Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

[Dalamar]

I've had my new system for a little over a week now and I've learned a lot about the P45 platform and the UD3R specifically. However, I'm still uncertain on several things. If anyone could help clarify any of these up for me, I'd be very grateful. I realize it's a long post. Even if you only have an answer for one of the questions, I'd really appreciate it!

EDIT: Ok, it seems there is a general aversion to large posts. I will break this up so it is easier to digest.

Getting answers to these questions would help immensely in reducing the number of trials needed to dial in a stable configuration. I could then avoid spending time accidentally trying completely invalid or conflicting configurations, which I'm sure I must be doing now as I keep encountering failures that don't seem to make sense.

I've never had such a tough time trying to find stable settings for a system, even just mildly overclocked (or at anything other than complete stock!) and I've been building computers since the early 90s. I feel like I'm walking a labyrinth backward and blindfolded!

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

1) Is the 400MHz (D) strap looser or tighter than the 333MHz (B) strap? To me it seems like 400 should be the loosest, as it's intended to be used with the highest FSB frequency. As you go down 400 -> 333 -> 266 -> 200 they should be getting tighter and tighter, such that it would be very difficult to run the 266 or 200 straps with a high actual FSB. Is this right? I've seen conflicting information on this, with some posts mentioning that 333 is easier to run at high (400+) FSB, which doesn't make sense to me at all. It seems that 400 should always be the easiest. If this isn't correct, please help me understand why.

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

2) Does specifying the strap manually instead of leaving it at "auto" do anything differently at all in the BIOS except limit your choice of memory multipliers? For instance, if I manually set the 333 (B) strap and then the 2.40B memory multiplier, is the effect exactly the same as leaving the strap at "auto" and choosing 2.40B? I would assume that it is identical, but wanted to be sure.

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

3) What does the strap do, specifically, besides determine the default settings for tRD? For example, say I have to choose a "tight" (266 or 200) strap because I want to use a particular memory multiplier that is only available on that strap. Then, since I'm using a high FSB I manually loosen tRD so that it should still run. Are there still other things besides tRD that the strap sets in the MCH which could cause it to fail at high FSB?

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

4) How, precisely, does the MCH derive it's internal operating clock speed? I've read that it is:

(default CPU multiplier / current set CPU multiplier) x FSB x (some internal multplier?)

Does the strap affect it somehow as well? Otherwise you'd get a very different MCH core clock speed running a 266MHz FSB processor vs. a 333MHz FSB processor even at their default multipliers. Does the MCH core clock speed even matter that much when trying to tune in stable settings?

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

5) Some FSB and memory multiplier/strap (if choosing the memory multiplier does indeed set the strap too, as I assume in #1) combinations seem to be completely invalid and fail regardless of whether I'm using a very low FSB and very relaxed memory timings. Is this expected? If so, is there a central source of documentation about which combinations just won't work? I'm not talking about tRD stuff here, I'm just talking about combinations of FSB clock values and memory multipliers, even with tRD greatly relaxed. For instance, I've found that the 2.66D memory multiplier (which is on the 400 strap) won't work stably at/near 400MHz FSB for me regardless of how much I relax memory timings, reduce CPU multiplier, or increase component voltages. This seems really odd to me, since I'd assume that the 400 strap should fit perfectly with 400MHz FSB.

[Lsdmeasap]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

1) Is the 400MHz (D) strap looser or tighter than the 333MHz (B) strap? To me it seems like 400 should be the loosest, as it's intended to be used with the highest FSB frequency. As you go down 400 -> 333 -> 266 -> 200 they should be getting tighter and tighter, such that it would be very difficult to run the 266 or 200 straps with a high actual FSB. Is this right? I've seen conflicting information on this, with some posts mentioning that 333 is easier to run at high (400+) FSB, which doesn't make sense to me at all. It seems that 400 should always be the easiest. If this isn't correct, please help me understand why.

400 Mhz strap is a faster Freq on the NB with looser internal timings.

As I said previously those who said 333 worked for them at higher FSB but 400 would not likely had some incorrect settings causing 400 to fail, such as tRD to tight or MCH Voltage to high or low.

400 May not always be the easiest, but after 400 FSB it would be eaiser to run and eaiser on the NB and overall voltages then running 333 on a 400+ FSB.

333 will of course work at 400 + FSB, and would be better then 400, but would also require more voltages

2) Does specifying the strap manually instead of leaving it at "auto" do anything differently at all in the BIOS except limit your choice of memory multipliers? For instance, if I manually set the 333 (B) strap and then the 2.40B memory multiplier, is the effect exactly the same as leaving the strap at "auto" and choosing 2.40B? I would assume that it is identical, but wanted to be sure.

They are indentical. The manual selection of MCH Strap just limits the multi's available to the user for said strap to make things easier as those would be the stable mutli't for the set strap. You can set it or leave it in auto and make your Memory Mutli choice and it is the same either way

3) What does the strap do, specifically, besides determine the default settings for tRD? For example, say I have to choose a "tight" (266 or 200) strap because I want to use a particular memory multiplier that is only available on that strap. Then, since I'm using a high FSB I manually loosen tRD so that it should still run. Are there still other things besides tRD that the strap sets in the MCH which could cause it to fail at high FSB?

That is about spot on thoughts. You may need higher MCH or ram voltage to run like that, but you have the general idea correct for sure

4) How, precisely, does the MCH derive it's internal operating clock speed? I've read that it is:

(default CPU multiplier / current set CPU multiplier) x FSB x (some internal multplier?)

Does the strap affect it somehow as well? Otherwise you'd get a very different MCH core clock speed running a 266MHz FSB processor vs. a 333MHz FSB processor even at their default multipliers. Does the MCH core clock speed even matter that much when trying to tune in stable settings?

Like I said before, that guide was written for a old board, with a older chipset where strap selections were not fully possible and I believe tRD was not even a BIOS option when that was written as well. And it does not apply to todays boards, specifically the P35/45.

What MCH Strap you choose to run does just that, runs the NB on that strap speed with Intel's programmed MCH Freq and Timings for the specified speeds

Yes, MCH Speed does matter, but how it matters is up to you. If you want maximum stability with less voltages then you need to use the strap for the FSB Speed you are running. If you want best performance and dont mind tuning in voltages and other timings then your best choice would be a lower then FSB Strap

5) Some FSB and memory multiplier/strap (if choosing the memory multiplier does indeed set the strap too, as I assume in #1) combinations seem to be completely invalid and fail regardless of whether I'm using a very low FSB and very relaxed memory timings. Is this expected? If so, is there a central source of documentation about which combinations just won't work? I'm not talking about tRD stuff here, I'm just talking about combinations of FSB clock values and memory multipliers, even with tRD greatly relaxed. For instance, I've found that the 2.66D memory multiplier (which is on the 400 strap) won't work stably at/near 400MHz FSB for me regardless of how much I relax memory timings, reduce CPU multiplier, or increase component voltages. This seems really odd to me, since I'd assume that the 400 strap should fit perfectly with 400MHz FSB.

There are some Boards with BIOS issues related to or causing this. Other times ram quality comes into play as well as all other settings

I'd have to see the Full timings for this set of failed 2.66D examples to best answer this.

Overall yes, some straps/multi's just will not work at certain exact FSB and one would need to raise or lower the FSB and or voltages and other settings a little and it will work.

P45 is VERY Tricky, sometimes when you expect higher voltage is needed, lower is what it really needed and visa versa

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Thank you very much, Lsdmeasap!

I'd have to see the Full timings for this set of failed 2.66D examples to best answer this.

I will post the exact settings combination(s) I've tried this evening.

[Outer Marker]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Dalamar, I'm one of those forum members that have had some issues with the 400 Strap. I'm in the curious position of being able to provide lots of measurements for evidence, but cannot adequately explain what I am seeing, but here are my generalizations so far with regards to the EP45-UD3R:

If you're using high-speed RAM (1066 or higher) and modest FSB, you need to be on the 333 (B) Strap. When you do this, you'll actually see more stability using lower MCH Core Voltage... default even 1.100

If you're going for high FSB (435+) you need to be on the 400 (D) Strap, and have some readily overclockable 800 MHz RAM. You can get things stable with 800 MHz RAM by using the driving strength profiles. These profiles do something more than just adjust Static tRD because I've tried all manner of values.

I think the magic that we're seeing here is that on the EP45-UD3R, the 333 (B) strap is good up to a CPU FSB of approximately 433 and requires low MCH Core Voltages. I interpret that as a great bonus! Low voltages with a generous overclock.

Now, those of us that aren't happy with "generous" and want "sick" overclocking... will need to switch to the 400 (D) Strap around 435 FSB and then look for higher MCH Core voltages and loose timings for stability.

If you want my opinion, Gigabyte is doing this because they are a leader in power saving and "green" technology. Seems like we can get very generous overclocks with very little voltage over stock - up to a point. After that, normal overclocking rules apply... more power, more heat, more stability etc.

[Lsdmeasap]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Just so you know 2.66D does work in P45, Here is some images from my Crucial Memory Reviews using 400+ FSB and 2.66D Multi >>>

Also, if you read my review you will see I used B strap at 450-500 FSB as well. It all just depends on MANY other things you can set. But most of the time you can use one above or below any given FSB speed with the proper settings

Crucial 2x1GB *Click Images to Enlarge*

450 FSB Memory @ 1200 2.66D

<a href="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial2x1-1200-266D.png" target="_blank"><IMG align=middle WIDTH=800 HEIGHT=500 SRC="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial-2x1-1200-266D-small.png" border="0"></a>

Crucial 4x1 *Click Images to Enlarge*

435 FSB Memory @ 1160 2.66D

<a href="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial4x1-1160-trd8-266.png" target="_blank"><IMG align=middle WIDTH=800 HEIGHT=500 SRC="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial-4x1-1160-trd8-266-small.png" border="0"></a>

Crucial 2x2GB *Click Images to Enlarge*

428 FSB Memory @ 1140 2.66D

<a href="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial2x2-1140-266D-tRD8-3.png" target="_blank"><IMG align=middle WIDTH=800 HEIGHT=500 SRC="http://i2.photobucket.com/albums/y4/lsdmeasap/Crucial-2x2-1140-266D-tRD8-small.png" border="0"></a>

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Dalamar, I'm one of those forum members that have had some issues with the 400 Strap. I'm in the curious position of being able to provide lots of measurements for evidence, but cannot adequately explain what I am seeing, but here are my generalizations so far with regards to the EP45-UD3R:

If you're using high-speed RAM (1066 or higher) and modest FSB, you need to be on the 333 (B) Strap. When you do this, you'll actually see more stability using lower MCH Core Voltage... default even 1.100

If you're going for high FSB (435+) you need to be on the 400 (D) Strap, and have some readily overclockable 800 MHz RAM. You can get things stable with 800 MHz RAM by using the driving strength profiles. These profiles do something more than just adjust Static tRD because I've tried all manner of values.

Excellent information! Thank you very much.

I think the magic that we're seeing here is that on the EP45-UD3R, the 333 (B) strap is good up to a CPU FSB of approximately 433 and requires low MCH Core Voltages. I interpret that as a great bonus! Low voltages with a generous overclock.

Now, those of us that aren't happy with "generous" and want "sick" overclocking... will need to switch to the 400 (D) Strap around 435 FSB and then look for higher MCH Core voltages and loose timings for stability.

If you want my opinion, Gigabyte is doing this because they are a leader in power saving and "green" technology. Seems like we can get very generous overclocks with very little voltage over stock - up to a point. After that, normal overclocking rules apply... more power, more heat, more stability etc.

I'm definitely not looking for insane overclocking. In fact, I'm not even looking to increase CPU speed at all. I just want to expand memory bandwidth and minimize memory latency at that expanded bandwidth.

If I can get a stable setup (i.e. 24-hr Memtest86 and 24-hr Prime95 blend stable) at 400MHz FSB with a 7.5x CPU multiplier (keeping it stock at 3.0GHz) and my RAM running near spec (~960-1066 range) with reasonably low latency, I'll be happy. I'd rather keep voltages stock too, so your reports of stability even at stock MCH core voltage are very encouraging.

Thanks again for the reply!

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Just so you know 2.66D does work in P45, Here is some images from my Crucial Memory Reviews using 400+ FSB and 2.66D Multi >>>

Also, if you read my review you will see I used B strap at 450-500 FSB as well. It all just depends on MANY other things you can set. But most of the time you can use one above or below any given FSB speed with the proper settings

Thanks! It helps to know that it is indeed possible, after having so many frustrating experiences.

[Lsdmeasap]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Anytime, I hope you have found enough info now to start your venture

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

Ok, I'm not sure if it was due to the new F7 BIOS, or if it's just that I wasn't changing as many settings at a time. In any case, I seem to have found a setup that gets 2.66D working fairly well. Here are the settings I'm using currently:

Robust Graphics Booster ...............: Auto
CPU Clock Ratio .......................: 7x
Fine CPU Clock Ratio...................: +0.5
CPU Frequency .........................: 3.0GHz

Clock Chip Control 
Standard Clock Control
CPU Host Clock Control.................: Enable
CPU Host Frequency (Mhz) ..............: 400
PCI Express Frequency (Mhz) ...........: 100

C.I.A.2 ...............................: Disabled

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

DRAM Performance Control
Performance Enhance....................: Standard
Extreme Memory Profile (X.M.P.)........: Disabled
(G)MCH Frequency Latch.................: Auto
System Memory Multiplier ..............: 2.66D
Memory Frequency (Mhz) ................: 1066
DRAM Timing Selectable ................: Manual

Standard Timing Control
CAS Latency Time.......................: 5
tRCD ..................................: 5
tRP....................................: 5
tRAS...................................: 15

Advanced Timing Control
tRRD...................................: Auto
tWTR...................................: Auto
tWR....................................: Auto
tRFC...................................: Auto
tRTP...................................: Auto
Command Rate (CMD) ....................: Auto

Driving Strength Profiles
Driving Strength ......................: 1066

Channel A
Static tRead Value.....................: 7
tRD Phase0 Adjustment..................: Auto
tRD Phase1 Adjustment..................: Auto
tRD Phase2 Adjustment .................: Auto
tRD Phase3 Adjustment..................: Auto

Trd2rd(Different Rank).................: Auto
Twr2wr(Different Rank).................: Auto
Twr2rd(Different Rank).................: Auto
Trd2wr(Same/Diff Rank).................: Auto

Dimm1 Clock Skew Control...............: Auto
Dimm2 Clock Skew Control...............: Auto
DDR Write Training.....................: Auto

Channel B
Static tRead Value.....................: 7
tRD Phase0 Adjustment..................: Auto
tRD Phase1 Adjustment..................: Auto
tRD Phase2 Adjustment .................: Auto
tRD Phase3 Adjustment..................: Auto

Trd2rd(Different Rank).................: Auto
Twr2wr(Different Rank).................: Auto
Twr2rd(Different Rank).................: Auto
Trd2wr(Same/Diff Rank).................: Auto

Dimm1 Clock Skew Control...............: Auto
Dimm2 Clock Skew Control...............: Auto
DDR Write Training.....................: Auto

Motherboard Voltage Control
Voltage Type...........................: Manual

CPU---------------------Normal-----------Current

Load Line Calibration..................: Disabled
CPU Vcore...............1.25000V.......: 1.25000V
CPU Termination.........1.200V.........: 1.200V
CPU PLL.................1.500V.........: 1.500V
CPU Reference...........0.760V.........: Auto

MCH/ICH
MCH Core................1.100V.........: 1.100V
MCH Reference...........0.760V.........: Auto
MCH/DRAM Reference......0.900V.........: Auto
ICH I/O.................1.500V.........: 1.500V
ICH Core................1.100V.........: 1.100V

DRAM
DRAM Voltage............1.800V.........: 2.100V
DRAM Termination........0.900V.........: Auto
Channel A Reference.....0.900V.........: Auto
Channel B Reference.....0.900V.........: Auto

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

I'm also using MemSet to try dialing in some of the advanced memory settings a bit more to match the EPP from the modules. I will do extended testing overnight.

[Dalamar]

Re: Ok, a deep breath, and some specific EP45-UD3R BIOS questions.

So far so good. That configuration survived 6+ hours running Prime95 blend and an instance of MemTest (the Windows one, from HCI Design) to hit the remaining RAM not covered.

I stopped it this morning, then rebooted and kicked off a new Memtest86+ run. We'll see how that fared when I return home from work.

Once I found this configuration last night I made several attempts to get to tRD 6, but despite pushing the MCH to 1.320V it wouldn't POST. I'm fairly happy with tRD 7, especially on the 400 strap. Sandra reports memory latency at a consistent 73ns.