8500cl5d-4gbpk

[Inmani]

Are these settings good or would I be better off not running the FSB 1:1 and clocking the ram higher? It's 2.1v trying to get the tightest timings and bandwidth. My motherboard P43 chip walls at 420FSB won't post higher than that, but it does have ram dividers to run 2:1 and 5:6 I believe something like that.

[Inmani]

So I did a bit of a testing with few different settings. The left side results are CPU 2520MHz, 1:1 divider, 420FSB, 840MH DDR2. The middle results are CPU 2400MHz, 4:5 divider, 400FSB, 1000MHz DDR2. The right side results are CPU 2460MHz, 4:5 divider, 410FSB, 1024MHz DDR2.



The timings for the middle and right side results were the same 5 5 5 15. The left side results they were tightened some to offset the lower memory clock rate to 4 4 4 10 timings. I didn't test any other benchmarks, but do you think I'd be best with the left side results or right side results? Will I have better benchmark scores in other tests by keeping the divider 1:1 or would they still be better with the 4:5 divider based on the memory results? For the most part the results summary would be a net gain of about 450MB/sec to memory reads and 3.2ns lower memory latency.

[GSKILL TECH]

Furthest right is best, if you OC the CPU, the gain would be even better.

[Inmani]

It's OC as high as the FSB will go as it is can only reach 420FSB max on the P43 chipset with the mobo the multiplier was just set to 6 rather than 8.5 for energy efficiency to conserve power, but will handle 420x8.5 easily enough. That would defiantly bump up the results for each accordingly of course.

It would probably do 5 5 5 15 timings 420 FSB with 4:5 divider at 2.3v-2.4v if voltage wasn't limited to 2.2v though still pretty good I guess.

[Inmani]

Some new results tweaked timings with memset plus. I broke the 9GB barrier and below 60ns barrier with DDR2 1032MHz rated speeds 430FSB 5:6 ram ratio. Seems to be stable thus far no crashes and didn't see any errors with Throttle Stop which usually is a good indicator though maybe not 24/7 benchmark stable same voltages 2.1V or actually 2.08V since *shakes fist* my board has a bit of vdroop.



Discovered my P43 can OC a bit higher than 420FSB wall by OC PCIe bus from 100MHz up to 105MHz/440FSB though I have it set at 430FSB (faster ram speeds and even more importantly higher CPU OC) PCIe103 because the on board NIC disappears at PCIe 104MHz.

I might toss in a PCI 100MB NIC I have laying around eventually it should be more tolerant of bus OC since PCIe bus actually gets OC x3 more than PCI bus. In fact on older VIA chip sets PCI adapter SATA/IDE cards were a work around to data corrupt from the chip set OC on them that did the same exact thing OC PCI/AGP/SATA/IDE buses since they are all tied to chip set and didn't have a lock.

Oddly enough is I knew PCIe OC helped on my G31 board, but didn't realize it did for P43 as well since it would only OC from PCIe 100-105MHz. It's doesn't quite offer the same head room though requires less severe PCIe bus OC.

[Inmani]

Added a PCIe NIC I had in another PC and disabled on board NIC can OC PCIe to 105MHz/440FSB now though stable which means even more bandwidth & lower latency not to shabby for a shoddy P43 chip set with stock NB cooler.