How to lock down Memory's Bus while increasing only CPU's FSB?

[vaporize]

I am thinking of buying a 2.6C and overclocking it to 230 FSB so it would run at 3.0 GHz. But at 230 FSB my ddr400 ram will run at 460 FSB. Is there anyway to lock the ram's bus speed at 200 while still running the CPU FSB at 230 so that my CPU will run at 920 FSB and my mem will run at ddr400? I dont want to use dram ratios because at 5:4 ratio my ram will run below 200 and the only way i can run my ram at its rated speed is if i overclock to 250 FSB w/ 5:4 ratio, but i dont want to push the FSB to 250. Getting to the point, I want my pc3200 to always run at its max speed (ddr400) while i overclock to 230FSB. Is there a way to accomplish this? -sorry of the long explanation. thanks for any help.

 

[MangoTBG]

To be honest dude, just do the 5:4. I have a 2.4C and it is flying at 250X12 and 5:4 ratio. I have a Zalman 7000 super silent HSF on it and I don't see 50s!!

 

[Lord Evermore]

Unless someone comes up with a hacked BIOS for your board, there's no way to make the memory run at anything other than the specified ratios. I'm not even sure a BIOS can be made to have simple unlocked memory settings with an Intel chipset, as they're very keen on people only using specific ratios and stable memory speeds.

Your processor should be able to handle a 250MHz bus speed just fine, at most maybe a slight voltage increase.

If not, you can test the various ratios and speeds and see what gives you the best performance. Having the FSB at 230MHz with the memory at a 5:4 ratio might be fast enough, but it might be faster to use a 1:1 ratio with only a 215MHz bus speed for example. The matched memory bandwidth to the CPU can make up for the reduced CPU clock speed in many cases. You can then play with the exact settings to get the highest memory speed possible.

 

[oldfart]

The thing is, if you are memory limited, you have to use slow timings if you push the limits of 1:1. I've done a bunch of testing and have found that 1:1 with slow timings runs ~ the same as 5:4 with fast timings. You are better off running a higher FSB/CPU speed with a 5:4 ratio and fast timings.

 

[Lord Evermore]

Yes, that's true, but only up to a point. If the faster FSB is only faster by a few MHz, then the faster timings and CPU speed won't make up for the reduced memory speed. (Of course it is all dependent on whether the stuff one is doing depends on memory bandwidth or latency or CPU speed more.)

It'd be nice to have a comparison of like, whether 10MHz faster FSB/CPU speed with slower memory and fast timings, is better than 10MHz slower FSB/CPU but matching memory speed with slightly slower timings. I mean like 2-2-2-5 compared to 2-3-3-6, or even 2-2-2-6, not drastic changes. Then figure out what the turnaround point is, where the extra FSB/CPU speed does help more than the bandwidth of matched memory speed.

 

[oldfart]

I didn't go that far...but here are some tests I did:

My system can hit one of two walls
1) Mem speed wall. My ram maxes out ~ DDR 426 with slow 2.5-3-3-7 timings. This equals 213 FSB max in this config.

2) CPU speed wall. My 2.6C can go to 257 FSB @ 1.6 Vcore.

This gives me a choice of

1) 213 FSB | 2.77 GHz | 1:1 | DDR426 | 2.5-3-3-7
2) 257 FSB | 3.34 GHz | 5:4 | DDR 412 | 2-2-2-5

LAME MP3 Encode
(lower is better)
257 FSB | 5:4 | 2-2-2-5
Encoded 11 files in 0:02:17

213 FSB | 1:1 | 2.5-3-3-7
Encoded 11 files in 0:02:45

213 FSB | 5:4 | 2-2-2-5
Encoded 11 files in 0:02:46

213 FSB | 3:2 | 2-2-2-5
Encoded 11 files in 0:02:46

Obviously, MP3 encode is all about CPU speed. 1:1, 5:4, 3:2 made no difference at all.

MPEG2 Video encode benches
TMPGEnc AVI -> SVCD CBR (lower is better)

257 FSB | 5:4 | 2-2-2-5
2:10

213 FSB | 1:1 | 2.5-3-3-7
2:35

213 FSB | 5:4 | 2-2-2-5
2:36

213 FSB | 3:2 | 2-2-2-5
2:40

Again, CPU speed is king. Also 1:1 slow timings and 5:4 fast timings are about the same. 3:2 is a little farther behind.

Q3 benches:

FSB | Ratio | cas settings | performance mode | Q3 FPS
257 FSB | 5:4 | 2-2-2-5 | disabled | 390.6
257 FSB | 5:4 | 2-2-2-5 | fast | 391.3

Next, 213 FSB since this is my max 1:1 FSB setting due to the limit of my ram. To get there I have to use slow timings:

FSB | Ratio | cas settings | performance mode | Q3 FPS
213 FSB | 1:1 | 2.5-3-3-7 | disabled | 335.8
213 FSB | 5:4 | 2-2-2-5 | disabled | 336.2
213 FSB | 5:4 | 2-2-2-5 | fast | 338.3

You can see that 5:4 fast timings and 1:1 slow timings are ~ the same.

 

[extro]

Agreed with everyone above. Unfortunately, we don't get a lot of choices in cpu : memory ratios. They're chipset controlled and we're lucky Intel gave us 5:4 and 3:2.

If your goal is to run at 3.0GHz, then you can save some bucks by getting a 2.4C and clocking the bus to 250Mhz with a 5:4 divider, your memory will run at 200Mhz and you could spend the extra bucks on better cooling for the CPU and Northbridge if required. I've read plenty of reports of people running the 2.4C at 3Ghz with the stock Intel sink/fan. I don't think 250Mhz FSB is such a scary number, it doesn't require exotic cooling of the CPU, just be sure the board has a good sink/fan on the NB.

 

[Lord Evermore]

My roommate's 2.4C runs fine at 3GHz, stock cooling.

I can see that with a 570MHz difference in clock speed the slower memory with faster timings is better, but what about if the difference was between using a 1:1 ratio with a 230MHz bus speed but very slow timings (as vaporize did indicate that's as high as he planned to go), or using a 5:4 ratio with a 230 or even 250MHz bus speed but good timings?

1) 215MHz | 2.80GHz | 1:1 | DDR430 | 2.5-3-3-8
2) 250MHz | 3.25GHz | 5:4 | DDR400 | 2-2-2-6

With that, there's only a difference of 425 MHz CPU speed. But there's 30MHz more memory bandwidth with the slower CPU. The faster CPU also has lower memory latency. Will that result in wide enough differences to make the 5:4 ratio with fast timings worth it? Or is the extra bandwidth enough in this instance to make the timings less meaningful? The comparisons oldfart makes are a very wide gap in bus speeds. We need to see what happens at more than just a couple of speeds for each bus and timings, for those who can't get their CPU up quite as fast, or who have better or worse memory that can handle timings differently.

But then, if I had the time and hardware to test all that, I'd be running a site. Or actually I wouldn't, I'd be doing other expensive things.

 

[oldfart]

With that, there's only a difference of 425 MHz CPU speed. But there's 30MHz more memory bandwidth with the slower CPU. The faster CPU also has lower memory latency. Will that result in wide enough differences to make the 5:4 ratio with fast timings worth it?

Yes. I'd say it will be faster.

My testing shows
1:1 2.5-3-3-7 = 5:4 2-2-2-5

Those being equal, all that is left is the gain of 425 MHz and an increase in FSB of 35 MHz with the 5:4 ratio.

Run some benches!

 

[extro]

Overclockers.com has been looking into the dilemma between higher FSB and slower memory speeds and how it impacts real world performance. The testing so far supports oldfart's observation that higher CPU speed, even at the expense of slower memory speed, yields faster overall performance on the 865/875 platform.

"Is 3:2 For You?"

"More On Memory"

 

[Lord Evermore]

Do they take timings into account? Ignoring timings, a small increase in CPU speed at the expense of memory bandwidth may not automatically produce higher performance. There very likely is a range where that doesn't happen, until you reach a particular level of increased CPU speed.

 

[oldfart]

That usually doesn't happen though. With P4C overclocks (typically 2.4C and 2.6C), you will hit a mem speed wall WAY before you hit a CPU speed wall. There is pretty much always a large CPU speed diff with these setups. Also, these is no reason to ignore timings. A lower mem MHz speed can run faster timings. This will compensate for the loss of mem MHz. If you are running a 3.0C, that is not the case.

 

[extro]

Originally posted by: Lord Evermore
Do they take timings into account? Ignoring timings, a small increase in CPU speed at the expense of memory bandwidth may not automatically produce higher performance. There very likely is a range where that doesn't happen, until you reach a particular level of increased CPU speed.

True, a small increase in FSB that required a reduction in memory timings wouldn't be worth it, they'd offset each other and you'd end up with no net gain and only more heat stress on components. But right now the P4C has as much as 70Mhz of head room over the fastest memory modules currently available. The FSB ceiling for the P4C is roughly around 300Mhz, while the best PC3500 hits the wall at around 230Mhz.

When you take advantage of a chunk of that 70Mhz, you render tinkering with memory timings to get over a 5-10Mhz hump irrelevant. It would only matter in cases where someone was intent on holding the CPU/DRAM ratio to 1:1 so they could have PAT enabled; but from all the benchmark tests I've seen I think PAT is mostly hype and can't match the percentage gains delivered by a 25% overclock.