Is it really over for Intel®?

[kimsix]

aida64 also not say authoriative .

can try this Intel MLC tool

https://www.intel.com/content/www/us/en/developer/articles/tool/intelr-memory-latency-checker.html

 

[Encrypted11]

Good post on in-mem delta (VDD/VDDQ deltas) and sampling suitability of timings for further mhz scaling.

https://www.hardwareluxx.de/communi...-ram-oc-thread.1306827/page-452#post-30286747

Test with VST VT3 or Google Stressapptest (ubuntu/mint live usb)

to install via terminal in live usb
sudo apt install stressapptest

to run
stressapptest -M 28000 -W -s 3600
if surge test cannot resume work
stressapptest -M 28000 -W -s 3600 --pausedelay 7200

monitor PMIC temp to scale your profile.

edit: for AHBD / AGBD SS/DS will be different deltas. Maybe can ask the expert for the probable range.

 

[Phen8210]

@kimsix

INTC back to $20. After trump announce tariff all US stock fall

 

[matique]

Making progress, hopefully I get 1ns down everyday haha

4K Ultra Preset + DLSS DLAA + Raytracing High - had 21777 prior to the CPU tweaks. Think I can squeeze out more performance if I can get more tighter tuning to CPU.

4K Ultra preset nothing else touched in the settings

I saw on the AMD AM5 thread with similar 4K Ultra preset, almost 10k lower score but about 20fps higher with 7900XTX. I don't understand how this benchmark works at all.

Nice bandwidth! But that latency is rough haha. I'm sub 50ns on aida, which doesn't mean much anyway.

Could you run 1440P on your bench for MHW? I don't have a 4K screen.

 

[fkchon]

Nice bandwidth! But that latency is rough haha. I'm sub 50ns on aida, which doesn't mean much anyway.

Could you run 1440P on your bench for MHW? I don't have a 4K screen.

Yeah I'll run it over the weekend.

I think Veii's custom bios for Z890 OCF isn't stable. I was using 2 days no problem (passed short 1 hour session memory tests) then the next day it won't even let me boot into Windows. I'll have to wait for the official Asrock fix (currently OCF has about 6ns added latency over Apex).

What voltage are you running for 8000C32?

 

[matique]

New ICs dropped by Nanya. Tight trcd & trp but looser CL.

 

[matique]

Yeah I'll run it over the weekend.

I think Veii's custom bios for Z890 OCF isn't stable. I was using 2 days no problem (passed short 1 hour session memory tests) then the next day it won't even let me boot into Windows. I'll have to wait for the official Asrock fix (currently OCF has about 6ns added latency over Apex).

What voltage are you running for 8000C32?

1.67v for 8000c32. I have swapped back to 8000c34 for daily as it requires less voltage at 1.62v. Using greensticks on this board needs a lot of voltage, for me at least, in order to stabilise daily boot to boot stability.

 

[fkchon]

New ICs dropped by Nanya. Tight trcd & trp but looser CL.

Z890 Unify didn't know people actually use those, judging by the number of posts on OCN I thought enthusiasts didn't touch that board.

 

[matique]

Z890 Unify didn't know people actually use those, judging by the number of posts on OCN I thought enthusiasts didn't touch that board.

yeah some folks still do. You can see some of them in the gskill competition right now.

 

[fkchon]

1.67v for 8000c32. I have swapped back to 8000c34 for daily as it requires less voltage at 1.62v. Using greensticks on this board needs a lot of voltage, for me at least, in order to stabilise daily boot to boot stability.

Ahh those are the voltages I'm comfortable running on AM5. I'm not sure if its different with CUDIMMs I don't see anyone running above 1.6v on Arrowlake and they're getting 9200C38. I wonder there is a reason they don't go higher perhaps the CKD multiplier doesn't like high frequency.

 

[watzup_ken]

In the end, I did not support Intel for their new CPU, but bought the B580 GPU instead. Great performance for the price. Of course, the caveat is that one should have a relatively new CPU to get the most out of it. I think last gen like Intel Raptor Lake or Ryzen 7xxx is actually good enough to alleviate the heavier CPU overhead.

 

[Encrypted11]

1.67v for 8000c32. I have swapped back to 8000c34 for daily as it requires less voltage at 1.62v. Using greensticks on this board needs a lot of voltage, for me at least, in order to stabilise daily boot to boot stability.

Not an isolated case. ODT and slope autos start becoming weird after 8000. But these sticks have no problem on Apex.

Guess MSI won Z690, Asus won Z790, ASRock won Z890 in memory tuning.

Ahh those are the voltages I'm comfortable running on AM5. I'm not sure if its different with CUDIMMs I don't see anyone running above 1.6v on Arrowlake and they're getting 9200C38. I wonder there is a reason they don't go higher perhaps the CKD multiplier doesn't like high frequency.

likely stability related and intel MRC + vendor custom tuning hasns't unlocked that speed at a usable level yet.

 

[fkchon]

A bit confusing coming from ASUS boards, AMD boards to Asrock Intel board.

AMD is quite easy just increase VSOC and that usually handles most of the tuning on motherboard side (DRAM separate tuning ofc).

VCCSA is System Agent and it also partially controls the memory controller.
ASUS Intel boards has a separate IMC_VDD for the memory controller.
I can't find any documentation on the ASrock manual, when I search IMC in the manual it comes with 0 results. Did some google, it would appear the ASrock one is called VDD2 Voltage and then there is another one called VDD2mv Voltage.

Quite confusing, just want to make sure I have the correct terminology down before I mess with voltages.

 

[Encrypted11]

Skatterbencher (Pieter-Jan Plaisier) is a good info source than typical influencers to understand the platform.
https://tw.linkedin.com/in/pieter-plaisier


Intel currently has a lot of FIVR based 1.8 VIN voltages to various subsystems and the CPU divides it into granual rails. With CPU GPU current draw going up this is a necessary mitigation to socket/connector/pin burnout which was seen on X3D XOC, and on a broader scale PCs using ATX12V systems with flagship GPUs taking 12V 50A+++. If X3D sram was an afterthought initially turned product, then AMD should pull forward AM6 socket redesign plans to 2025 rather than protect socket compatibility to 2027 and risk more socket burnups esp. with high vSOC (and Current Amps) to be forward looking.

I expect more FIVR high input voltage rails being implemented onto CPUs, even AMD which is abit behind in this aspect and who knows, car automotive systems have already moved away from 14V/12V. It's about time to see ATX18V or ATX24V to mitigate these 18-20AWG PSU wire burning up issues.

 

[fkchon]

Skatterbencher (Pieter-Jan Plaisier) is a good info source than typical influencers to understand the platform.
https://tw.linkedin.com/in/pieter-plaisier


Intel currently has a lot of FIVR based 1.8 VIN voltages to various subsystems and the CPU divides it into granual rails. With CPU GPU current draw going up this is a necessary mitigation to socket/connector/pin burnout which was seen on X3D XOC, and on a broader scale PCs using ATX12V systems with flagship GPUs taking 12V 50A+++. If X3D sram was an afterthought initially turned product, then AMD should pull forward AM6 socket redesign plans to 2025 rather than protect socket compatibility to 2027 and risk more socket burnups esp. with high vSOC to be forward looking.

I expect more FIVR high input voltage rails being implemented onto CPUs, even AMD which is abit behind in this aspect and who knows, cars have already moved away from 14V/12V. It's about time to see ATX18V or ATX24V to mitigate these 18-20AWG PSU wire burning up issues.

Yes I've looked at some of his sources, the problem is his labels are mostly how its shown on ASUS motherboards and the labels seem a little different on Asrock. I will review again.

 

[Encrypted11]

How the RPL / ADL layout was like:

DRAM PMIC Voltages
VDD
VDDQ

CPU VDD2 (ASUS - IMC)
CPU VDDQ TX (CPU Side VDDQ linked rail)

Previously, Mem_VDD and Mem_VDDQ
Now due to the IMC tile, I think CPU VDDQ is moved to MRC_VDDQ with a couple more IMC related rails.

Mem_VDD & Mem_VDDQ had deltas in set voltage
Mem_VDDQ & MRC_VDDQ (CPU_VDDQ back then) had deltas in set voltage

SA is very different in ARL, with the set voltage override being the max VID that can be requested in a number of subsystems.

All I know for now. helping a friend with Z890I Nova. May not have findings for likely over a mth as I don't know fully what's going on, but with a bit of platform knowledge coming from RPL.
There are 2 BCLKs now, SOC and CPU.

 

[Yongkit]

This video mentioned from 25:00 time onwards, higher memory soc will result weaker pcie including GPU & m.2 on North bridge.

 

[fkchon]

How the RPL / ADL layout was like:

DRAM PMIC Voltages
VDD
VDDQ

CPU VDD2 (ASUS - IMC)
CPU VDDQ TX (CPU Side VDDQ linked rail)

Previously, Mem_VDD and Mem_VDDQ
Now due to the IMC tile, I think CPU VDDQ is moved to MRC_VDDQ with a couple more IMC related rails.

Mem_VDD & Mem_VDDQ had deltas in set voltage
Mem_VDDQ & MRC_VDDQ (CPU_VDDQ back then) had deltas in set voltage

SA is very different in ARL, with the set voltage override being the max VID that can be requested in a number of subsystems.

All I know for now. helping a friend with Z890I Nova. May not have findings for likely over a mth as I don't know fully what's going on, but with a bit of platform knowledge coming from RPL.
There are 2 BCLKs now, SOC and CPU.

Thats very helpful thanks.

This video mentioned from 25:00 time onwards, higher memory soc will result weaker pcie including GPU & m.2 on North bridge.

I think its a bit different on Intel platforms, when I make changes to SOC voltage it will always decouple from the GPU, M2 etc, but yeah I don't plan on going too memory soc anyway.

 

[Encrypted11]

Intel since skylake has many asynchronous buses and subsystems. Which is good whereby stability of A doesn't affect B (to a larger extent)

But Toppc's video is cool.
He confirmed as a vendor rep that AMD running fclk 2200 and 1:1 DRAM possibly 6400 tight is much better for using less vSOC resulting in better PCIe bus and IO stability aside from socket burning than following Intel users maxing out Gear 2 speeds at 8xxx-9xxx MHz.

 

[fkchon]

Could you run 1440P on your bench for MHW? I don't have a 4K screen.

I'm just playing around with the bios settings with limited time I have, these settings are done in a minute Asrock Tuning Stage 2, DDR5 8800 Tight timing.

The CPU is insanely good, I just need to get to around 60-62ns and I will be very happy. I do seem to have a weird invisible wall at 9000MTs where its not very stable no matter what voltage I give it, the DIMMs were running cool during benches.

The 4K Score I have shows 94.64fps vs 87.07. I'm not sure whether because now I unlock full 4096 power limit vs default 250? or if its because Arrow Lake just likes Windows 10 that much more than Windows 11.