Exposing my ignorance: I don't know what's going on with my RAM. (Renamed)

DarkSideA8

Gawd
Joined
Apr 13, 2005
Messages
738
I don't necessarily know if it is worth the extra money or not in your case though. But at least it's an option
Well - the goal is to buy a 4k 32 later this year. I know a 3070 is going to struggle with that, but the option to plus up to a 5800 is a definite nice to have.

I started poking about, and even though 5800x 'likes' 3600 / 4000mhz RAM... 2133 appears to be still the default. ( https://premiumbuilds.com/features/zen-3-ram-speeds-benchmark-analysis/ ) -- a fact that quite surprises me. IDKW - but I would have thought that 2133mhz would have been a legacy default, and the 'default' timing / mem speed would move up with each iteration of CPU.

Guessing that is b/c Mem makers are pushing out RAM for multiple generations and allowing users / builders to enable profiles on their own?
 

mda

[H]ard|Gawd
Joined
Mar 23, 2011
Messages
2,023
Keep your RAM at 1.35V, it's totally safe even when doing 2933.

Now that you are stable at 2933, start pushing until it becomes unstable.

Try 3000 next, 3066, 3133.

The CPU and memory WILL affect your minimum frames with some CPU bound gaming (will manifest by stuttering), so it's not a surprise that you can notice the difference in your games.
 

DarkSideA8

Gawd
Joined
Apr 13, 2005
Messages
738
Keep your RAM at 1.35V, it's totally safe even when doing 2933.

Now that you are stable at 2933, start pushing until it becomes unstable.

Try 3000 next, 3066, 3133.

The CPU and memory WILL affect your minimum frames with some CPU bound gaming (will manifest by stuttering), so it's not a surprise that you can notice the difference in your games.
I'm kind of surprised at how much I did not know - and I've been building my own systems for years. This has been a good thread and experience for me.

I've always been of the impression that performance was largely dictated by the amount of RAM and less by the speed (to the point of not being overly concerned about the speed) - but as I read more about it I'm seeing a real benefit to this
 

mda

[H]ard|Gawd
Joined
Mar 23, 2011
Messages
2,023
I'm kind of surprised at how much I did not know - and I've been building my own systems for years. This has been a good thread and experience for me.

I've always been of the impression that performance was largely dictated by the amount of RAM and less by the speed (to the point of not being overly concerned about the speed) - but as I read more about it I'm seeing a real benefit to this
Some CPU architectures are apparently more RAM speed dependent than others especially in certain workloads. Ryzen for one likes fast RAM, although more recent Ryzen generations (3000 and 5000 series in particular) try to mitigate the drawbacks of having slower RAM.

Techspot has some decent RAM scaling speed tests for Ryzen. They test a 3rd gen Ryzen here. It's for 3rd gen though and, you'll notice there is a potential for added performance from moving to 3000mhz to 3600. What more 2133 to 3000?

1st and 2nd gen Ryzens are even more RAM speed sensitive for CPU bound tasks, although the 1000 and 2000 series CPUs can't clock RAM very well. It will be extremely rare for a 2700X to get RAM to 3600, but 3200 is certainly possible with a decent sample CPU and a decent board.

If you can't crank your RAM speed up, look into manually tuning your RAM timings to grab some performance back. The primary timings, TRC, TFAW, TWR, TRFC, the two "SCL", Gear Down Mode and Command Rate 1T settings seem to have the biggest impact on performance. But you'll have to run a stability test to prevent file data corruption when you do decide to try and tweak the timings.

Also, if you have two sticks of RAM, make sure you're running dual channel in the correct slots. It's normally the 2nd and 4th slot away from the CPU.
 
Last edited:

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
I'm running on 40 minutes right now at 2933 with 1.30v, and have Tarkov open. Ran one raid and it 'seemed' like an improvement. Did not notice any tearing.
Congrats 2933 is the rated speed for you MC (Which is why I pointed it out. and why it should be first testing point)

Tou mentioend tearing. The memmory speed will not have any effect on tearing. you you confussing tearing with microstutter /lag?)

Tearing is somthing that happens because your front frambuffer gets switches out ind the midle of a refresh cycle of your monitor, causing the display image to consist of 2 different images, making it appers like ther eis a tear in the picture
https://en.wikipedia.org/wiki/Screen_tearing
The only reason for tearing is lack of syncronisation between the monitor refresh cycle and the frambuffers beeing swapped. and can only be fixed with synchronizing them.

I believe you might be thinking of something else
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
or youre making things up...
Yeah lets just ignore AMD and Intel spec sheet and wikipedia articles because that makes more sense than there was something you didnt know....
You know the links was posted for a reason.
 
Last edited:

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
Some CPU architectures are apparently more RAM speed dependent than others especially in certain workloads. Ryzen for one likes fast RAM, although more recent Ryzen generations (3000 and 5000 series in particular) try to mitigate the drawbacks of having slower RAM.
This is again because just clocking the ram does not only affect the ram (even though that decade old fact is news to pendragon1)
on the first generations of ryzen the memory clock sped was directly tied to the core to core communication part of the CPU/CCX units. So increasing the ram speed made cores able to communicate faster to each other even if you task had very little RAM I/O needs

This was also why my project mercury has the CCX separation method, for thread load distribution. On many workloads you could get the most of the speed up by avoiding the slow CCX communication speeds (the one we raised with faster ram) thereby giving you part of the performance boost you would get with faster RAM.
 
Last edited:

mda

[H]ard|Gawd
Joined
Mar 23, 2011
Messages
2,023
This is again because just clocking the ram does not only affect the ram
Probably unrelated but I'd like to add - yeah, I first really took notice of this on a friend's 7700K that started throttling on his Z170 ITX board (which had crappy VRMs) when RAM speed was raised (Being an ITX hotbox, I'm going to guess that the whole setup was on the brink of throttling anyway). The CPU in his particular setup also apparently got up to 5 degrees hotter when running RAM 2133 vs 2666.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
Probably unrelated but I'd like to add - yeah, I first really took notice of this on a friend's 7700K that started throttling on his Z170 ITX board (which had crappy VRMs) when RAM speed was raised (Being ITX, I'm going to guess that the whole setup was on the brink of throttling anyway). The CPU in his particular setup also apparently got up to 5 degrees hotter when running RAM 2133 vs 2666.

I do not know the motherboard on top of my head so this next part is purely speculation
It could be the BIOS is increasing the voltage to the MC to be stable at 2666 speed and therefore more power goes into the CPU, creating more heat.
This extra load could also be an issue for VRM's heating those op generation a warmer environment for the CPU cooler to work in.

So absolute on point with the fact that things are not just isolated compontents. They all impact each other either in communication, powerload or heat.
 

pendragon1

Fully [H]
Joined
Oct 7, 2000
Messages
26,978
Yeah lets just ignore AMD and Intel spec sheet and wikipedia articles because that makes more sense than there was something you didnt know....
You know the links was posted for a reason.
lol it didnt say what you are saying and fuck wiki, its not a valid source of anything. running higher spec ram is overclocking the ram not the cpu. ps: you come off as insulting in almost all of your posts, thatts a nice high horse youre riding.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
Sorry for going of topic DSA8
But im happy you got the speed up to at least 2933 (funny enough the exact rated speed for the MC)

If you cant get the ram to work at 3200 and if you havent tried it ( i might have missed a post) then increasing voltage to the MC.
I do not know what the bios will call on top of my head but many times I've seen it as "offcore" voltage aka the part of the CPU that is not just the cores for calculations aka the MC as well)
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
lol it didnt say what you are saying and fuck wiki, its not a valid source of anything. running higher spec ram is overclocking the ram not the cpu. ps: you come off as insulting in almost all of your posts, thatts a nice high horse youre riding.
I know you are not saying what im saying.

I have never said running the ram at above rated speed is not overclocking the ram. that is purely something you are adding in on your own
I am saying that forcing the MC to run at above its rated speed by having it communicate faster to the ram IS overclocking of the MC. RAM ratings are not part of my statements
The MC is in the CPU so yeah running a part of the CPU at above rating is overclocking the CPU. despite whatever this is new to you or not.

From someone with the needs to drop the F bomb because you got corrected (with supporting evidence), I'm having a hard time putting value to you high horse comment.
I would think some that thinks he is just right because he says so and provide not supporting evidence vs someone that linked to external verification, is probably the more arrogant person.

Now you have not providing a single technical argument for why increasing the communication between the MC and RAM, is not increasing pat of the CPU. So Im not sure what information you need to understand this.
But her is Intel's Architecture optimization manual
https://www.intel.com/content/dam/w...4-ia-32-architectures-optimization-manual.pdf
In page 2-7 you should be able to see that in fact this device is in the CPU and thereby having it run faster than rated is making part of the CPU run faster than rated.

Offcause if that is not clear we can go look at corsairs take on it as well
https://www.crucial.com/support/articles-faq-memory/understanding-cpu-limitations-with-memory

So there you have it from the manufacturer of both RAM and CPU informing you how that the CPU is part of this memory speed issue. both of them confirming my statement and the wiki article

If you have any questions to my statements please feel welcome to post them. But I sense that you are not really interested in learning how stuff works, but more in just being right on a forum..Which is typically why you end up not knowing stuff in the first place.
Just saying "it not so because i didn't know it" is not really an argument in a technical debate.
 
Last edited:

DarkSideA8

Gawd
Joined
Apr 13, 2005
Messages
738
Sorry for going of topic DSA8
But im happy you got the speed up to at least 2933 (funny enough the exact rated speed for the MC)

If you cant get the ram to work at 3200 and if you havent tried it ( i might have missed a post) then increasing voltage to the MC.
I do not know what the bios will call on top of my head but many times I've seen it as "offcore" voltage aka the part of the CPU that is not just the cores for calculations aka the MC as well)
The XMP offered 'Profile 1' which set the mhz to 3200 at 1.35v. That was not stable.

I have it working at 2933 with an arbitrarily chosen 1.3v. Don't know enough about voltage to know what it should be - but there are suggestions above to try 1.35 and 1.4 as I try incrementally higher mhz ratings for faster speeds. Timings were chosen by referring to the 'supported' page on the Gigabyte site.

Learning as we go!
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
The XMP offered 'Profile 1' which set the mhz to 3200 at 1.35v. That was not stable.

I have it working at 2933 with an arbitrarily chosen 1.3v. Don't know enough about voltage to know what it should be - but there are suggestions above to try 1.35 and 1.4 as I try incrementally higher mhz ratings for faster speeds. Timings were chosen by referring to the 'supported' page on the Gigabyte site.

Learning as we go!


If Im understanding you correctly you are increasing the voltage to the RAM. which is absolutely sane to try. but if that leaves you nowhere and you are unable to find a satisfactory results
You might need to find another volt options in your bios that is for the MC part of THE CPU
vUncore
vI/O
or something along thet line sadly there is no real naming standard for it ( to my knowledge) so it might be tricky to find.

But boosting this voltage might help the MC to better run at higher than 2933 speeds. it is mostly not needed but you could just be that you are unlucky with this CPU.
 

DarkSideA8

Gawd
Joined
Apr 13, 2005
Messages
738
If Im understanding you correctly you are increasing the voltage to the RAM. which is absolutely sane to try. but if that leaves you nowhere and you are unable to find a satisfactory results
You might need to find another volt options in your bios that is for the MC part of THE CPU
vUncore
vI/O
or something along thet line sadly there is no real naming standard for it ( to my knowledge) so it might be tricky to find.

But boosting this voltage might help the MC to better run at higher than 2933 speeds. it is mostly not needed but you could just be that you are unlucky with this CPU.
I'll look into this - and yes I have been tweaking the RAM voltage
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
image try with a visual of what im thinking could be the issue, as mis communication can easily happen on a forum.

MCstuff.png


Grey is CPU
Green is RAM
Orange is the motherbaord traces from the CPU to the RAM

My suspicious is that ram is rated at 3200. the motherboard is rated at 3200 but the MC in the CPU is only at 2933 and might not OC well

Purple is the area you are increasing voltage right now
Blue is the device I think would be worth testing with some extra voltage as that is the component that is running/going to run outside of specs.

No promises though. Just that I've seen cases before that increasing MC voltage was needed for high speed memmory. or sometimes it could evne mean you could lover voltage to ram down to something more nice



Also this is the absolut fun of overclocking. Testing out stuff. Testing stability. Establishing new progress , Revalidation.
 

pendragon1

Fully [H]
Joined
Oct 7, 2000
Messages
26,978
I know you are not saying what im saying.

I have never said running the ram at above rated speed is not overclocking the ram. that is purely something you are adding in on your own
I am saying that forcing the MC to run at above its rated speed by having it communicate faster to the ram IS overclocking of the MC. RAM ratings are not part of my statements
The MC is in the CPU so yeah running a part of the CPU at above rating is overclocking the CPU. despite whatever this is new to you or not.

From someone with the needs to drop the F bomb because you got corrected (with supporting evidence), I'm having a hard time putting value to you high horse comment.
I would think some that thinks he is just right because he says so and provide not supporting evidence vs someone that linked to external verification, is probably the more arrogant person.

Now you have not providing a single technical argument for why increasing the communication between the MC and RAM, is not increasing pat of the CPU. So Im not sure what information you need to understand this.
But her is Intel's Architecture optimization manual
https://www.intel.com/content/dam/w...4-ia-32-architectures-optimization-manual.pdf
In page 2-7 you should be able to see that in fact this device is in the CPU and thereby having it run faster than rated is making part of the CPU run faster than rated.

Offcause if that is not clear we can go look at corsairs take on it as well
https://www.crucial.com/support/articles-faq-memory/understanding-cpu-limitations-with-memory

So there you have it from the manufacturer of both RAM and CPU informing you how that the CPU is part of this memory speed issue. both of them confirming my statement and the wiki article

If you have any questions to my statements please feel welcome to post them. But I sense that you are not really interested in learning how stuff works, but more in just being right on a forum..Which is typically why you end up not knowing stuff in the first place.
Just saying "it not so because i didn't know it" is not really an argument in a technical debate.
oh noes a scary f bomb. sure i guess you can call it that, but its not the normal overclocking that everyone is used to. have at er. give your horse a carrot for me.

ps: way to overcomplicate things, good job, you showed me...
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
oh noes a scary f bomb. sure i guess you can call it that, but its not the normal overclocking that everyone is used to. have at er. give your horse a carrot for me.
I dont' care what everyone is used to, that is not part of the argument.
You statement was that I made it up. So can we now agree it was not made up. you just didnt know it, and didn't understand what was said?.

Or are you ever going to provide some actual technical argument to why it is made up?
 

pendragon1

Fully [H]
Joined
Oct 7, 2000
Messages
26,978
I dont' care what everyone is used to, that is not part of the argument.
You statement was that I made it up. So can we now agree it was not made up. you just didnt know it, and didn't understand what was said?.

Or are you ever going to provide some actual technical argument to why it is made up?
seems like your calling it something they arent. no im not going to dig for things to try and prove you wrong, its already provided. and my chip below is running stock with overclocked RAM, not the cpu.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
seems like your calling it something they arent. no im not going to dig for things to try and prove you wrong, its already provided. and my chip below is running stock with overclocked RAM, not the cpu.
I can help notice that you have yet to answer any questions in you eager to not admit your fault nor provided any technical supporting argument.

But lets take the next goal post move as well.
https://en.wikipedia.org/wiki/Overclocking ( i know you hate wiki for some reasone but lets start here)
"Technically any component that uses a timer (or clock) to synchronize its internal operations can be overclocked"
So I am calling it exactly what it is tecnically.

and for the claim that its is commonly not accepted as so
https://www.urbandictionary.com/define.php?term=overclock
"To overclock is to force some hardware mentioned above to run at faster, often non-intented speeds"
Not specified to CPU cores only

So again not made-up. Not using words incorrectly. Just you not knowing how components works, and being to prideful to admit fault.
 

pendragon1

Fully [H]
Joined
Oct 7, 2000
Messages
26,978
wiki is garbage and lol urban dict. i did say have at er, so carry on feeling superior....
you can save yourself another long winded rant, gonna use that internal ignore list now...
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
wiki is garbage and lol urban dict. i did say have at er, so carry on feeling superior....
you can save yourself another long winded rant, gonna use that internal ignore list now...
Still not a single technical argument. I guess putting the fingers in your ears and yelling "lalalal cant hear you" is a win in his book.
But honestly thank you at least this way i will not have to deal with you incorrect statements towards me again.. its appreciated.



Anyway back on topic or at least related to the topic
The act of boosting the voltage to the memory controller ( or chipset) to gain overclocked stability is not news
Even pre MC-inside-CPU times it was something we did when the buss speed was overclocked. FSB or Chipset to Ram.

The old celeron 360 I could get to run at 540, However due to FSB increases and the memory ratio/multiplier the ram ran to fast for the MC in the chipset.
Giving the old BX northbridge a little notch up in voltage made it stable.


Its really not news how the MC and RAM works... just "new" (decade or so ago) that it is inside the CPU no
 

pendragon1

Fully [H]
Joined
Oct 7, 2000
Messages
26,978
would you stop with the fucking rambling. most of what you have wrote in this thread is pointless rambling and trying to prove yourself "right. "internal ignore" as in "in my head", i dont use the actual ignore list, kids do.
if youre not pimping your shit, youre spouting off about irrelevant shit. im done.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,268
I though you had me on ignore. This does not look like ignoring me.
Please note I simply reply to what you are reaching out to me about.
If you don't like replies don't reach out.

About rambling accusation. My post has ben served with technical explanations and links to references, where as you has just been "Because I say so"
Not sure what you consider rambling on a technical forum.

And off cause I'm trying to prove myself right when someone come in with an incorrect statement that about what I said. That is actually a pretty normal part of a technical debate with different perspective.
That you see putting forth proof as something bad is just more telling about you, and gives good explanation as to why you have yet to put forth a single proof if your statement...

Again: If you don't like replies don't reach out. You invited the debate, not me.
 
Last edited:
Joined
Apr 16, 2018
Messages
957
Theres no way to increase voltage to the memory controller that i know of, at least one that works. If there were im pretty sure it would be all over the internet.
To stabilize your memory DarkSideA8 you either have to increase voltage to the modules themselves (up to 1.45 is perfectly safe) or do what pendragon1 suggested and increase voltage to vsoc or both. Dont be afraid of increasing the voltage to your memory. It isnt an uber fragile cpu, memory is very durable. Even if you accidentally jack the voltage up to 1.7v for a few minutes you wont kill it. As long as you arent running 1.7v thru ddr4 with no active cooling for long periods of time they will more than likely outlive the rest of your rig. If youve tried upping your voltages unsuccessfully you can try loosening up your four primary timings (18-18-18-40 for example) one by one. Again dont worry about loosening your timings. You wont blow up your win10 install. It will either help stability or it wont. If that doesnt cut it you have to take drastic measures and start replacing gear! Or just run them at less than their rated speed ;D
As youve found.
 
Top