XP @ 166

[Skirmash]

Hello,

I have no real desire to overclock a new XP1800 as such - I cant afford the additional noise from the heavy duty HSF that might be needed to cope with additional heat. I use my PC on music projects ! But I would like to make full use of the KT333 chipset and run with a 166 FSB synchronously with DDR2700. This would seem to give performance benefits without heat problems.

My understanding is that I then need to unlock the CPU multiplier using THG's method demonstrated on video, so that I can lower the multiplier (cos I am increasing the FSB from 133 to 166) so that the CPU is still close to its core speed.

If I can achieve this, then pesumably I should not be causing any stability problems - as long as the motherboard allows me to set the PCI speed so I can keep it running at 33.

My questions are

1) is my general understanding about needing to unlock the multiplier to lower it true ?
2) is it true that by keeping the cpu core at roughly the same speed using 166FSB and lower multiplier I would not be introducing any inherent instability or additional heat problems ?
3) is it true that if I am to maintain stability using my PCI cards, then I will need to keep the PCI bus at 33. Does anyone know if the Asus board based on the KT333, or any of the others, permit you to set the PCI bus at 33 regardless of how you set the FSB ? I know the IWIll does.

thanks in advance,

Skirmy

 

[FatBurger]

1) I don't remember

2) No, that's not true. It should work, but that doesn't mean it will.

3) The only board that actually lock down the PCI bus so you can move the FSB freely are the P4 boards from Abit. A 1/5 divider is all that is needed to run 166MHz FSB with the PCI @ 33MHz.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

 

[bront]

1) Yes, unless your MB has ways around it. Some let you set it in the Bios or on the MB. Not sure how reliable that is without unlocking though.

2) That should be the case.

3) You wouldn't be running too much out of spec, so you might be fine with an OCed PCI bus.

Bad trolls Bad trolls... Whacha gonna do... Whacha gonna do when they post here too...

 

[Matisaro]

1) Yes.
2) There isa .00001% chance it will not be stable at that fsb, however if the core is at spec it is virtually guarenteed to work properly.
3) The 333mhz fsb boards should have diviers to keep pci/agp in spec.

"The Cash Left In My Pocket,The BEST Benchmark"

 

[Raystonn]

Note that you will not have as much processing power per FSB clock if you lower your multiplier. Thus, the efficiency of the use of your memory bandwidth will go down. To summarize, if you increase your FSB by 25%, yet lower your CPU multiplier to compensate, you will not see a 25% increase in actual effective memory bandwidth. It will be somewhat lower.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[eden]

No but note that if he doesn't plan to OC his CPU's MHZ, staying at 1.53GHZ with a bigger FSB and RAM will result in a good 10-20% increase in overall performance, so yes it's worth it if you don't want to OC. Same as a NW 2.2GHZ, but upped to 533MHZ FSB and PC1066RDRAM, to see how much you gain from that at same clock. My guess is a lot for P4, so it's worth it on later higher clocked NWs since they don't OC as well as the 1.6A.

--
For the first time, Hookers are hooked on Phonics!!

 

[Guest]

Epox 8KHA+ (with a beta version of BIOS), EPOX 8K3A+ and Abit KR7A can automatically chance de PCI divider to 1/5, 1/6 and 1/7 (only for Abit).
This is observed in here: http://www.aoaforums.com/forum/showthread.php?s=&threadid=1284
and http://www.amdmb.com/vb/showthread.php?threadid=83654

 

[ath0mps0]

Note that you will not have as much processing power per FSB clock if you lower your multiplier. Thus, the efficiency of the use of your memory bandwidth will go down. To summarize, if you increase your FSB by 25%, yet lower your CPU multiplier to compensate, you will not see a 25% increase in actual effective memory bandwidth. It will be somewhat lower.

Ray, for someone who works for Intel, you don't seem to get how this works. While you are technically correct - and this is the same with the P4 - you left out the most important issue - TOTAL SYSTEM PERFORMANCE. If someone spends the extra money on the faster RAM and system board, they aren't too worried if they are using the memory bandwidth to its maximum efficiency. They want to know how much performance increase they will see. Having more memory bandwidth available than you need is what you want. A problem only occurs when you don't get any/enough performance increase for the extra money you spend.

If you leave the FSB at a set speed - 133/266MHz in this case - and increase the clock multiplier, the overall performance increase of the system is less for each subsequent increase due to the fact that the memory bandwidth to CPU frequency ratio decreases. This is already proven fact via benchmarks for both AXP and P4a. When you increase the CPU MHz by only increasing the multiplier, you get less and less increase in performance as you increase the clock after you cross the efficiency threshold. When you reach the efficiency threshold for 133/266MHz, increasing the FSB to 166/333Mhz will yield very little performance increase as the CPU doesn't need the extra memory bandwith except with specific tasks. This is the point at which you are least efficiently using your RAM.

The trick is to find a "sweet spot" and stay above it with your FSB and multiplier. It is true that the higher your FSB is above the "sweet spot" the less actually performance increase you'll see when you raise the FSB speed. After you pass the sweet spot, though, you need to increase your FSB to maintain a more linear performance curve with each successive jump in CPU clock speed.

Intel has already demonstrated that the threshold for the P4 is around 2.2GHZ (22x multiplier) - the memory bandwidth utilization of these systems is approaching 100% of usable bandwidth at 100/400Mhz. Increasing the FSB to 133/533Mhz yields substantial performance increases for the 2400 and higher parts.

I would venture an educated guess that for the AXP this threshold is close to the 12-13x multiplier range (or 2000-2100+) as the AXP is far more memory efficient than the P4. Some may claim that the AXP is just memory starved, but this is refuted by Anand's latest tests that show only a maximum 6% increase in performance when the FSB is increased to 166/333MHz on the 2000+(1666Mhz). The performance gap between the 133/266MHz and 166/333MHz systems will widen, however, as the CPU clock is increased.

Theoretical Example:

A Tbred AXP 2000+ (1666MHz) with a 133MHz FSB has a clock multiplier of 12 (12 * 133 = 1666). Assuming the threshold is at 12x, if you increase the FSB to 166MHz and lower the mult. to 10x, you maintain the same CPU clock of 1666MHz. As shown by Anand, this doesn't significantly increase real-world performance (at least not too noticeably). But, as you continue to increase the mult. to say, 11-12x, you start to reach the sweet spot and end up with a 2250+ (1833MHz) to 2500+ (2000MHz) CPU clock - and as these speeds increase you will most definitely see the performance difference - just like with the P4.

Based on the above info, I would only advocate - for most users - increasing your FSB to 166/333MHz if your multiplier unlocked AXP will be running at a 2100+ (1733MHz) or higher clock speed. At lower speeds the performance increase is not significant enough - to me - to justify the additional expense. If you will be overclocking, I wish you good luck in getting to the highest stable FSB/CPU clock possible whether or not you unlock the multiplier.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.