INTEL CPUs *OC Benchmark & Discussion*

[matique]

Possible solutions for Intel's gen 13 and gen 14 Core desktop processors crashing in Unreal Engine games​

https://www.notebookcheck.net/Possi...crashing-in-Unreal-Engine-games.805728.0.html

4096w/a is out of intel spec though, even on MSI boards you would have to manually toggle the limits to be turned off for that to occur. If pure stock this should not be an issue.

If you turned limits off, it's on you to test your stability. That's why the seasoned intel folks here do hard hitting thermal loads on our CPUs, such as running VST/VT3 from ycruncher. If it can survive that it'll survive any game.

 

[Encrypted11]

Quite honestly primary issue is lack of memory stability testing by user on either platforms. It is unlike DDR4, now more motherboard components are offloaded to the RAM modules as the small board designs in notebooks/AIO/desktops are running out of space.

Ever since, XMP certified =/= stable since DDR5 with PMIC thermal management being a high priority matter. With proper testing, those annual OS reinstalls / Windows Update broke my OS etc type of stuffs (game instability aside) are really non existent.

 

[Encrypted11]

Testing the newer VDDQ-VDDQ and in-mem delta methods. In later OC discussions Veii, tibcsi etc (well ahead of old rules posted by shamino).

Start with a baseline 7200MT/s C34 GSAT 2H profile on the 96GB. Probably will tighten and stop, wait for next BIOS before scaling up MT/s since it is very time consuming and my 7200MT/s is usable. Plenty of voltage headroom on CPU side (same old 13700K)

SA: 1.160V
VDD2: 1.220V
VDDQ CPU: 1.165V
VDDQ Mem: 1.335V (170mV VDDQ-VDDQ Delta)
VDD Mem 1.425V

With the new methods it looks like I can support 8200MT/s on 24GB SRs at 1.160V SA. Min 50-100 pages of reading/collecting data on hardwareluxx before starting. The discussion is now way ahead of OCN, patterns on RTT/ODT, slopes, skews, are already discovered instead of brute forcing voltages of the past.

 

[Mach3.2]

8200MT/s on Z690 board is very impressive.

 

[kimsix]

Testing the newer VDDQ-VDDQ and in-mem delta methods. In later OC discussions Veii, tibcsi etc (well ahead of old rules posted by shamino).

Start with a baseline 7200MT/s C34 GSAT 2H profile on the 96GB. Probably will tighten and stop, wait for next BIOS before scaling up MT/s since it is very time consuming and my 7200MT/s is usable. Plenty of voltage headroom on CPU side (same old 13700K)

SA: 1.160V
VDD2: 1.220V
VDDQ CPU: 1.165V
VDDQ Mem: 1.335V (170mV VDDQ-VDDQ Delta)
VDD Mem 1.425V

With the new methods it looks like I can support 8200MT/s on 24GB SRs at 1.160V SA. Min 50-100 pages of reading/collecting data on hardwareluxx before starting. The discussion is now way ahead of OCN, patterns on RTT/ODT, slopes, skews, are already discovered instead of brute forcing voltages of the past.

Ya not sure what is Intel equivalent, for zen4, these termination values and vddp phy voltages helps ddr5 overclock. 100% boot everytime.

my sample like lowering vddp somemore. seems is finding your components signaling sweet spot.

increasing vdd and vddq >1.5v actually make my unbootable.

somemore our SG temps, inside case already start at +42c @ 1.48v vdd/q..

 

[matique]

Ya not sure what is Intel equivalent, for zen4, these termination values and vddp phy voltages helps ddr5 overclock. 100% boot everytime.

my sample like lowering vddp somemore. seems is finding your components signaling sweet spot.

increasing vdd and vddq >1.5v actually make my unbootable.

somemore our SG temps, inside case already start at +42c @ 1.48v vdd/q..

That's why for us who oc pretty high, just waterblock the ram hahaha to get rid of any temperature issues.

Testing the newer VDDQ-VDDQ and in-mem delta methods. In later OC discussions Veii, tibcsi etc (well ahead of old rules posted by shamino).

Start with a baseline 7200MT/s C34 GSAT 2H profile on the 96GB. Probably will tighten and stop, wait for next BIOS before scaling up MT/s since it is very time consuming and my 7200MT/s is usable. Plenty of voltage headroom on CPU side (same old 13700K)

SA: 1.160V
VDD2: 1.220V
VDDQ CPU: 1.165V
VDDQ Mem: 1.335V (170mV VDDQ-VDDQ Delta)
VDD Mem 1.425V

With the new methods it looks like I can support 8200MT/s on 24GB SRs at 1.160V SA. Min 50-100 pages of reading/collecting data on hardwareluxx before starting. The discussion is now way ahead of OCN, patterns on RTT/ODT, slopes, skews, are already discovered instead of brute forcing voltages of the past.

Will be starting on my 8400c34 tune soon on these 16*2 adie greens, see if i could get it stable. It boots no issue, but gotta play with some RTT/ODT and skews. These sticks still need high juice though, anything below 1.6v and it won't even go past 30mins on karhu.

8200MT/s on Z690 board is very impressive.

Z690i unify was one of the best boards for the generation, only losing to binned Z690 dark KPE & apex. Z790i Edge carries the same trait, so far a very very solid board.

 

[NightRaven49]

 

[elmariachi]

Sold my old setup to a friend who wanted a white build and got a 14900KS, Encore Board and some 2nd hand rams from someone who was selling his 8400C40 kit. Delidded my 14900KS and was just playing around to see what is the limit of the CPU and board. I think my board can more than handle 8800C38 but my CPU has the infamous SA bug at 1.23V so I am currently running it at 1.22V. At 1.23V and above, TM5 will simply just lock up the system even though there are no errors.

So final DDR5 OC looks to be 8600C36! 8800C38 will need at least 1.3V SA which makes it impossible for me to get it to run without locking up. I probably can use the same voltages with some minor adjustments to timing to run 8800C38 if my CPU did not have this dumb SA bug. Seems others have it worse with their golden CPUs where SA bugs kicks in at 1.2V and they need to run below that which makes even 8600 or maybe 8400 not possible to run.

Voltages:

VDD - 1.62V
VDDQ - 1.6V
MC - 1.45V
TX - 1.35V

Skews:

DQ RTT NOM RD 80
DQ RTT NOM WR 80
RON ODT 48-40-40-48
tXSR 746
tXP 32
PPD 2

Max temps during TM5 - 37.5 degrees with a small 80mm fan. The Bitspower heatsinks are really good with Gelid Extreme pads and some Thermalright TFX paste between the sinks. I was thinking at 1.62V, temps would be well past 40 degrees but it only stood at 37.5 degrees tops. It runs much cooler than my 16Gbx2 7800Mhz A-Die sticks. Those sticks at 1.57V VDD/VDDQ is already at 41 degrees after running TM5. Exact same pads and heatsinks too.

 

[Koenig168]

water-cooled RAM

 

[elmariachi]

water-cooled RAM

No man my rams not watercooled. Just Bitspower heatsinks and a **** 80mm fan on the test bench. Max temps 38 degrees with 1.62v VDIMM.

 

[xonix]

Pity those unsuspecting people who buy 2nd hand pcs

Intel announces two extra years of warranty amid chip crashing and instability issues — longer warranty applies to 13th- and 14th-Gen Core processors
https://www.tomshardware.com/pc-com...-applies-to-13th-and-14th-gen-core-processors

 

[Encrypted11]

This might be a good way to affix RAM sinks with thermal putty getting more common.
No longer need to worry about pad thickness, cavitation, air gaps.

For single side RAMs, possibly 1.5-2mm pads on empty side of PCB, as adhesive/offset/spacer for the IC/PMIC side.
For DS PCB, use equal thickness pads as the spacer/epoxy.

Thermal pads used to affix heat sink can be substituted with temp resistant double sided foam tapes
(similar to what many stock heat sinks use)

How to clean off? Old (dry) toothbrush, IPA, cotton buds/wipes/pads etc.

Putty is very reusable until it has absorbed too much dust, and likely abit more eco friendly
(without the need for the plastic peeloffs, unusable small pad scraps being disposed.)

But the downside, putty is softer.
During removal of rams holding the edges of PCB than pulling the heatsink will be more ideal to prevent runout.

 

[elmariachi]

This might be a good way to affix RAM sinks with thermal putty getting more common.
No longer need to worry about pad thickness, cavitation, air gaps.

For single side RAMs, possibly 1.5-2mm pads on empty side of PCB, as adhesive/offset/spacer for the IC/PMIC side.
For DS PCB, use equal thickness pads as the spacer/epoxy.

Thermal pads used to affix heat sink can be substituted with temp resistant double sided foam tapes
(similar to what many stock heat sinks use)

How to clean off? Old (dry) toothbrush, IPA, cotton buds/wipes/pads etc.

Putty is very reusable until it has absorbed too much dust, and likely abit more eco friendly
(without the need for the plastic peeloffs, unusable small pad scraps being disposed.)

But the downside, putty is softer.
During removal of rams holding the edges of PCB than pulling the heatsink will be more ideal to prevent runout.

Nice! But the only problem with putty is that it is an arse to clean off compared to normal thermal pads. I think if you get proper heatsinks, using thermal pads is really non-issue. I've been using my Bitspower heatsinks and if you use the right pad thickness, it works incredibly well. At 1.55V VDD/VDDQ, I hardly cross 40 degrees on my ram temps during Tm5 runs with a small 60mm fan over it. Gaming barely hits 35 degrees. One thing is for sure, Gskill heatsinks is about as useful as a condom in a convent. Thrash. Wished more manufacturers will sell naked pcbs like Hynix or Klevv without all that rgb crap.

BTW what heatsinks are those? Looks dope.

 

[Encrypted11]

Iceman from Taobao, recommended by @matique !

 

[elmariachi]

Iceman from Taobao, recommended by @matique !

Good set of heatsinks. Hopefully CAMM2 will drastically change and simplify memory cooling. Looks pretty interesting so far!

 

[Phen8210]

Very very good

This might be a good way to affix RAM sinks with thermal putty getting more common.
No longer need to worry about pad thickness, cavitation, air gaps.

For single side RAMs, possibly 1.5-2mm pads on empty side of PCB, as adhesive/offset/spacer for the IC/PMIC side.
For DS PCB, use equal thickness pads as the spacer/epoxy.

Thermal pads used to affix heat sink can be substituted with temp resistant double sided foam tapes
(similar to what many stock heat sinks use)

How to clean off? Old (dry) toothbrush, IPA, cotton buds/wipes/pads etc.

Putty is very reusable until it has absorbed too much dust, and likely abit more eco friendly
(without the need for the plastic peeloffs, unusable small pad scraps being disposed.)

But the downside, putty is softer.
During removal of rams holding the edges of PCB than pulling the heatsink will be more ideal to prevent runout.

Thanks for sharing

 

[matique]

This might be a good way to affix RAM sinks with thermal putty getting more common.
No longer need to worry about pad thickness, cavitation, air gaps.

For single side RAMs, possibly 1.5-2mm pads on empty side of PCB, as adhesive/offset/spacer for the IC/PMIC side.
For DS PCB, use equal thickness pads as the spacer/epoxy.

Thermal pads used to affix heat sink can be substituted with temp resistant double sided foam tapes
(similar to what many stock heat sinks use)

How to clean off? Old (dry) toothbrush, IPA, cotton buds/wipes/pads etc.

Putty is very reusable until it has absorbed too much dust, and likely abit more eco friendly
(without the need for the plastic peeloffs, unusable small pad scraps being disposed.)

But the downside, putty is softer.
During removal of rams holding the edges of PCB than pulling the heatsink will be more ideal to prevent runout.

Very nice! Good advice and it seems like super great contact. I used putty before, on my 3090 back during mining season lol. But it cleaning it off was really annoying for me, even with ipa and toothpick. I'm happy enough with the temps I get from pads.

 

[elmariachi]

https://wccftech.com/intel-core-ult...xFYU6A5nWuGt2_Emdc_aem_5kI7BRpBIwgBd7NE9MLGKg

Looking pretty sweet. Just have to see how much more efficient and the rumours of it using 100w less than Raptor Lake turns out. If clocks are at 5.7Ghz without HT, these results could mean that IPC improvements is sizeable.

Looks like Intel may have finally enabled DVLR on Arrow Lake! Finally P and E cores get a seperate voltage rail. Not sure if ring bus will get their own voltage rail or is shared. This still could be speculation so take it with a pinch of salt.

 

[Encrypted11]

No thunderbolt 5 except for pcie cards though, and oregon instead of haifa

 

[memoriess]

just want to check if i am not able to wait for intel 15th gen, can I go for 14th gen one?
Does it still have stability issues?