RAM upgrade importance withe regard to Multi-Threading

[LucShep]

Gskill Ripjaws V 4000Mhz CL18 (F4-4000C18D-64GVK) is most likely Hynix CJR or Micron Rev.E.
There is no Samsung B-die at 4000Mhz 18-22-22-42 - those would be instead 18-18-18-38 or 18-19-19-39 (XMP).

That memory besides the 4000 18-22-22-42 (XMP) @1.40v, should also be able to do 3600 17-19-19-39 (maybe @1.35v).
You can also try 4000 20-24-24-44 (@1.40v) if you want the speed of that kit. 

Always use the "Command Rate" at 2T (for better stability).
And do change the DRAM GEAR MODE to "Gear 1", as explained above, for any memory at or below 4000Mhz on Intel 12th and 13th gen.

Edited July 12, 2023 by LucShep

[BitMaster]

Maybe you can find out which Hynix IC that is and then find the preferred latency values for them.

 

*Luc posted faster...

 

 

Edited July 12, 2023 by BitMaster

[Bucic]

@BitMaster

@LucShep

I'm pretty much through all the testing runs. The result is I cannot get past 3700 MHz 18-22-22-42 701 tRFC even at 1.41 V true voltage. And I'm perfectly fine with it, saved it in UEFI as an OC profile. If you guys could recommend a timing tightening strategy (EDIT: the 3600 17-19-19-39 by Luc is a good reference point) I could give it a try but I'm satisfied with the results so far.
As for the latency values: Thaiphoon states the same the stickers on the packaging do and I made sure to enter those values manually. The mobo did a poor job reading those off.

Now, let's see what's the final Cinbebench R23 score

Now I need to bring back some of those fast booting settings as right now my system boots slower than my 90's rig

Also, the Lenovo ideapad nightmare hit me again. For the life of me I can't enter UEFI by smashing the F1 key. The 'Go To UEFI' (hold power to enter UEFI) doesn't work either. I have to use the 'Recovery options' in Windows for now.

 

Edited July 12, 2023 by Bucic

[dankmaster]

On 7/2/2023 at 11:30 AM, skywalker22 said:

And also make sure you create the pagefile on your the fastest disc.

Curious to know why the consensus seems to be that no matter how much physical RAM is installed, that a pagefile/virtual memory/swap is still necessary?  Perhaps a Windows "feature" to use a slow pagefile despite having unused RAM?  Has anyone observed performance degradation with sufficient RAM but without pagefile?

[LucShep]

1 hour ago, dankmaster said:

Curious to know why the consensus seems to be that no matter how much physical RAM is installed, that a pagefile/virtual memory/swap is still necessary?  Perhaps a Windows "feature" to use a slow pagefile despite having unused RAM?  Has anyone observed performance degradation with sufficient RAM but without pagefile?

Some old programs need a minimum of (at least) 80MB+ pagefile, otherwise they won't start or crash.
The problem is, at this point there is no list of which ones need it, and which ones don't.

So instead of going into that bad lottery (by not using any pagefile), if you have a lot of of RAM memory (64GB or more), and don't want to use an unnecessary big pagefile, then just create a small one for that effect only. For example, I use a pagefile manually set at 1024MB min and 2048MB max. 

Edited July 12, 2023 by LucShep

[Bucic]

Cinebench R23 final results

8100 Old corsair Vengeance 3200 MHz

8240 8290 new GSKILL Ripjaws 3700 MHz 18-22-22-42 701 tRFC 1.41 V true

8200 new GSKILL Ripjaws 3600 MHz 17-19-19-39 701 tRFC 1.41 V true

Conclusion: Those Corsair Vengeance sticks sure were a good bang for the buck! The Ripjaws can only touch them at higher clocks.

 

Edited July 13, 2023 by Bucic

[dankmaster]

Thanks for the response @LucShep.  It's not the size of the pagefile I'm concerned about.  It's the fact that RAM is much faster than virtual memory and I would rather my computer use the faster one when available.

If DCS is the "very old program that needs minimum 800MB pagefile" then I guess this is a moot point.  In any case, I hope to do some testing in the next little while on my system and report the results to all in the hopes of unravelling this mystery...

[LucShep]

1 hour ago, dankmaster said:

Thanks for the response @LucShep.  It's not the size of the pagefile I'm concerned about.  It's the fact that RAM is much faster than virtual memory and I would rather my computer use the faster one when available.

If DCS is the "very old program that needs minimum 800MB pagefile" then I guess this is a moot point.  In any case, I hope to do some testing in the next little while on my system and report the results to all in the hopes of unravelling this mystery...

Nah, DCS itself doesn't need a pagefile IF you have a lot of RAM memory (64GB or more).  

DCS does need it if you're stuck with 32GB or less of RAM. Because it can use all that RAM (on busy missions and in MP) and, once it's fully eaten, it'll look out for more elsewhere to devour as emergency resort... that's right, on the pagefile (and yes, it's much much slower!).  
That's why it's recommended to set the pagefile at 32GB (32768MB) if you have 32GB of RAM memory, and always in the fastest drive.

Edited July 12, 2023 by LucShep

[dankmaster]

5 minutes ago, LucShep said:

and, once it's fully eaten, it'll look out for more elsewhere to devour

Oh the horror!

After having already consumed my time, bank account, RAM, CPU, SSD & GPU you're telling me DCS is still hungry for more? 

[LucShep]

30 minutes ago, dankmaster said:

Oh the horror!

After having already consumed my time, bank account, RAM, CPU, SSD & GPU you're telling me DCS is still hungry for more? 

Yep, I am... its insatiable apetite...   every DCS user is an S&M fetishist with it. 

Edited July 12, 2023 by LucShep

[Hiob]

2 hours ago, Bucic said:

With 3600 MHz @ 1.39V 18-22-22-42 701 tRFC I'm already back and beyond the old 3200 RAM, with 8200 points in Cinebench R23 multicore. Sounds a bit meh but at least I'm no longer worried I paid plenty of money for almost a downgrade.

Have you controlled all the other bios settings that could potentially affect performance. Sometimes a hardware change resets the bios or switches to the backup bios if your mobo has one.

[Bucic]

1 hour ago, Hiob said:

Have you controlled all the other bios settings that could potentially affect performance. Sometimes a hardware change resets the bios or switches to the backup bios if your mobo has one.

The UEFI configuration is unchanged from the Corsair Vengeance times. I'd notice a reset immediately as my Noctua IPPC fans would have spun  from 800 to 2000 RPM

As for other settings, this is my very first UEFI computer. I knew the boomer BIOS by heart. UEFI/AMI - not so much. But my mobo is a basic one. Not many OC settings to speak of and not many I have changed from the stock settings.

Edited July 12, 2023 by Bucic

[BitMaster]

I recommend not to turn off pagefile, there are plenty reasons.

First one, even with 64GB DCS can run into issues. I have seen my DCS hitting 58GB in use and about 30ish GB pagefile. Luckily I had my pagefile at System controlled and it had enough space to do that on a 980Pro. Not saying DCS breaks thru the 64GB barrier with ease, no, not yet, but even with 64GB you can run into very high RAM usage in DCS and a pagefile will safe your a** that day.

With 32GB it's as above, just more common and almost certain on heavy MP that your 32GB will not suffice and a pagefile is your only savior. 

 

I will link a video about RAM I watched lately. It is basic knowledge everyone should be aware off. Like an empty gas tank light and tyre thread, you should know those if you drive a car...

 

 

[Bucic]

2 hours ago, Hiob said:

Have you controlled all the other bios settings that could potentially affect performance. Sometimes a hardware change resets the bios or switches to the backup bios if your mobo has one.

Care to take a look? I myself don't see anything suspicious apart from some settings like the power limits I completely don't understand. I took the opportunity to disable Hyperthreading today, leaving Resize Bar and Virtualization in their default enabled state.

 

 

[Hiob]

As I said, I‘m currently running a Ryzen system. Not really up on speed regarding Intel. 

Unless you‘re running windows sandbox or some virtual machines, I would disable cpu virtualization. Hyperthreading or SMT is usually not a problem for performance.

Perhaps somebody else knows that bios better, but I really don’t see anything suspicious. Leaving most things on auto/default is certainly not a bad choice.

[LucShep]

2 hours ago, Bucic said:

Care to take a look? I myself don't see anything suspicious apart from some settings like the power limits I completely don't understand. I took the opportunity to disable Hyperthreading today, leaving Resize Bar and Virtualization in their default enabled state.
 

Please don't disable Hyper-Threading!  You'll want that ENABLED on that i3 12100F, that's for sure! 
There are no "E-Cores" (Economy Cores) on that particular Intel 12th gen processor, they're all "P-Cores" there (Performance Cores, i.e, the good kind!).
You'll want that CPU's 4-Cores assisted by its 8-Threads (Hyper-Threading) at all times, including for DCS.

Regarding your BIOS settings, it seems good to me (looks like already "optimized defaults"?) but then I confess to only having read it "in the diagonal".... 
I did notice the "Command Rate" in the "Dram advanced settings" set at "Auto".... As I've said before, IMO it should always be set manually at "2N" (same thing as 2T) for best stability (even more so if with manual settings in place) but that's your call.

So, anyways, I see the 3600 17-19-19-39 RAM new settings in place there, does it work nicely?
If yes, how does it compare to your previous RAM working settings?

Edited July 12, 2023 by LucShep

[Bucic]

9 hours ago, LucShep said:

Please don't disable Hyper-Threading!  You'll want that ENABLED on that i3 12100F, that's for sure! 
There are no "E-Cores" (Economy Cores) on that particular Intel 12th gen processor, they're all "P-Cores" there (Performance Cores, i.e, the good kind!).
You'll want that CPU's 4-Cores assisted by its 8-Threads (Hyper-Threading) at all times, including for DCS.

Regarding your BIOS settings, it seems good to me (looks like already "optimized defaults"?) but then I confess to only having read it "in the diagonal".... 
I did notice the "Command Rate" in the "Dram advanced settings" set at "Auto".... As I've said before, IMO it should always be set manually at "2N" (same thing as 2T) for best stability (even more so if with manual settings in place) but that's your call.

So, anyways, I see the 3600 17-19-19-39 RAM new settings in place there, does it work nicely?
If yes, how does it compare to your previous RAM working settings?

 

I'm pretty sure I set 2N. I'll verify, thanks.

  

16 hours ago, Bucic said:

Cinebench R23 final results

8100 Old corsair Vengeance 3200 MHz

8240 new GSKILL Ripjaws 3700 MHz 18-22-22-42 701 tRFC 1.41 V true

8200 new GSKILL Ripjaws 3600 MHz 17-19-19-39 701 tRFC 1.41 V true

Conclusion: Those Corsair Vengeance sticks sure were a good bang for the buck! The Ripjaws can only touch them at higher clocks.

 

 

I might loosen the timings again to hit that 4000 or at least close. What timings should I try initially?

XMP/sticker says 18-22-22-42 701 tRFC

XMP ?misread? by mobo was 19-19-19-43 467 tRFC

 

Funny thing is I was about to bash those sticks because 'they are not stable even at 3600!'... What do you know...
https://forum.dcs.world/topic/330420-crash-on-missing-sound-asset-ed_sound-5760-cant-load-wave/

 

EDIT:

As for hyperthreading - yeah, I ran CapFrameX yesterday and figured the same. See the black and white image. Don't ask

Do you also have some SOP for when ST runs better than MT overall? It seems to be the case on my system, granted I tested it with a simple instant action track file.
 

Edited July 13, 2023 by Bucic

[BitMaster]

Looks ok to me, but as said above..an i3 needs HT due to the lack of cores.

One minor thing you might want or need. USB state in S4/S5 mode basically defines if your USB can charge a phone or tablet while in sleep or even powered off. You might want to enable that. Usually the board's manual states if there is a tool to install in Windows to enhance this with even more wattage ( Asus does with AiCharger or similar name and my GB board does it too, I guess MSI has an equivalent tool/function.

 

Edited July 13, 2023 by BitMaster

[LucShep]

2 hours ago, Bucic said:

I'm pretty sure I set 2N. I'll verify, thanks.

  

I might loosen the timings again to hit that 4000 or at least close. What timings should I try initially?

XMP/sticker says 18-22-22-42 701 tRFC

XMP ?misread? by mobo was 19-19-19-43 467 tRFC

 

Funny thing is I was about to bash those sticks because 'they are not stable even at 3600!'... What do you know...
https://forum.dcs.world/topic/330420-crash-on-missing-sound-asset-ed_sound-5760-cant-load-wave/

 

That motherboard seems overly sensiitive... or the RAM. One of the two for sure.

19-19-19-43 is too tight for anything over 3600Mhz on that Hynix memory, even at 1.41v, I think.

Just curious, if you load XMP on BIOS, and then put all "Dram advanced settings" at Auto (exception to "Command Rate" , always 2N), then Save and Exit.....
....what does the motherboard choose for timings on next reboot (if it doesn't crash, that is) ?

Edited July 13, 2023 by LucShep

[Bucic]

1 minute ago, LucShep said:

That motherboard seems overly sensiitive... or the RAM. One of the two for sure.

19-19-19-43 is too tight for anything over 3600Mhz on that Hynix memory, even at 1.41v, I think.

Just curious, if you load XMP on BIOS, and then put all "Dram advanced settings" at Auto (exception to "Command Rate" , always 2N), then Save and Exit.....
....what does the motherboard choose for timings on next reboot (if it doesn't crash, that is) ?

It seems it shoves the RAM into the following:

"XMP ?misread? by mobo was 19-19-19-43 467 tRFC "

[LucShep]

21 minutes ago, Bucic said:

It seems it shoves the RAM into the following:

"XMP ?misread? by mobo was 19-19-19-43 467 tRFC "

That's odd for sure. 

Anyways, try more relaxed timings if you wish to keep 4000 MHz RAM speed, for example 20-24-24-44 (@ 1.40v or 1.41v) as a baseline, still plenty good for 4000 DDR4.

That said, 3700 MHz 18-22-22-42 is actually not bad at all on Intel. If that works rock solid, keep it I'd say.

Edited July 13, 2023 by LucShep

[Bucic]

For now it fails 3800 1.41V

both 21-21-21-45 and 21-24-24-45

Edited July 13, 2023 by Bucic

[LucShep]

16 minutes ago, Bucic said:

For now it fails 3800 1.41V

both 21-21-21-45 and 21-24-24-45

 

hmmmm  still find that odd.

Like said, 3700 MHz 18-22-22-42 is actually not bad at all on Intel. If that works rock solid, keep it.

Notice that your other choice of Gskill Ripjaws V (F4-3600C18D-64GVK, also based on Hynix memory) was 3600 18-22-22-42, so....  

Edited July 13, 2023 by LucShep

[Bucic]

17 minutes ago, LucShep said:

hmmmm  still find that odd.

Like said, 3700 MHz 18-22-22-42 is actually not bad at all on Intel. If that works rock solid, keep it.

Notice that your other choice of Gskill Ripjaws V (F4-3600C18D-64GVK, also based on Hynix memory) was 3600 18-22-22-42, so....  

 

Yeah, I'm staying with the 3700 which gives 3290 cinbenches over the old 3100 cinebenches I too find that weird and I'm also leaning toward blaming it on the motherboard.

[Bucic]

New relevant info?

Quote

Nice-Dog-7904

· 8 mo. ago

Intel 12, 13th gen non-K CPUs have locked SA voltages. Even if you give more voltage to the ram in the bios, the system actually doesn't. It seems that the limit of ram overclock on these non-K CPUs are 3600mhz. Most people will fail to reach 3600, while 3500 is completely possible. The very rare few who did succeed 3600mhz are the ones who won the memory controller lottery.

https://www.reddit.com/r/MSI_Gaming/comments/uzissc/does_ram_xmp_profile_issue_still_persist_on_the/

That is exectly what I have on my to-do list. Leave 1.4 SET voltage, bring back the frequency to 3600 and tighten the timings. Some day...

 

EDIT:
shiiiiiii@#$@#

Quote

TLDR: important voltage for memory OC is locked on 12th gen non K. Don’t expect anything above 3600MT/s G1 stable on those SKU’s.

https://www.reddit.com/r/intel/comments/sluk4z/psa_12th_gen_non_k_skus_have_locked_vccsa_voltages/

 

Edited July 15, 2023 by Bucic