What would you try next?

[KayossZero]

Re: What would you try next?

Well the main site says OCZ Technology | Products | Memory | OCZ DDR2 PC2-8500 Reaper HPC 4GB Edition "2.1-2.2" Newegg also says 2.1 Newegg.com - OCZ Reaper HPC 4GB (2 x 2GB) 240-Pin DDR2 SDRAM DDR2 1066 (PC2 8500) Dual Channel Kit Desktop Memory Model OCZ2RPR10664GK - Desktop Memory Amazon says 2.2 Amazon.com: OCZ OCZ2RPR10664GK PC2-8500 DDR2 Dual Channel Reaper Series 1066MHz 4G Kit CL 5-5-5-18 Memory: Electronics so I'd say either can work depending, and while TRAS says "18" mine seem to do 15 okay, I haven't tried to go say 4-4-4-12 on them or anything yet. (though I doubt at 533~ FSB they would go lower anyways)

[corlay]

Re: What would you try next?

Things I would try, in order:

• Try the 2.66C multiplier.

Tried it, with worse results.

• MCH voltage up to 1.32,

tried this as well. Funny how that MCH core works, eh?
I started high and then worked my way down the ladder, skipping every other rung. It didn't like anything on my first pass through, with everything failing p95 in < 5min.. So then I tried MCH=1.26 (I hit 1.28and 1.24 on my first pass...) and it liked it better, but still not stable. Only got ~1hr. of p95 with that setting.

• Set Vcore to as near to 1.325V as you can with LLC enabled.
• Add VTT (CPU Term) up to 1.3V while maintaining the 1.325Vcore

Well, I sort of followed your advice here. I set Vcore=1.30 and Vtt= 1.26 without LLC; and it didn't really help things.

• Set CPU TErm temporarily to 1.2V, then set MCH Ref manually to 0.740V and increase CPU Term back up. Again try 1.24-1.3V CPU term.
• Again set CPU Term to 1.2V, this time decreasing MCH ref to 0.720V, then test 1.24-1.3V CPU Term

I never made it this far to try these last few suggestions...

But I've had a change of heart:
I'm getting tired of trying to find the needle in the haystack by running my 8gb @1066.

One thing I did new yesterday, was install MemSet. I wanted to see what the MB was giving me with my advance DRAm timings all set on Auto, because any attempts by me to set reasonable timings manually always fails a prime95 test almost immediately. Well, MemSet revealed very loose settings. Like a tREAD=12 and tWR=11 (subtracting 5 from the MemSet reported value of 16 for Write to Precharge Delayed...).Combine this with my inability to run the Standard Timings at spec (I had to use 5-5-5-18 vs. 5-5-5-15), and I just think that this is becoming a lost cause. With DRAM so loose and still no complete stability achieved, maybe I should just concede that underclocking DRAM is the smart (and easy) way to go?

So, I set my System Memory Multiplier to 2.00D, set my Standard Timings back to 5-5-5-15, and manually set my Advanced Timings to 4-4-8-72-4-2 and tREAD=8. I also reduced my DRAM Voltage to 2.14 from 2.24, but kept MCH Core=1.26; and re-used my previously "best" CPU settings of Vcc=1.275, Vtt=1.22, PLL=1.5.

I set prime95 to run a blended test late last night, and as of this morning, she was still running strong after 8hrs. If it's still running when I return home this evening, I'll try and further reduce my Standard Timings to 4-4-4-12 and maybe try and pinch tREAD down to 7. And if that all goes well, I'll then start adding FSB and see how far I get.

One thing that I've learned in all of this, is that my core temps under load @1.250 vs. 1.300 are pretty much negligible. And I have a cooling fan on that chip. Whereas if I need to push MCH core up to 1.32 (or above?) to try and get northbridge stability @DDR-1066, that's creating a hot spot that I don't have any additional cooling for, nor any reliable way to report actual temps. And I figure if I can get to (or close to...) [email protected] vs. [email protected] I'm better off, anyway.

Your thoughts?

[Psycho101]

Re: What would you try next?

I would still encorage you to set a higher Vcore with LLC, and combine it with lowering MCH ref to 0.74 @ 1.2VTT and then trying 0.720V if that doesn't work.

You've made good progress. I'd use the 1hr+ stable settings for 1066
Your plan may just well be better though, but it's worth giving this a shot. I know it seems like all fannying about and annoying,; set this, reboot, wait 1 hour to have it fall over on you etc, but MCH ref is like voodoo.

For 500 FSb, I can gurantee that, short of going phase change you won't get a 9.5 multi working. You may squeeze 500*8.5, but 500*8 is a more realistic aim for now. You may have meant this, and it was just a typo, so if that's the case, NM.

For your RAM timings, you may find you can run 4-4-4-12 up to a point. My particular RAM will do it up to 890 at 2.1V and to 910 at 2.25V before it gives up and needs at least CAS5-4-4-12 or looser. Depending on the ICs used in the OCZ you may get better or worse results.

[corlay]

Re: What would you try next?

yep, I'm hummin' along now!

I stopped my p95 blended test @ the 14hr. mark, and set my timings lower, and re-started it.

going with 4-4-4-12 with tRFC=54 and tREAD=7 (all other settings the same...) and it's into hour 5 and still going strong...

at first I tried tREAD=6 and it wouldn't boot. (series of short beeps), so I guess 7 is the limit.

Before that I got ****y, and went for CPU Host Frequency=422 with my current settings, and it wouldn't fully boot, and I couldn't get back into the BIOS either, so I had to manually re-set with the screwdriver trick on the MB pins, and start over.

I'm going to stick with the [email protected] scenario for the weekend, and assuming my current test makes it through until morning, I'll then start squeezing-down my Vcc and MCH as see what's possible. At the very least, I should establish a good, reliable baseline for further overclocking later.

Once I get this nailed, I may try the 2.40B strap and see what happens, or maybe lower my CPU multiplier and go for a larger CPU host frequency; just to get my DRAM bandwidth closer to DDR-1066. The final goal for me would be somehwere around 4.0Ghz with somewhere near DDR-1066 and stay below 60c core temps under load.

Who knows if I'll get there...

But at least I've got something positive to build upon, finally. I was gettting really frustratred...

Thanks for everyone's help so far!

[corlay]

Re: What would you try next?

I'm going to stick with the [email protected] scenario for the weekend, and assuming my current test makes it through until morning, I'll then start squeezing-down my Vcc and MCH as see what's possible. At the very least, I should establish a good, reliable baseline for further overclocking later.

ok,
the weekend testing went well!

I've got my DRAM timings squeezed-down to:
Standard Timings: 4-4-4-12
Advanced Timings: 3-4-4-48-4-2
tREAD = 7

Couldn't reduce MCH Core below 2.16v
And my Vcc still sits @ 1.275v

Question: When it comes to DRAM does bandwidth rule supreme? Or Do timings figure significantly? What I'm getting at is: does DDR-800 come close to the performance of DDR1066, *if* the DDR-800 timings are super tight, and the DDR1066 timings are awfully loose for stability purposes?

I'm wondering if I should just continue along on the 2.00D strap, and see how far I can push my e8500 with a 9.5 multiplier? Or should I get back on to trying to get that 2.66D strap stable?

(my almost stable [email protected] DRAM timings were 5-5-5-18 and 4-5-11-72-4-2 w/ tREAD=12)

For kicks, I raised CPU Host Frequency=422, and Vcc=1.30, Vtt=1.24 with all my other settings left @ my stable [email protected] config - and she wouldn't fully boot. After re-setting the BIOS, and re-entering my overclock values, I then tried CPU Host Frequency=411, and that booted but failed p95 pretty quickly.

This OC'ing stuff is finicky stuff, eh? <g>

Here's my latest stable settings, just for record:

Motherboard: Model: EP45-UD3P Rev#: 1.1 Bios Version: [B]F9[/B]
Cpu: e8500
Ram: 4x2gb OCZ, OCZ2RPR10664GK, www.ocztechnology.com/products/memory/ocz_ddr2_pc2_8500_reaper_hpc_edition
Power Supply Unit: Corsair, HX-620, 620w

MB Intelligent Tweaker(M.I.T.)
Robust Graphics Booster -------------- : [Auto]
CPU Clock Ratio ---------------------- : [9]
Fine CPU Clock Ratio ----------------- : [0.5]
CPU Frequency ------------------------ : [400]

Clock Chip Control
Standard Clock Control
CPU Host Clock Control --------------- : [Enabled]
CPU Host Frequency (Mhz) --------------: [400Mhz]
PCI Express Frequency (Mhz) ---------- : [100]
C.I.A.2 ------------------------------ : [Disabled]

Advanced Clock Control ------- [Press Enter]
CPU Clock Drive ---------------------- : <<< 800
PCI Express Clock Drive -------------- : <<< 900
CPU Clock Skew (ps) ------------------ : <<< 0
MCH Clock Skew (ps) ------------------ : <<< 0 

DRAM Performance Control
Performance Enhance ------------------- : [Disabled]
Extreme Memory Profile --- (X.M.P.) --- : [Disabled]
(G)MCH Frequency Latch ---------------- : [Auto]
System Memory Multiplier -------------- : [2.00D]
Memory Frequency (Mhz) ---------------- : [1066, 800]
DRAM Timing Selectable ---------------- : [Manual]

Standard Timing Control
CAS Latency Time -------- 5 ---------- : [4]
tRCD  ------------------- 5 ---------- : [4]
tRP --------------------- 5 ---------- : [4]
tRAS  ------------------- 15 --------- : [12]

Advanced Timing Control
tRRD  ------------------- 4 ---------- : [3]
tWTR  ------------------- 4 ---------- : [4]
tWR --------------------- 6 ---------- : [4]
tRFC  ------------------- 68 --------- : [48]
tRTP  ------------------- 4 ---------- : [4]
Command Rate (CMD) ------ 0 ---------- : [2]

Driving Strength Profiles (A/B)
Driving Strength --------- : [Auto]

Static tRead Value (A/B) ---------------- : [7]
(Note: All other settings in this section left on Auto)

Motherboard Voltage Control
CPU
Voltage Types------------ Normal -------- Current -----
Load-Line Calibration ----------------- : [Disabled]
CPU Vcore --------------- 1.25 -------- : [1.275]
CPU Termination --------- 1.200 ------- : [1.22]
CPU PLL ----------------- 1.500 ------- : [1.50]
CPU Reference ----------- 0.760 ------- : [Auto]
CPU Reference2 ---------- 0.800 ------- : [Auto]

MCH/ICH
MCH Core ---------------- 1.100 ------- : [1.26]
MCH Reference ----------- 0.760 ------- : [Auto]
MCH/DRAM Ref ------------ 0.900 ------- : [Auto]
ICH I/O ----------------- 1.500 ------- : [1.50]
ICH Core ---------------- 1.100 ------- : [1.10]

DRAM
DRAM Voltage ------------ 1.800 ------- : [2.14]
DRAM Termination -------- 0.900 ------- : [1.07]
Channel A Reference ----- 0.900 ------- : [1.07]
Channel B Reference ----- 0.900 ------- : [1.07]

Advanced Settings
Limit CPUID Max. to 3 ------------------ : [Disabled]
No-Execute Memory Protect -------------- : [Enabled]
CPU Enhanced Halt (C1E) ---------------- : [Disabled]
C2/C2E State Support ------------------- : [Disabled]
x C4/C4E State Support ----------------- : [Disabled]
CPU Thermal Monitor 2 (TM2) ------------ : [Enabled]
CPU EIST Function  --------------------- : [Disabled]
[URL="http://forums.tweaktown.com/#"]Virtualization Technology[/URL] -------------- : [Disabled]

Integrated Peripherals
Legacy USB Storage Detect  ------------- : [Disabled]

[corlay]

Re: What would you try next?

After (finally) solidifying my stable [email protected] settings, I built upon what I learned and made some further progress last night "climbing the ladder".

First, I tried my stable settings with the 2.00B strap vs. 2.00D; and it balked. I just figured I'd give it a shot to try and realize the performance benefits of the B strap. Oh well..

Then, I decided to see how far i could push the FSB with my 9.5 multiplier. I added Vcc, Vtt, PLL, MCH core, and adjusted MCH Ref downward. Within a few tries I was booted into Windows with [email protected], and I ran some quick p95 tests. Things looked pretty stable, so I pushed further.

I was able to get what looks like stable settings with [email protected]. Currently running in hour-10 in p95 (blended), and looking good.

What was interesting is that with the 422 FSB, I first tried loosening my DRRAM timings from 4-4-4-12 to 5-5-5-15, but the system wouldn't boot, and gave the continuous short beeps error code at startup (listed as "power error" in the MB manual). I wonder why that is? I figured looser would mean an easier road to stability. 4-4-4-12 booted fine, BTW. I did have to relax tREAD from 7 to 8, though.

My max core temps through 10 hr.of testing is 61c. Is that about the "safe limit" for an every-day configuration? I certainly don;'t want to push things too far.

And one final question: Will I benefit from lowering my multiplier and pushing FSB even further (to the 444-450 range) to get my memory bandwidth up, and keeping my CPU around 3.9-4.1 GHz? Maybe something like [email protected]? ANd if I attempt this, will I likely be able to use my current settings for Vcc, Vtt becaus ethe reasultant CPU speed is roughly the same? (I don't think I want to push my core temps too much further...) A raise in FSB will likely require more MCH Core voltage, though, eh?

Here's my [email protected] settings, just for record:

Motherboard: Model: EP45-UD3P Rev#: 1.1 Bios Version: [B]F9[/B]
Cpu: e8500
Ram: 4x2gb OCZ, OCZ2RPR10664GK, www.ocztechnology.com/products/memory/ocz_ddr2_pc2_8500_reaper_hpc_edition
Power Supply Unit: Corsair, HX-620, 620w

MB Intelligent Tweaker(M.I.T.)
Robust Graphics Booster -------------- : [Auto]
CPU Clock Ratio ---------------------- : [9]
Fine CPU Clock Ratio ----------------- : [0.5]
CPU Frequency ------------------------ : [4.0GHz]

Clock Chip Control
Standard Clock Control
CPU Host Clock Control --------------- : [Enabled]
CPU Host Frequency (Mhz) --------------: [422Mhz]
PCI Express Frequency (Mhz) ---------- : [100]
C.I.A.2 ------------------------------ : [Disabled]

Advanced Clock Control ------- [Press Enter]
CPU Clock Drive ---------------------- : <<< 800
PCI Express Clock Drive -------------- : <<< 900
CPU Clock Skew (ps) ------------------ : <<< 0
MCH Clock Skew (ps) ------------------ : <<< 0 

DRAM Performance Control
Performance Enhance ------------------- : [Disabled]
Extreme Memory Profile --- (X.M.P.) --- : [Disabled]
(G)MCH Frequency Latch ---------------- : [Auto]
System Memory Multiplier -------------- : [2.00D]
Memory Frequency (Mhz) ---------------- : [1066, 844]
DRAM Timing Selectable ---------------- : [Manual]

Standard Timing Control
CAS Latency Time -------- 5 ---------- : [4]
tRCD  ------------------- 5 ---------- : [4]
tRP --------------------- 5 ---------- : [4]
tRAS  ------------------- 15 --------- : [12]

Advanced Timing Control
tRRD  ------------------- 4 ---------- : [4]
tWTR  ------------------- 4 ---------- : [4]
tWR --------------------- 6 ---------- : [6]
tRFC  ------------------- 68 --------- : [54]
tRTP  ------------------- 4 ---------- : [4]
Command Rate (CMD) ------ 0 ---------- : [2]

Driving Strength Profiles (A/B)
Driving Strength --------- : [Auto]

Static tRead Value (A/B) ---------------- : [8]
(Note: All other settings in this section left on Auto)

Motherboard Voltage Control
CPU
Voltage Types------------ Normal -------- Current -----
Load-Line Calibration ----------------- : [Disabled]
CPU Vcore --------------- 1.25 -------- : [1.3250]
CPU Termination --------- 1.200 ------- : [1.24]
CPU PLL ----------------- 1.500 ------- : [1.57]
CPU Reference ----------- 0.760 ------- : [Auto]

MCH/ICH
MCH Core ---------------- 1.100 ------- : [1.30]
MCH Reference ----------- 0.760 ------- : [0.765] dropped value to 0.74 when Vtt=1.20
MCH/DRAM Ref ------------ 0.900 ------- : [1.07]
ICH I/O ----------------- 1.500 ------- : [1.50]
ICH Core ---------------- 1.100 ------- : [1.10]

DRAM
DRAM Voltage ------------ 1.800 ------- : [2.14]
DRAM Termination -------- 0.900 ------- : [1.07]
Channel A Reference ----- 0.900 ------- : [1.07]
Channel B Reference ----- 0.900 ------- : [1.07]

Advanced Settings
Limit CPUID Max. to 3 ------------------ : [Disabled]
No-Execute Memory Protect -------------- : [Enabled]
CPU Enhanced Halt (C1E) ---------------- : [Disabled]
C2/C2E State Support ------------------- : [Disabled]
x C4/C4E State Support ----------------- : [Disabled]
CPU Thermal Monitor 2 (TM2) ------------ : [Enabled]
CPU EIST Function  --------------------- : [Disabled]
Virtualization Technology -------------- : [Disabled]

Integrated Peripherals
Legacy USB Storage Detect  ------------- : [Disabled]

[Edit: I was able to lower Vcc from 1.3275 to 1.3250 and maintain stability]

[Psycho101]

Re: What would you try next?

Depending a little on luck you may find you don't need any more MCH voltage. Some can run really low MCH Core at 500 FSB +, you may be one of them now you've found a stable RAM multi. Also, don't worry about 2.00B, as at 400FSB + 2.00D would likely be the better performer.

Usually a higher FSB and a lowered multi will require a little more Vcore than using the maximum multiplier, but not always. It's often a surprise to people that they need to ass a little CPU volts for the same speed, but it is possible. You may also need to change VTT, however if you bench with Prime Large FFT, which is great at finding problems at high FSB and fail, first thing I would do is temporarily raise Vcore 2 ticks to eliminate it as the cause and lower MCH ref to 0.720 @ 1.2VTT.

At 61c your temps are good. I wouldn't worry until I get to 65c in Prime, but as always lower's better, so work Vcore down as low as possible while maintaining stability.

[corlay]

Re: What would you try next?

O-K!

My [email protected] config has proved to be 24hr. stable in p95.

So, getting a little ****y;
I went for [email protected].

I had trouble booting to Windows at first,
and then once I tweaked so windows would boot,
I would blue-screen in p95 Small FFT.

Well, as it turns out, at least in part,
it's just like you said: no more MCH core v may be needed.

Since I initially raised voltages almost across the board for the 444FSB,
I tried reverting back to my stable 422FSB, and then only added a bit of Vcc and Vtt:

Vcc=1.350 < max. Vcc for my e8500, per Intel spec.
Vtt=1.26

That almost worked.

I then kept everything the same, but enabled LLC;
and am now running a p95 blended test with ~2hrs. under my belt.
Core temps are ranging from 58c-64c; with a solid avg. of 61c;
and an idle temp of 41c.

I may try tomorrow to disable LLC to see if it's really needed,
but other than that I'm completely satisified with this config.
It represents a bit more than my initial goal, and is probably the limit temp-wise of what I should run 24-7.

If this latest config proves stable,
you likely won't hear from me again...

[Update: Whelp, it looks like y'all are stuck with me for a while longer.
I awoke this am to a windows error message saying that prime 95 experienced and unexpected problem,
and had to be halted. p95 ran for 2.5hrs. before windows stopped it.
I suppose that's better than a full-on windows blue-screen crash.
And also better than an error encountered within p95?
I just re-started p95 on a new blended torture test,
just to see if the same Windows error happens again.
Maybe it was just a fluke?]

Thanks for all the great advice!

[corlay]

Re: What would you try next?

Ok,

It looks like my system is going to require a large jump in Vcc, Vtt, and maybe MCH for a BIOS config yielding much over 4.0GHz.

I attempted to get things stable at both [email protected] and [email protected], and while I could get things to boot to windows ok, one of two cores would fail a p95 blended test usually within the first hour (or minutes!).

It's possible that I might be able to play around with voltages and timings some more and get one of these stable, but I'm concerned about running a system 24-7 with Vcc set over max. spec (1.350v) and an MCH core over 1.30v. The heat is just too much for my system's cooling capacity, I think.

I could also try lowering my CPU multiplier from 9.5 to 9.0; but fear that FSB values over 450 will also put too much strain on the MCH.

So, I'm settling-in on my stable [email protected] config, and am actually p95 testing right now with the Vcc reduced from my previous "good" value of 1.3275 to 1.3125.

Once I finialize how tight I can get my timing and voltages for this config, I'm going to benchmark against my [email protected] config. The biggest difference between these two is the tREAD setting. I was able to run tREAD=7 @ 411FSB, but had to raise it to 8 for 422FSB stability. (all other DRAM timings and voltages are identical between the two configs)

I did some research, and read some interesting articles on the AnandTech website yesterday; and they make a pretty compelling case for not sacrificing tREAD for high FSB:
AnandTech: Intel X38 Tango - Is High FSB Overclocking Worth It?

And this article is also very interesting, mostly about how the MCH is key in translating data between the CPU and the DRAM:
AnandTech: ASUS ROG Rampage Formula: Why we were wrong about the Intel X48

So, I'm going to test each of my 2 stable configs, and see if a lower tREAD trumps a higher FSB, at the same CPU multiplier.

"Efficiency vs. Power"
Should be interesting....

[Psycho101]

Re: What would you try next?

If you get errors instantly in prime, try dropping VTT to 1.20 and set MCH Ref to 0.720, then raise VTT back up. If that fails, try 0.700V @ 1.2VTT, then increase VTT bach to where you had it.

A lower tread won't be better than a higher FSB at the same system multiplier. That wasn't the point of the AnandTech article. The article was refering to something like 8*500 @ tRead=11 versus 9.5*422 @ tRead=8. The increase in FSB at the same multiplier needed to necessitate an increase tRead will always give better performance simply due to the higher CPU speed. To compare as Anand did, you need to raise the FSB and increase tRead while keeping the CPU constant.

[corlay]

Re: What would you try next?

A lower tread won't be better than a higher FSB at the same system multiplier. That wasn't the point of the AnandTech article. The article was refering to something like 8*500 @ tRead=11 versus 9.5*422 @ tRead=8. The increase in FSB at the same multiplier needed to necessitate an increase tRead will always give better performance simply due to the higher CPU speed. To compare as Anand did, you need to raise the FSB and increase tRead while keeping the CPU constant.

understood.

But in my case, given that it seems that I've found my comfortable limit for FSB on the 9.5 CPU multiplier; I thought it would be interesting to test FSB=411 & tREAD=7 vs. FSB=422 & tREAD=8 because the FSB are so close to eachother.

To definitively reveal which is more significant:

an increase of 11 FSB?
(on a 9.5 multiplier)

or

a decrease of 1 tREAD?
(on the 2.00D strap)

[Psycho101]

Re: What would you try next?

You'd have to benchmark it to be sure. Not just a RAm bandwidth test and SuperPI, but something more robist like PC Mark Vantage. Register two email addresses to get two free runs and do a before and after.

I'd guess almost 100% that although RAM bandwidth would drop by 200-350MB/s, the extra 100MHz CPU speed would improve some applications and the RAM bandwidth others. Depends how you use your PC.

[corlay]

Re: What would you try next?

Not just a RAm bandwidth test and SuperPI, but something more robist like PC Mark Vantage.

I've been using PassMark PerformanceTest for this purpose:
http://www.passmark.com/products/pt.htm

Will that do?

It cycles thorough a bunch of 2d and 3d apps, with some video tossed-in. Seems comprehensive enough.

[Psycho101]

Re: What would you try next?

Looks good. Certainly better than synthetic benchmarks. Only plus about Vantage is it contains exerts from real world programs, where as these are still kind of simulated. Good enough for an approximate idea though.

[corlay]

Re: What would you try next?

right you are, Psycho!

the [email protected], tREAD=8 performed about 3% better than the [email protected], tREAD=7 config - across the board in all categories.

I expected the CPU test to be better,
but certianly not the DRAM!

Must be the slightly increased memory bandwidth (DDR-844 vs. DDR-822) trumped the tREAD affect?

So, now I have a renewed interest in OC higher,
tREAD be damned!

Starting working on [email protected], tREAD=8 again.
Wish me luck!