Dealing with ASUS P45

[MachineLearning]

Hey everyone,

RaspberryHex and I have recently been experimenting with ASUS P45, and want to share our findings.

It has been known for quite some time that ASUS P45 boards are outright inferior to Gigabyte P45 for both raw FSB & quad-cores, as well as ASUS X48 for dual-core FSB and efficiency. The intention of this post is to help people who have ASUS P45, and want to get the best out of them, by providing relevant data we have collected with our respective boards. The primary focus in this post is on clocking FSB.

Important FSB tips:
• Windows 2000 is highly recommeded (also for other boards, memory and CPU frequency valids). This alone gave my P5Q Turbo a +14MHz boost with identical BIOS settings & +7MHz to EP45-UD3R.
    ◦ Also recommend downloading MSCONFIG.exe for Win2K, in order to quickly and easily disable unnecessary services and limit in-OS RAM. You can find a link here.
• Saving a CPU-Z validation at a lower frequency will assist significantly with validating at much higher frequencies. It appears to both quicken and lessen the stress of validation.
    ◦ Ex: A 550MHz FSB validation assisted me in going from 678.5MHz to 683.3MHz valid.
• CPU and NB skews are important, although Auto/Auto is usually competent here.
    ◦ RHex
        ▪ CPUskew/NBskew = 400/100 is best (P5Q Deluxe, P5Q Pro Turbo). Detailed below.
    ◦ ML
        ▪ CPUskew:NBskew = ~4:3 with a delta of 200-300. 400/100 clocked poorly (P5Q Turbo). Detailed below.
• On both of my P5Q Turbos, it was necessary to change "Performance Margin Enhancement" to "Performance." The other two options made it difficult to even attain 500-550MHz.
    ◦ RHex's boards did not exhibit this trait.

I saw the Windows 2000 tip from the description on one of A6M-Reisen's submissions. I don't recall hearing about the CPU-Z trick before, but I'm most likely not the first to figure this out.

~~~

RHex Skew Testing (P5Q Deluxe)

Settings: 1.52VTT, CPU 0/2 GTLs=0.590x, CPU 1/3 GTLs=0.580x, NB GTL=0.545x

CPU/NB Clock Skews (Max Achievable using SetFSB w/o CPU-Z @ Ambient=23C°)

Auto/100	662MHz
100/100	No POST @ 620MHz
0/100	No POST @ 620MHz
300/100	652MHz
400/100	662MHz (equal to Auto/100)
300/0	632MHz
500/100	No boot @ 620MHz
500/200	652MHz
600/300	653MHz
700/300	No Boot @ 620MHz
700/400	644MHz
700/500	636MHz
800/500	652MHz
900/600	No POST @ 620MHz
900/700	645MHz
1000/700	646MHz
1000/800	No Boot @ 620MHz
1100/700	No POST @ 620MHz
1100/800	No Boot @ 620MHz
1200/800	No POST @ 620MHz
1200/900	No POST @ 620MHz
1200/1000	No POST @ 620MHz

"I suspect clk skews to be pretty much consistent for each mobo model
I mean if it does best 400/100 for one P5Q Dlx/Prem/-E then it should be the same for another"
~~~

ML Skew Testing (P5Q Turbo) - Batch L831A638 (Wk31/2008)

Universal settings: E8400 E0 = 690MHz+ ambient capable | 550MHz boot | 1.4vCore | 1.48vMCH | 1.5VTT | 1.5vPLL | 0.63x CPUGTL; NBGTL unknown | 2.4vDIMM | 1GB Ballistix .16FD3 @ A1, Loose RAM timings i.e. 6-9-9-24-2 | PL"22"(=6), Clock Twister = Light, CPU Margin Enhancement = Performance

- Win2K, MSCONFIG.exe -> Diagnostic mode + 64MB RAM enabled, SetFSB + CPU-Z running
- BAR FED14382 = 03020202 -> 07070706 (mirror EP45-UD3R @ 600MHz boot; negligible impact on FSB OC)

Common step: 550 -> 600 -> 620 -> 640 -> 650 -> 660 -> 670 -> 672 -> 674 -> 676 -> 678 -> 680 -> 681 -> 682 etc.

Max BCLK seen before/at lockup is listed, not necessarily validated; "Normal" is assumed = 0

Auto/Auto	678.5MHz
0/0	600 -> 620
100/0	640 -> 650
200/0	680
300/0	672
400/0	660
400/100	674
500/0	620 -> 640
600/0	600 -> 620
400/200	676
400/300	620 -> 640
500/200	674
500/300	670
600/300	674
600/400	676
700/400	672
700/500	678.5
800/500	670
800/600	678.5 -> 680 (restarted instead of freezing)
800/600 (2nd try)	686
900/600	670
900/700	681.69
1000/700	670
1000/800	660 -> 670
1100/800	678.5 -> 680
1100/900	670 -> 672
1200/900	683.3 -> 684
1200/1000	660
1300/1000	676 -> 678
1300/1100	660 -> 670
1400/1100	682.5
1400/1200	676
1500/1200	674
1500/1300	672 -> 674

+200 delta mean: 672.78 approx.
+300 delta mean: 674.63 approx.

Best skew	Approx. Ratio
800/600	4:3
1200/900	4:3
900/700	4:3.10
1400/1100	4:3.14

Rule?: Best = 4:3 CPUskew:NBskew; with +200 or +300 delta

~~~

ML PL Testing (appears to also apply to other 31-value PL boards like TPower I45, though more testing is needed to confirm)

@ 400MHz FSB, 800MHz RAM:

BIOS Set	Memset 4.1
1	No POST code D4
2	No POST
3	No POST
4	No POST
5	No POST
6	No POST
7	No POST
8	No POST
9	No POST
10	10
11	11
12	12
13	13
14	14
15	15
16	0
17	1
18	2
19	3
20	4
21	5
22	6
23	7
24	8
25	9
26	10
27	11
28	12
29	13
30	No POST code D5
31	No POST

I suspect that either:
A) BIOS set PL also adjusts other internal MCH subtimings, which is why 16+ is reported as lower PL than <15
B) MemSet is somewhat being fooled by ASUS BIOS

I will also link photos of the OC section of my P5Q Turbo, as it was tuned to validate 683.3MHz. Note that this is not perfectly tuned, nor should these settings apply perfectly to other ASUS P45 boards. They are simply a point of reference.

These boards do relatively well with quad-cores. This is what I got after about 30 minutes of testing, 550MHz on a random Q9550 & P5Q Turbo. Any more was very difficult, but could be down to me using Auto skews.

ASUS P45 is also known to do relatively well with clocking DDR2. RHex's result is top 1 on air and 4th overall.

MCHBAR edits are currently in the works. This post will be updated as more testing and information is available. So far, RaspberryHex has not found any benefit from addresses FED14320 -> FED14450.

Here are our current best results with these boards:

686.45MHz - P5Q Turbo #1 for ASUS P45 | ML
671.38MHz - P5Q Deluxe #3 " " (3-way tie) | RHex

Here is the current ASUS P45 leaderboard. Not including non-reference frequencies where FSB was clocked higher - if they exist.

It's my opinion that with correct tuning, and taking the CPU & motherboard cold, 700MHz+ is possible on ASUS P45. I have already seen 690MHz on ambient, although it is too unstable to valid (yet).

~~~

P5QC Comments

This board is extremely poorly behaved. Not only was my maximum FSB with the golden E8400 just 610MHz, it clocks RAM extremely poorly. Most of the RAM ratios are simply broken at any frequency (down to 200MHz and up to 500MHz FSB) for both DDR2 and DDR3, regardless of strap. Only the middle RAM ratio selections for 266, 333, and 400MHz worked reliably. I know some people have managed 1730MHz, but I couldn't even get much further than ~1550MHz on DDR3 or ~1100MHz on DDR2. Replacing EBB with M2F or P5Q Deluxe did nothing significant. Old BIOSes clocked the same or worse.

In short, the board is complete e-waste, and is certainly the absolute worst ASUS P45 model ever made. Not worth the time of detailed notes.

------------

Potential future additions: P5Q Pro Turbo, M2F

This post will be continually be updated with new information as we learn more about these boards. Changes will be listed below.

v1.0 - Initial post

v1.1 - Add P5Q Turbo batch & comment on P5QC

[MachineLearning]

Sorry for this post being so long, I want to get the information out there first. Better formatting will come later.

Also want to add, that the P5Q Turbo is compatible with the P5QD Turbo and P5Q Pro Turbo BIOSes. In fact, for OC, they appear identical (same options, tested Pro Turbo BIOS to hit at least 680MHz). Plain P5Q would likely work as well. P5Q-E will boot with a larger BIOS chip, but will throw errors at boot:

[Mr.Scott]

I have P5QC should you need one.

[MachineLearning]

Dec 30, 2023 12:16:49 GMT -5 Mr.Scott said:

I have P5QC should you need one.

I actually grabbed one a couple days ago, should be on its way pretty soon. Still - thank you for the offer.

Hopefully these notes can help out if you want to run it again. How high could you clock DDR3 on it? Not seeing over 1500-1600 from anyone on the bot. Expecting DDR2 to be relatively normal for ASUS P45.

[Mr.Scott]

1600 was about it on DDR3. DDR2 was picky as hell. Only about 1100.
Got 520 pretty easy with 9550 on air though. Never tried higher. Ram is the problem on this board.
Runs both DDR2 and DDR3, but neither very well.

[freeagent]

P45 pushes 9550 that hard? Dagnabbit. I was running an 8600 up to 600 with my RF DDR2 board, and then got the 9550 and was capped to like 471.. boo.

[MachineLearning]

Dec 30, 2023 19:59:50 GMT -5 Mr.Scott said:

1600 was about it on DDR3. DDR2 was picky as hell. Only about 1100.
Got 520 pretty easy with 9550 on air though. Never tried higher. Ram is the problem on this board.
Runs both DDR2 and DDR3, but neither very well.

Thanks - good to know. After I posted that comment I found this valid of >1700, but that's clearly the exception not the rule. valid.x86.fr/2sws7s

freeagent , Gigabyte/ASUS/ASRock P45 all do well with quads. Actually, they can be brought to ~600MHz on a good board. hwbot.org/submission/5435529_thequentincc_reference_frequency_ga_ep45t_ud3r_596.6_mhz
X48 can do okay here too, I think my REX hit >500-520 but it's been a while.

[freeagent]

Kimd of makes me want to put it together now lol..

[twopic]

I used to have p45 premium back in 2019. The board was alright but a month into the purchase of that board and it randomly dies. Doesn't post no matter which ddr2/cpu i put inside it. I ended switching to Asus p35 and had a far more enjoyable experience. hwbot.org/submission/4383538_twopaca_reference_frequency_p5k_premiumwifi_ap_591.38_mhz

Not sure if you experience the same with random dying p45 as i did. But thanks for sharing your data!

[MachineLearning]

Jan 9, 2024 1:28:43 GMT -5 twopic said:

I used to have p45 premium back in 2019. The board was alright but a month into the purchase of that board and it randomly dies. Doesn't post no matter which ddr2/cpu i put inside it. I ended switching to Asus p35 and had a far more enjoyable experience. hwbot.org/submission/4383538_twopaca_reference_frequency_p5k_premiumwifi_ap_591.38_mhz

Not sure if you experience the same with random dying p45 as i did. But thanks for sharing your data!

Nope, haven't had them die on me (fingers crossed). I think you unfortunately got a lemon Sorry to hear it.

But: Glad P35 worked out better for you. ASUS did way better there, their P45 is notoriously fickle and difficult. Their P35 is best in-class.

[twopic]

Jan 9, 2024 7:27:03 GMT -5 MachineLearning said:

Jan 9, 2024 1:28:43 GMT -5 twopic said:

I used to have p45 premium back in 2019. The board was alright but a month into the purchase of that board and it randomly dies. Doesn't post no matter which ddr2/cpu i put inside it. I ended switching to Asus p35 and had a far more enjoyable experience. hwbot.org/submission/4383538_twopaca_reference_frequency_p5k_premiumwifi_ap_591.38_mhz

Not sure if you experience the same with random dying p45 as i did. But thanks for sharing your data!

Nope, haven't had them die on me (fingers crossed). I think you unfortunately got a lemon Sorry to hear it.

But: Glad P35 worked out better for you. ASUS did way better there, their P45 is notoriously fickle and difficult. Their P35 is best in-class.

I had some issues with P35 in terms of which memory divider worked best. But overall my p5k was really fun and I kind of want to get back to 775.

[austin86]

Dec 30, 2023 19:59:50 GMT -5 Mr.Scott said:

1600 was about it on DDR3.

I second this, I seldom got far past 1600 on a 775 board outside the 790i and a small handful of higher end x48 and p45 boards.

[somerandomtechyboi]

have a p5q with bios mod to p5qd turbo and ebb has been swapped to maximus ii formula, from 530 -> 570 fsb max bios boot but highest ive stabilized is only 490, real trash board if you want any decent oc out of it with the exception of ram oc as i was able to do 1520 7-9-6-9 trfc 80 2.38v with xtreem dark 1066c6 somewhat p95 stable but hitting ram stick limit and one of my sticks died so no more 1x3 and for whatever reason ocs worse now with just 2 sticks so i would like to find some rams that can clock even higher on air than nanya d die if possible, i mean ddr3 has a few ics capable of pushing well into 3200+ on air so id assume ddr2 would have something similar for 1600+ just that info is basically non existent and noone ever pushes their ram ocs this far unlike ddr3 where 3000+ is a common sight and there are already lots of ics known to do 3000+ (gdie, 1gbit bfr/dfr, 2gbit bfr/cfr, all the 4gbit hynixes, etc.)

bios is crippled but modding the bios is very easy with just mmtool 3.22 fix bkmod, bios of your board, and maximus ii formula bios to take the ebb off the maximus bios and put in your mobos bios. p5q are socketed bios so very easy to use a ch341a and crossflash, the boards with 1mb chips you can insert 2mb chips and run the higher end bioses, yes ive tested with a p5q premium bios before on my p5q by dropping in a 2mb bios chip from a scrap board and got the extra options like nb gtl but a quirk being insta auto bios recovery every time you boot but this doesnt really matter as you can cancel and go into bios menu, did not see any improvement in fsb nor ram clocking capabilities and actually saw more of a regression compared to that modded p5qd turbo bios, though i didnt really do much research and used giga ep45 values as reference for nb gtl so maybe someone can make these bioses work better than the 1mb bioses on lower end p5q considering the extra options

also these boards are really inconsistent fsb wise as in the past ive gotten a p5q pro thatd only run at like 410-420fsb stable and not boot much past that then a p5q thatd only do 460fsb max in bios, didnt help that i was using a meh p5q pro turbo bios at the time rather than a p5qd turbo with maximus ii ebb but still very inconsistent boards so another reason why id rather avoid these if possible especially the lowend boards

currently if i wanna get somewhat serious with 775 again im considering buying a p5k(r) or a p5k deluxe/premium over a p5q as those seem to clock fsb alot better, but i have no idea how well they oc rams compared to p5q since thats gonna be the main focus, id assume theyd be better than atleast the giga ep45 boards (never got more than 1380 stable with my xtreem dark) but i do wanna shoot for 1600+ so id rather not run into northbridge limit. also wanna buy a giga eg31mf board just because quad slot g31 and maybe itll clock better than my g31m s2c that tops out at only 542fsb especially once voltmodded, also looking for g31 s3g and a giga g33 but those seem to be basically non existent so gonna have to continue the hunt for em, not that fun to push max fsb on boards known for high fsb so id rather oc the balls off lowend g31/g33/g35 boards