Rocket Lake 32M Pi Tweaks for Noobs

[Splave]

32M is a blast on rocket lake! Thought we could compile some tips to get everyone started. Feel free to add yours and I will quote you in the OP

• Gear 1 Mode necessary duh
• TRAS no longer limited to 28 go as low as you can. 16-17 should be ok to 4k mhz
• You need blast the SA voltage 1.55-1.65 all day VCCIO_M on the other hand doesn't need Z490 levels 1.35/1.45 has not limited in any way. 
• wazza doesnt help much, maybe 1 second. 
• unlike other gens you can pretty much just torque each timing as low as it will go and see a benefit
• RTL you can try Relaxing channel A and running them at the same timings IE 53/53 54/54 has netted some solid scores
• BCLK is king, run as high a bclk as you can. It will allow you to lower the multiplier of the CPU to a level that you can run at exactly 5005 for the entire run
• Binning for Gear 1 is a thing. The same setup I have cpus that wont boot 3900 and ones that run 4060 on air. 
• a0 might be better than a2 dimms. I wouldnt call this a fact yet but it seems to be leaning this way. 
• IMC needs cold to max out (obvious) 
• take advantage of training lower and raising bclk in OS (usually 2-3bclk is fine)
• leaving cache freq very low on training may help (needs more science)
• Certain microcode version may be more efficient (needs more science)
• per core OC needed for highest clock. Try to change only core 02 as high as it will go. usually gains 50-100mhz depending on the cpu. Also makes it fail less after the bench completes and or when opening cpuz. Right click CPUz to select the high core. 

Updated as of 04/29/2021

Edited August 31, 2021 by Splave

[flanker]

Ha, so this is the reason of my issues tight timings and no post on 3866/3733 (without 2T). 1T only up to 3666 MHz

[speed.fastest]

New motherboard, bios, memory combo, idk if i can improving again, but idk why benchmate is not working properly? Try many os same result.

[_mat_]

Would fit better in the BenchMate support thread, but my guess: You are not using the bundled HWiNFO version. They have to be compatible, otherwise they don't load next to each other. It's a limitation by the SDK/HWiNFO.

[speed.fastest]

29 minutes ago, _mat_ said:

Would fit better in the BenchMate support thread, but my guess: You are not using the bundled HWiNFO version. They have to be compatible, otherwise they don't load next to each other. It's a limitation by the SDK/HWiNFO.

True, not really fit in this thread, but im noob with superpi and benchmate. But i use bundled hwinfo, after checking that, version on hwinfo and on benchmate is not the same, maybe misswriting or something, but on hwinfo its v7.03-4435, but on benchmate it says v7.3-4435.

[Broken APEX49]

I have been suffering from this symptom for some time
11600KF for APEX
Not overclocked
I was turning it for a trial

[zeropluszero]

Subbed

[Splave]

Dual rank seems very odd.

Seems like 4800mhz vs 5000mhz cache is less than .125s where single rank is considerably more. 

anyone else seeing similar?

 

[Guest endpunkt]

On 5/3/2021 at 2:55 AM, GTI-R said:

I have been suffering from this symptom for some time
11600KF for APEX
Not overclocked
I was turning it for a trial

Same error too

[Alex@ro]

17 hours ago, Splave said:

Dual rank seems very odd.

Seems like 4800mhz vs 5000mhz cache is less than .125s where single rank is considerably more. 

anyone else seeing similar?

 

Interesting, what's the difference though at let's say 4000 12-11-11 sr vs dr ? 

[Splave]

 

50/48 4000mhz

12-11-11-18-180 SR 5min04.922

12-11-11-28-280 DR 5min04.422

Edited May 6, 2021 by Splave

[Splave]

Old school Trident Dual Rank "A0" style PCB is flying. If you have any old dual rank b-die its worth trying. Smashes my ripjaws and royal 4000c16 sets. Possible that its coincidence but not probable. 

Gear 1 needs binned to max out. 2 out 12 CPUs I have can do 4000 12-11, only one can do tras under 28. Some wont even train 3800 12-11.

Edited May 10, 2021 by Splave

[Broken APEX49]

It's not for beginners

[Arni90]

On 4/29/2021 at 9:49 PM, Splave said:

• RTL you can try Relaxing channel A and running them at the same timings IE 53/53 54/54 has netted some solid scores

How do I go about doing this on an ASUS M13 Apex?

The only RTL "control" I can find is to enable RTL optimization.

[bl4ckdot]

12 hours ago, Arni90 said:

How do I go about doing this on an ASUS M13 Apex?

The only RTL "control" I can find is to enable RTL optimization.

If I'm not mistaken, thats labelled "Round trip latency" under memory training algorithms.

[Splave]

im not sure on asrock you just type the number in you want 

 

[SparkysAdventure]

sounds like ASRock built diff

[Rauf]

Some experience from me playing around with both 2x8GB and 2x16GB the last few weeks:

• It's hard to find IMC that does good 2x8GB on air (2000+). On LN2 should not be a problem for most cpus. But it's much harder to find IMC that does good with 2x16GB when cold! For example my ES, which has the best IMC so far, can loop PYP at 2x16GB C12-11 very tight at 2050MHz all on air. When full pot IMC tops out at just 2000. The same CPU can do 2x8GB 2200++ C12 on LN2 full pot.
• You have to pay attention to timings. Lower is definitely not always faster. And then balance performance vs stability. For example, with my timings TRFC 180 is worse than 220 for both PYP and 32M. Same goes for some subtimings. Some timings give slightly faster times but ends in crash 9/10 times. For me that is not worth it when another setting passes 10/10 times.
• For me A2 is better than A0/A1 for 2x8GB. But I haven't played so much with A0/A1 because my A2 sets are fantastic. I know A0/A1 likes slightly different timings, so maybe I didn't tweak that enough... For 2x16GB I only have A2. Best 2x16GB kit so far did around 1970MHz C12-11 32M, 2050MHz PYP. Best 2x8GB does 4300+ C12-12 32M and 4400+ PYP.
• 32M with benchmate is really fast on win7. I use the same OS as I do for 3D, so nothing special really. Don't have to do waza either. A good win 7 run can compete with XP for low clock challanges, but it varies more than XP. For full out XP will give you higher clocks, so XP is still better there.
• On ASUS make sure Round Trip Latency is enabled on mem training settings. And use latest SPI bios and maximus tweak mode 2.
• For 32M with 2x16GB on XP higher maxmem than 600 is better.
• When benching cold I would recommend 2x8GB, because unless you have a killer IMC it will have trouble handling 2x16 when cold. At least when you bench full pot. You will most likely loose MHz compared to air, unlike 2x8GB where you can see massive gains.

[flanker]

On 5/2/2021 at 6:55 PM, GTI-R said:

I have been suffering from this symptom for some time
11600KF for APEX
Not overclocked
I was turning it for a trial

lol, I had the same with 11600KF, interesting, no this issue with 11900K and KF...

[bl4ckdot]

Yep, its only affecting RKL 6c. Got this issue too on my 11600k

[Broken APEX49]

23 hours ago, flanker said:

笑、私は 11600KF でも同じでした。興味深いのですが、11900K と KF ではこの問題はありません...

You’re mocking me, aren’t you?

[SparkysAdventure]

3 hours ago, GTI-R said:

You’re mocking me, aren’t you?

I think he meant that he experienced the same Benchmate compatibility issue with his own 11600kf. The 11900k and 11900kf do not have the same problem.

[Broken APEX49]

whatever

[superpatodonaldo]

Hi y'all

Newbie rkl here  looking for some help, 11700K on XII, A0 b-die, air test

VCCIO Mem oc voltage = ? 

VCCIO = ?

VCCSA = ?

on M9A run spi32m with IO SA voltages around 1,2-1,25v, something different now?

[SparkysAdventure]

6 hours ago, superpatodonaldo said:

Hi y'all

Newbie rkl here  looking for some help, 11700K on XII, A0 b-die, air test

VCCIO Mem oc voltage = ? 

VCCIO = ?

VCCSA = ?

on M9A run spi32m with IO SA voltages around 1,2-1,25v, something different now?

VCCIO mem = 1.45, sometimes higher

VCCSA = 1.45 to 1.65

VCCIO = whatever, not important