thats not prime/memtest stable. it runs superpi 1mb in 18 seconds though 1 time i swear i got 17

 

[BonzaiDuck]

Well, you may have misunderstood me.

The heat is a linear function of speed and a parabolic or exponential function of voltage. So even if your voltage is not far from "default," your over-clock is closer to a 100% increase than it is to a 50% increase.

Of course, these processors may exhibit different characteristics now than did processors three years ago. I thought that I understood from forums and friends that a 50% over-clock to certain AMD Athlon processors was decently good. And at that time, a 25% over-clock to an Intel processor was what you might expect without running the voltage too high or putting stress on the Northbridge.

We're in the middle now of a spate of motherboards that actually anticipates compatible processors that run at a higher FSB. You could have such a motherboard-- I'm not that familiar with Gigabyte products -- but if it uses a 965 chipset, it may not be among those I speak of here.

The heat may be more of an indication that you've gone too far, and not so much of a reason that you can't go farther, or that your stress-tests are failing.

Following my own strategy, benchtest the default settings for your processor and memory and create a log-file of 8-second-interval samples of Core-Temp readings at PRIME95 load. The averages would give you a baseline for each core.

Then I would methodically over-clock the processor to the point where my stress-tests are showing no more than 5 to 10C above those averages, while stress-testing for voltage settings that remain rock-stable for six hours or more. At that point, I'd attempt to hold that particular over-clock, and focus on tuning the memory for better bandwidth.

Have you read reviews of boards that use the 965 chipset? What are the speeds anticipated for that board? Does it provide for CPUs with stock FSB's of 1,333 Mhz? Or is it a board spec'd at only 1,066? If the latter, then a 266 Mhz clock increase is not within the board's official spec. But a 25% over-clock to both the processor and the board is probably "reasonable."

It was only two years ago that I "blew" a motherboard -- its chipset -- with a full, 25% over-clock.

 

[Tempered81]

think its 1066 spec
but cpuz says like 1430

 

[BonzaiDuck]

If the manufacturer says 1066, it's 1066. Are you sure CPU-Z is not reporting your current over-clocked FSB value?

I'm sure you can "go farther" with your CPU than I can go with mine. I've still got a B3-stepping of the Q6600. But even a G0-stepping of the Q6600 pretty much tops out at a reasonable over-clock around 3.4+. That's what OTHERS are doing with theirs. It is still less than a 50% OC above stock.

But even if the G0 voltages are close to spec or "default" values, the boards most people are using are anticipating release of processors with 1,600 FSB, and either spec'd at 1,333 or 1,333 and 1,600.

My top-end over-clock on the B3-stepping was only 3.20 Ghz. But I've dropped it back to 3.16 because it's the best way to capture memory bandwidth with a 1T command-rate setting -- IF I continue to tune the system at a 1:1 CPU/RAM ratio. Bandwidth is better with DDR= 704 Mhz and 1T than at 712 and 2T, and I can't set a 1T command rate at 712 without increasing memory voltage above spec.

If I run a divider with the 3.2 setting, I can get the memory to run 7:8 with 4,4,4,8 timings, but the bandwidth at 800+ Mhz doesn't look any better than it does at 704 and 3,3,3,6.

And for everyday use, I'm still debating whether to just leave the system at 3.0 GHz 25% over-clock, or use default settings. I might be more inclined to kill the processor, since I'm planning to replace it with either a G0 or a Penryn, but having to do that before I want to pay for a new processor will be too inconvenient.

Remember that I'm starting from a point where default CPU speed is 2.4 Ghz and 1,066 FSB, and a mobo spec'd to handle processors with default FSB of 1,333. You're starting from 1.8 Ghz with 800 FSB and a mobo that in all likelihood is spec'd at 1066. At FSB 1,333, you will have OC'd the processor by 66%, and the motherboard by 25%. You may be inclined to let the processor die early and buy another. But at least there's less a chance of destroying the motherboard with that level of over-clock.