Wolfdale/Yorkfield overclocking thread...

PangingJr

just installed a Q9300 in EX38-DS4 Rev.1.1 board...

3:4 (Memory multi 2.66Dx. MCH Strapping - FSB 400MHz)



5:6 (Memory multi 2.40Bx. MCH Strapping - FSB 333MHz)

[hobo]eclipse

***NOT OPTIMIZED***

this has been running for 3 hours now, pretty impressed, i thought for sure i would come back to at least one of the p95 instances crashed. what is the best way to test with prime95 nowadays. i use 2 instances, run the torture test and select custom, using one thread for each instance and setting a max FFT of 448 (K). Is this a good way to be stress testing or should i be messing with the affinity? using the 2 instances i get the memory usage up to 3.6-3.8 GB.

thanks panging for the ex38 info, how do you get your alpha timings to default so low? before i thought that was p35 thing but i guess not, currently im using the same memory multi 2.40B. Do you know off hand which settings affect the
performance level the most? maybe i need to try with something other than a 9 CPU multi. Just that im not running the RAM any higher than before but the PL and tRFC keeps getting higher and higher the higher i run the FSB. Has to be
the memory multi/strap effecting this? Or maybe because im running 2x2gb modules?

what bios were you using on the ex38-ds4 panging?

**DOH! I see now, F3 bios**

swimtech

Great overclock hobo[eclipse], and on less than 1.3volts! Wow, that is really encouraging!

Nice volts on that 9300 PangingJr, especially on the EX38-DS4 board.

[hobo]eclipse

Still messin.....





1 hour 10 mins prime95 (as described above) stable
11.833 seconds super pi 1M

Speed Step off
2.40B RAM multi
+0.30V RAM
+0.15V FSB
+0.25V MCH
Load Line calibration on
1.3375 Bios CPU V

seems my RAM is willing to play a bit all memset options are bios auto except the usual 5.5.5.15 and I also set the tRFC to 68 as it defaulted to 74 with it on auto in the bios. Dont believe there is a PL setting in the bios on the ex38-ds4. Only thing that seems odd is the memory read bandwidth in everest. seems a little low doesnt it?

PangingJr

that's because the Performance Level is a bit high.

the RAM module itself effects Performance Level the most.
however, motherboard's chipset, BIOS's Memory Performance Enhance setting, voltage settings, MCH strap and processor, also tends to affect the automatically-set Performance Level.

the "tRD" value (or Static tRead value) in Gigabyte board's BIOS setting = Memory's Performance Level.

the setting below the tRD is tRD Phase Adjustment (or Static tRead Phase Adjustment, or Read Delay Phase Adjust),
it usually cause some booting problem if i change the value in BIOS, so i always left this setting to Auto.

PangingJr

below is the actual setting in BIOS (and values in Memset in Windows)

******** Standard Timing Control ********
CAS Latency Time......................5 ..............(Memset = 5)
DRAM RAS# to CAS# Delay........5 ..............(Memset = 5)
DRAM RAS# Precharge...............5 ..............(Memset = 5)
Precharge delay(tRAS)................15 ..............(Memset = 15)
******** Advanced Timing Control ********
ACT to ACT Delay(tRRD).............4 ..............(Memset = 4)
Rank Write to READ Delay...........4 ..............(Write to Read Delayed in Memset = 12.............. BIOS-set value +8 (for CL5), +7 (for CL4), +6 (for CL3))
Write To Precharge Delay............7 ..............(Memset = 15.............. BIOS-set value +8 (for CL5), +7 (for CL4), +6 (for CL3))
Refresh to ACT Delay.................48 ..............(Memset = 48)
Read to Precharge Delay..............4 ..............(Memset = 6.............. depending on BIOS, in some BIOS's the tRTP value that's set in BIOS will be the same as Memset's)
Static tRead Value.......................7 ..............(Performance Level in Memset = 7)
Static tRead Phase Adjust............Auto
Command Rate (CMD)................Auto

ChaosMinionX

Ugh.... that Q9300 did better than the X3350..... and Newegg is being slow shipping my other chips

PangingJr

now Q9300 prices start dropping quite a bit as far as i know



if i can get the Q9300 up to 3.55-3.60GHz. (or 480 FSB), i should be able to get all 10K MB/s in Read/Write/Copy and maybe below 50.0 ns in Latency.

mkk

My low cost E7200 upgrade seems to settle at 3.6GHz using a modest 1.31 core voltage. With a Thermaltake Ultra-120E that I got cheap it could easily do more but I don't feel like bumping the voltage much just to get a little higher. If anything I'll try 9x400 instead of 9.5x379, but I'm also glad not having to raise any other voltage in the system yet.

This relatively budget board EP35-DS3 from Gigabyte has a peculiar behavior on cold boots that puts the CPU voltage at default until I power cycle once more. Googling indicates it's a known problem and that something similar has ocurred on other early P35 based boards, but solveable through the BIOS. Being an old model board and a fresh CPU I guess I might have to live with it, and I could. Everything else is peach.

PangingJr

the cold boot problems in Gigabyte P35 was in some very early BIOS's i think it was gone (and never came back) for some time now.

my P35-DS4 rev.1.0 is a very good board, very stable and easy to overclock (i don't like the first one and i had to return it once for a new replacement board from local distributor),
i checked the new EP35-DS4 and it's selling for the same price of EX38-DS4, i like to buy an X48-DS4 but noone has them in stock (but X48-DQ6, which is too expensive for a board that i want to use it ).

mkk

Good to hear, I wrote to Gigabyte to keep them interested in updating this BIOS.

PangingJr

there's also a cold boot loop problem, like the system takes several power up and down cycles and then instead of loading the overclocked settings, it loads CPU's default clock speed and default memory multiplier but using all the voltage from the previous settings... in this case, the BIOS might be using a not so right setting, likes memory timings or memory multiplier, or might be voltage settings.

PangingJr

This is what i like to boot the system with and use it daily...

400x7.5 3.0GHz,
CPU's core voltage is set at below Normal,
Memory multiplier 3.0x 400 = DDR2-1200, 5-5-5-15, Performance Level 5, 2GB of RAM.





Another setting that i plan to use with this board is the CPU's default clock speed (2.50GHz 333x7.5),
Memory multiplier 3.20x 333 = DDR2-1066, 5-5-5-15, PL 5, 4GB of RAM,
All voltage settings at Normal,
CPU Enhanced Halt (C1E) [Enabled]
CPU EIST Function [Enabled]
Gigabyte's Dynamic Energy Saver is launched.... CPU's core voltage is at around 1.0875 at most of the time.

PangingJr

the Yorkfield Q9300 was a bit hard to overclock with the EX38-DS4 board at first, and i almost hate this board.
tried to add any of the voltages, either Vcore, MCH or FSB, or sometimes even PCIe and Vdimm to the overclock was a bit hard, because it won't get stable.
so i decided to check on the web just to see who else is having the same problem, then i found this pages The Official Q9450/X3350 Overclocking Thread - Overclockers Forums yesterday.
there is a few guys at the forums who are using the EX38-DS4 also having some problem with a high FSB overclocking on Q9450,
it's at around 454-460 MHz under the EX38-DS4, and anything higher will require a fairly amount of chipset voltages increases,
but for FSB below 450 it won't require any of an additional chipset voltages, and the core voltage use in the overcloking is fairly low as well.

anyway, i now think, i end up liking this board after a few days of playing with it.
i hope it's not too soon to say this now, the EX38-DS4 might be the cheapest X38 board in the market, but it's a good motherboard indeed.
right now, i'm running Q9300 at 435x7.5 with a core voltage of 1.184V (in Windows), which's still below its stock VID, without any other voltages increases, and it's quite stable.

btw, ChaosMinionX, i think i also saw your posts on the forum as well.

ChaosMinionX

Yeah I posted there trying to sort my issue, thought maybe I was missing something.... That X3350 I had needed absurd amounts of voltage to even get into windows with 3.6ghz....

I am waiting to try some new chips.

PangingJr

this (below) Q9300 has managed to be 3.5 GHz stable at very low core voltage and the G33's chipsets voltages use and with a mATX board (with a large vdrop/vdroop in Windows).
it's actually quite amazing to me.


Pretty good Q9300 - XtremeSystems Forums

[hobo]eclipse

well ive got to believe that something is definately wrong with my system, after the incident described in my Daed/deadly PSU thread my L2 cache everest benchmarks have dramatically changed.

heres a comparison of the screenshots in my prior post to screens i just took. i have reinstalled vista and have all the same drivers installed except the logitech one(didnt install)

Bios settings are the same


PREVIOUS





CURRENT




now where my confusion comes in is wheather this is damaged hardware, the DCP latency issue, or both.

notice the last screen and the yellow and red spikes, those are happening during the write phase of the L2 Cache benchmark, the one that has changed the most in terms of MB/s.

This issue has exhausted me as of now and i have tried everything that i can think of to try to get the previous numbers back. might try some games in abit to see about actual performance.

PangingJr

hmm i don't think that's caused by the DPC latency. may be the Vcore setting was too low.

try running the CPU at default clock speed and then run the Everest test again, and see how's it at the L2 Write?

forget about the DPC latency issue just for a while, and try your overclock again, start from 400 FSB and then 450... and run the Everest's each time...
and tell me how the L2 write's score came out.

PangingJr

yesterday morning i booted my computer and the Q9300 450 FSB at 1.30V (around 1.28xV in Windows before full load),
PCIe 100MHz and +0.10V, MCH/FSB at Normal, Memory multi Auto (2.4x and DDR2-1080), 5-5-5-15-AUTO.... 2.1V.
this has already been stable and i have it saved in a profile, i also have another one, the same 450 FSB but with 4:5 saved in CMOS.

anyway, this morning, i want to boot at 460 FSB, and i have to use 1.3625V, PCIe 100MHz, PCIe/FSB/MCH = +0.25V/+0.20V/+0.25V,
i get a BSOD 10 mins after running a 4x SuperPi 32MB, then i end up running these voltage setting with 456 FSB, which i'm running it now.



then, tomorrow i will try again by dropping Vcore down, but adding up Vdimm, and see if i able to run 456 FSB at a lower Vcore than this.

the memory multi and timing settings are also important to your overclocks, as for your CPU (and rest of 1333MHz CPUs) you can keep it at 2.4Bx (or 2.0Bx), and memory timing 5-5-5-15-AUTO...
i never use the same RAM that you're using, the tRD value in BIOS may require to be left AUTO on some FSB overclocking. this needs to run a system stable check.

the rest is all about your voltage settings. with this board, it has some vdrop, which i think sometimes add Vdimm and the overclock just worked out okay, however, you will need to give your CPU (and chipset) some volts, you may not have to add a lot of it to the chipset since you're using a C2D.

i find that every time i stick with the idea of keeping the Vcore, Vdimm, and rest of voltage settings as low as possible, then also want the overclock to be stable and all things come out right, then it'll harder than just gave plenty of voltages to the system and then bring them down after.

PangingJr

for people looking to buy Q9550....

DailyTech - Intel to Cut Prices on Existing CPUs and Introduce New Parts in Q3

Intel puts more pressure on AMD with price cuts and new parts

Two things that drive the CPU world are price and performance. Depending on the needs of the consumer one or the other of these two factors is the main driver behind what processor is purchased. For many AMD CPUs the only place they compete currently in on price and Intel is starting to whittle away at the price difference.

Recent roadmaps reveal the company has another cut coming in Q3 2008. Simultaneously, Intel will release new quad-core and dual-core processors. This includes the 3.0 GHz Core 2 Quad Q9650 with a price of $530 in 1000 unit trays. The current Core 2 Quad Q9550 will get a significant price cut from its current $530 to $316.

A few of the older Intel CPUs will be phased out including the Core 2 Quad 9450. The 9450 will be replaced with the Core 2 Quad running at 2.66GHz at $266. Other quad-core processors being phased out include the Q9300 and Q6700 leaving the Q6600 as the only 65nm Intel CPU on the market selling for $203.

Intel will also make some changes in its dual-core lineup. The company will announce its Core 2 Duo E8600 3.33 GHz processor for $266, and discontinue the Core 2 Duo E8300. The E8500 and E8400 dual-core parts will drop to $183 and $163 respectively. A new entry-level Core 2 Duo part will debut called the E7300 running 2.66GHz for $133. The existing E7200 will get a price reduction to $113.

PangingJr

P5Q D and Q9300

this P45-based P5Q-D can really clock this Q9300 Yorlfield quite well and use a quite low chipset volts as well.

Common Performance Level 8
CPU Voltage: 1.2875V
CPU PLL Voltage: 1.58
FSB Termination Voltage: 1.28
DRAM Voltage: 2.16
NB Voltage: 1.28
All CPU and NB GTL Ref set to: .655x




(this is the 1st attempt on maintaining CPU with core voltage at below 1.30V and to overclocking it at its default CPU multi setting.
this is not a max out for 7.5 multi overclocking)

PangingJr

wow, i like this yorkfield Q9300, It's pretty easy to overclock it.

Q9300 @ 3.6GHz (480x7.5) 4x1GB DDR2-1152MHz

swimtech

Looks like this board and your chip are really nice together - 500Mhz on a quad (CPU-Z Validation)! And your DDR2 (disregarding MemSet - reporting as DDR3) is really blazing! Are you using the PC2 6400HZ modules? I'm inspired to try mine alone - removing the Crucial chips - and seeing what happens...

PangingJr

i don't have G.Skill PC2-6400 HZ any more, i traded one kit in at my dealer for another kit of PC2-8500 HK.

those memories are 2x 2x1GB G.Skill PC2-8500 HK RAM kits (4 sticks with serial number that runs in order).

one of the PC2-8500 HK's kit i have will run DDR2-1333 at least (in a 2x1GB overclock configuration) in the Black DIMM slot of P5Q-D board.