Is it really over for Intel®?

[fkchon]

If you really are concerned about ram temps, just block the ram it will run around 2-3c above coolant under full load which is around 5-7w.

never took apart the ram before let along blocking it so quite anxious doing that, but yeah main goal getting MORA for now is to cool the ram, I'll play around with air cooling some more, hopefully there is a sale during Easter holidays (checked their FB there was one in 2024) and their lead time is 5 weeks right now haha

 

[Encrypted11]

So far Raptor Lake vs. Arrow Lake IMC (MCU 116, ME 1895) memory copy on GSAT. Identical Klevv AGBD.
Similar timings, 8000C34 vs 8066C34 (due ARL straps).
Per OCN, more MHz needed on ARL to hit RPL work level

Spoiler

 

[matique]

never took apart the ram before let along blocking it so quite anxious doing that, but yeah main goal getting MORA for now is to cool the ram, I'll play around with air cooling some more, hopefully there is a sale during Easter holidays (checked their FB there was one in 2024) and their lead time is 5 weeks right now haha

like i've said before, you don't need a mora just to cool the ram... a 360 rad for your cpu + ram would be adequate.

 

[fkchon]

like i've said before, you don't need a mora just to cool the ram... a 360 rad for your cpu + ram would be adequate.

yeah I know but to push higher clocks and running at higher voltage I think it definitely helps, like you said above 2-3 degrees of coolant temperature is very enticing. My understanding is air cooling will always be 10 degrees higher than ambient correct?

 

[Encrypted11]

No matter how much cooling, most important is quality of the DIMMs.

Anecdotally the same JEDEC (i.e. AGBD 5600) speed ICs can be as far as 4 CAS apart in terms of v/f scaling. If you have existing rads from older custom loops, reuse it first, hook that on to RAM blocks + heat sink.

Observe your thermal scaling factor at 1.5-1.6v vs temp deltas before further expanding your cooling needs, or if you'd attempt a different kit.

Find out where you're walled first, it may not be cooling (already waterblocked DIMMs) even if its a 360.

 

[fkchon]

I no longer have any custom loop components those were 10 years ago. I've been on AIO since AM3 until now (different kits).

How would I know if its the quality of the DIMMs?
I have this screenshot which may help

Currently I'm running 8800C42 on 1.30V which keeps the thermal stable, I do know if I go 1.5V even if at the same settings, the temperature goes above 50 degrees faster and thats usually when I get a memory related error.

 

[Encrypted11]

I think cold (dram temp) boot test is a good yardstick at least thats how I check and compare against the experts this gen. If someone can stabilize 9333, 95xx / 96xx is usually in their bootable range.

Usually -200 to 400MHz of the max single stick bootable frequency can be stabilized assuming no training/odt/slope issues. Slot B has better signal than A. It depends on how much time you want to put into mem testing. I do max boot test on slot B 1 by 1 and then worse stick match with slot B, better stick in slot A.

max bootable MHz with only primaries set (based on the experts, C38-42 seems to be the gold standard around 9000MHz) at maybe 1.6-1.65v VDD rest autos.

Don't boot into OS though, disable the boot disks if necessary. Unstable memory is bound to break OS. There's a good chance of BIOS corruption requiring rom flashback during experimentation, not exactly for the faint hearted.

From the way it scales likely in the high 30-low 40c's, and how well it trains and boots will usually give you a good yardstick on how well the chip cold scales. If it just walls at a low (sub 1.6 voltage) where it frequency doesn't respond to more voltage, its usually not worth blocking the ram.

 

[Encrypted11]

Also i think I said somewhere on another thread that the thin XMP heat sinks usually only are good enough for 1.4V sustained use thermal density at max. This is not even considering heat generated by stuff like running stressapptest or VST VT3 which drops vmax to possibly 1.25V-1.3V.

Anything beyond requires aftermarket ram sinks and possibly waterblocks. Very normal for things to be unstable no matter how you tweak, past 1.4V or so.

 

[fkchon]

Also i think I said somewhere on another thread that the thin XMP heat sinks usually only are good enough for 1.4V sustained use thermal density at max. This is not even considering heat generated by stuff like running stressapptest or VST VT3 which drops vmax to possibly 1.35V.

Anything beyond requires aftermarket ram sinks and possibly waterblocks. Very normal for things to be unstable no matter how you tweak, past 1.4V or so.

I see.

I think the sticks I got are relatively thick. They are the same as these ones here except its a 8800 kit instead of 8400 kit.

 

[Encrypted11]

Iceman, bitspower, kimdoole all sink the 5-7W under heavy load well at even 1.6V. (Single Sided)

For an instance I'm around 1.6V and VT3 stressapptest runs smoothly. So anything less just works without app instability.

The seemly semi stable feel comes from inadequate cooling past xmp VDD, where revalidation by the user is required based on typical operating DIMM temps, unlike DDR4.

Tests like dangwang, karhu, hci etc people go for at least 20000% because there's not enough load to allow RAM temps to hit the hot steady state quick, lower coverage usually results in false positives on memory stability. Unless the RAMs are on a custom loop with pretty stable coolant temperatures, going with less coverage is not a good idea.

Again, its all about PMIC/SPD temperature targeting with the case/MB PCB temperature. That's why I recommend torching the RAMs with hair dryer (low-medium heat) to the typical operating temps under load before starting the tests. May save you at least an hour of testing each time.

 

[elmariachi]

I no longer have any custom loop components those were 10 years ago. I've been on AIO since AM3 until now (different kits).

How would I know if its the quality of the DIMMs?
I have this screenshot which may help

Currently I'm running 8800C42 on 1.30V which keeps the thermal stable, I do know if I go 1.5V even if at the same settings, the temperature goes above 50 degrees faster and thats usually when I get a memory related error.

Teamgroup sticks are generally better thermally because they have a PMIC thermal pad. If it's Gskill, even at 1.4V will hit close to 60 degrees. I've tested before the stock heatsinks are dog5hit. I torched my heatspreaders and removed it gently. And attached the bitspower heatsinks. Even without fan, the temps are significantly lower than the stock ones.

 

[Phen8210]

Yeah, better avoid G.Skill sticks, I got them for my old rig, and it died a bit over 3 years, if the PMIC wasn't naked, it could have last longer. Nevertheless, I had a lifetime warranty, RMA-ed it to G.Skill Taiwan, they gave me an exchange and I finally got rid of that E-WASTE on Carousell.

For my current rig, i'm using kingbanks, it has decent out of the box heatsinks from what I can feel and see (the temps), and the thermal pad has full coverage according to @KleoZy .

 

[elmariachi]

These are how my 8400C40 sticks looked after tearing down and after blocking it with Bitspower Heatsinks + Gelid Extreme Pads + Thermalright TFX between the two halves of the heatsinks at the top for better thermals.

If you're already on Gskill and want to push your sticks further, you have no option but to tear down the heatsinks. Even running them naked has much better temps. Problem with the Gskill is it uses stupid double sided tape which hardly has any thermal transfer properties. I actually suspect alot of this kits dying or degrading over time is due to heat. I feel most RMA returns is due to their dumb choice not to have a PMIC pad which elevates idle temps and even load temps significantly. Not so bright after all that's for sure.

 

[alauz]

Sounds like at least they learnt from their lesson. Nice upgrade to your CUDIMMS!

 

[elmariachi]

Sounds like at least they learnt from their lesson. Nice upgrade to your CUDIMMS!

Sorry I just checked and G.Skill CKD Rams are still WITHOUT any PMIC thermal pads installed apparently!! Edited my earlier reply for clarity.

Mine is not CUDIMMS though. Just normal 8400C40 sticks which I blocked up and running it with a fan at 8600C38 @ 1.55V. No point for me to run CUDIMMS as I do not intend to move to 285K. CUDIMMs works on bypass mode on earlier platforms like Z790. So no point because you will not be able to use the CKD feature unless you are on Z890 which I am not on and not going to move to. Memory frequency scaling makes no difference for games due to the shitty architecture of arrow lake which have gimped the memory controller. Raptor Lake however had very good gains on playing with memory which translated into actual performance in games and general workloads.

 

[KleoZy]

Yeah, better avoid G.Skill sticks, I got them for my old rig, and it died a bit over 3 years, if the PMIC wasn't naked, it could have last longer. Nevertheless, I had a lifetime warranty, RMA-ed it to G.Skill Taiwan, they gave me an exchange and I finally got rid of that E-WASTE on Carousell.

For my current rig, i'm using kingbanks, it has decent out of the box heatsinks from what I can feel and see (the temps), and the thermal pad has full coverage according to @KleoZy .

Yup.. I checked out their rams. the padding is good and heatsink is decent. If you dont fancy logo and design. these rams will just run. Does not matter Intel or AMD. I have been testing mainly on AMD platform.. sorry I know it's Intel thread but just saying. they just run default XMP or EXPO without issues.

 

[alauz]

Sorry I just checked and G.Skill CKD Rams are still WITHOUT any PMIC thermal pads installed apparently!! Edited my earlier reply for clarity.

Mine is not CUDIMMS though. Just normal 8400C40 sticks which I blocked up and running it with a fan at 8600C38 @ 1.55V. No point for me to run CUDIMMS as I do not intend to move to 285K. CUDIMMs works on bypass mode on earlier platforms like Z790. So no point because you will not be able to use the CKD feature unless you are on Z890 which I am not on and not going to move to. Memory frequency scaling makes no difference for games due to the shitty architecture of arrow lake which have gimped the memory controller. Raptor Lake however had very good gains on playing with memory which translated into actual performance in games and general workloads.

Yeah CUDIMMS if they can use it for re-timing 4 DIMMS to get reliable EXPO/XMP settings, it would be much more attractive, right now 4 DIMMS Slots = JEDEC speeds for DDR5 which is pathetic.

 

[fkchon]

Haven't played my Fender Stratocaster in a while, in fact I never played it since I got this Intel build. When I played on my other AMD build 9800X3D/7950X3D with the ASUS X670E-A I would also get alot of electrical noise picked up from the guitar's pickup (its the loudest when the CPU is half idle ie. watching a video).

Initially I searched up why theres this noise picked up on the strat and not Les Paul Custom with PAF, it seems the pickup is more sensitive to the electromagnetic interference given off from electronics, so I just sit farther away from the PC.

Today I got the strat out and played a bit, 0 noise. Brought the guitar right next to the PC, still 0 noise. I don't know if its Intel systems have better power delivery, or if this Asrock Taichi OCF higher end mobo with more power stages or if its because the Asrock OC Memory Shield sticker.
https://www.asrock.com/mb/Intel/Z890 Taichi OCF/page03.asp

It's a bonus upgrade I didn't expect.

 

[fkchon]

Sorry I just checked and G.Skill CKD Rams are still WITHOUT any PMIC thermal pads installed apparently!! Edited my earlier reply for clarity.

I found the video I saw months back, Gskill actually does have PMIC pads installed for the CKD rams. Its the annoying glue thats the issue, V-color kits just come off without requiring anything.

 

[matique]

Sorry I just checked and G.Skill CKD Rams are still WITHOUT any PMIC thermal pads installed apparently!! Edited my earlier reply for clarity.

Mine is not CUDIMMS though. Just normal 8400C40 sticks which I blocked up and running it with a fan at 8600C38 @ 1.55V. No point for me to run CUDIMMS as I do not intend to move to 285K. CUDIMMs works on bypass mode on earlier platforms like Z790. So no point because you will not be able to use the CKD feature unless you are on Z890 which I am not on and not going to move to. Memory frequency scaling makes no difference for games due to the shitty architecture of arrow lake which have gimped the memory controller. Raptor Lake however had very good gains on playing with memory which translated into actual performance in games and general workloads.

Some kits are starting to have pads on the pmic. One OC friend got the 7200c36 24x2 mdie gskill kit and it had pads for the pmic. Still, feel it's quite stupid they didn't include it in the first place. Not sure if all future sticks will have them.