Trying to understand Ryzen 3000 series boost speed variations

[Zach_01]

Yes I know you are trying to learn as we all do the same thing because the Zen2 CPUs are fairly new chips/architecture. I like to observe and learn about them too.

VRM MOS temp has max value way over 100C. In fact, most of these mosfets have throttling points from 115 to 125C even the crapiest ones, so do not worry about it.

 

[lorry]

Yes I know you are trying to learn as we all do the same thing because the Zen2 CPUs are fairly new chips/architecture. I like to observe and learn about them too.

VRM MOS temp has max value way over 100C. In fact, most of these mosfets have throttling points from 115 to 125C even the crapiest ones, so do not worry about it.

Oh I know that they can go pretty high, seen plety of BZ breakdown videos of VRMs, all I was saying really was that the cpu temp wasn't being affected by anything else, as nothing else was under any stress.

One interesting thing I just noticed, every single effective thread today has a Min high of 4250MHz. That is around 200MHz higher than any of my other screenshots of HWinfo. The highest in fact is 4,373MHz !

 

[Zach_01]

Oh I know that they can go pretty high, seen plety of BZ breakdown videos of VRMs, all I was saying really was that the cpu temp wasn't being affected by anything else, as nothing else was under any stress.

Understood...

One interesting thing I just noticed, every single effective thread today has a Min high of 4250MHz. That is around 200MHz higher than any of my other screenshots of HWinfo. The highest in fact is 4,373MHz !

Can’t really say why...

I still see you have a bit high avg CPU usage (5.2%). I know that was after some 100% loads but still it’s 14+hours avg. I guess you run a lot of tests during that time.

 

[lorry]

Understood...


Can’t really say why...

I still see you have a bit high avg CPU usage (5.2%). I know that was after some 100% loads but still it’s 14+hours avg. I guess you run a lot of tests during that time.

Not really, maybe 4 or 5 CB R29 but Firefox has a Ton, as in 40+ tabs open!

 

[Zach_01]

Yeah... I also have minimum about ~25 tabs open (Chrome). At least on chrome don’t put any stress on the CPU, just Ram space.

 

[lorry]

Yeah... I also have minimum about ~25 tabs open (Chrome). At least on chrome don’t put any stress on the CPU, just Ram space.

no idea re Firefox

Watching ;
3rd-gen Ryzen Overclocking - Everything you need to know!
by that tubby guy from linus tech tips (forget his name right now)

So far he has said this -

"
for high speed memory enable gear down mode
disable power down mode

Manually change every voltage that you can to stock value"

 

[tabascosauz]

Oh I know that they can go pretty high, seen plety of BZ breakdown videos of VRMs, all I was saying really was that the cpu temp wasn't being affected by anything else, as nothing else was under any stress.

One interesting thing I just noticed, every single effective thread today has a Min high of 4250MHz. That is around 200MHz higher than any of my other screenshots of HWinfo. The highest in fact is 4,373MHz !

View attachment 139122

Are you applying a undervolt? Those voltages look okay to me. Maybe you have a fair bit of room for undervolting.

MOSFET temps don't scale like CPU temps. You'll see them start to creep up when applying full loads like P95 for an extended period of time. A short bench usually won't cause them to get too hot. Less than 50 is ice cold!

 

[lorry]

Are you applying a undervolt? Those voltages look okay to me, maybe even slightly on the low side.

MOSFET temps don't scale like CPU temps. You'll see them start to creep up when applying full loads like P95 for an extended period of time. A short bench usually won't cause them to get too hot. Less than 50 is ice cold!

no, every voltage is as the bios set them

Just seen this though, which vindicates what I was saying about 1.4V being average

 

[tabascosauz]

no, every voltage is as the bios set them

Just seen this though, which vindicates what I was saying about 1.4V being average

View attachment 139124

I hate general "recommendations" like that for Ryzen 3000 because of how dynamic Vcore is on this generation, and how each chip is of different quality. What is be interested in knowing is your SVI2 TFN voltage under maximum possible load, as it will drop a fair bit.

 

[lorry]

I hate general "recommendations" like that for Ryzen 3000 because of how dynamic Vcore is on this generation, and how each chip is of different quality. What is be interested in knowing is your SVI2 TFN voltage under maximum possible load, as it will drop a fair bit.

would R20 do that? As I can quickly run that now

 

[Zach_01]

no idea re Firefox

Watching ;
3rd-gen Ryzen Overclocking - Everything you need to know!
by that tubby guy from linus tech tips (forget his name right now)

So far he has said this -

"
for high speed memory enable gear down mode
disable power down mode

Manually change every voltage that you can to stock value"

View attachment 139123

Did not read it but I suspect is for a static CPU OC, right?
If yes then it does not apply to me or anyone leaving the CPU with stock boosting.
As for Gear/Power down settings yes I knew that but it is more for 3600+ Mem speed.
RyzenDRAMcalc also suggest these settings according to Speed/timings

 

[lorry]

Did not read it but I suspect is for a static CPU OC, right?
If yes then it does not apply to me or anyone leaving the CPU with stock boosting.
As for Gear/Power down settings yes I knew that but it is more for 3600+ Mem speed.
RyzenDRAMcalc also suggest these settings according to Speed/timings

ah! okay
and yes for a static OFC

I hate general "recommendations" like that for Ryzen 3000 because of how dynamic Vcore is on this generation, and how each chip is of different quality. What is be interested in knowing is your SVI2 TFN voltage under maximum possible load, as it will drop a fair bit.

R20 obviously wasn't enough, as the SV12 TFN didn't move from 1.087V during R20

Any other suggestions re a test?

 

[tabascosauz]

would R20 do that? As I can quickly run that now

If it stresses all 24 threads heavily, then I guess it would. Gets kinda difficult to tell with voltage on these bigger SKUs because Ryzen boost will call for different Vcore depending on the load (even if they all display "100% usage") and the amount of cores being utilized. Highest Vcore will be seen under light, almost "idle" load, and lowest will be the heaviest load across the most cores.

The voltage offset you can apply is applied all the time, so regardless of what Vcore the chip is calling for at that instant, the same offset will be applied.

P95 Small or smallest? But some people don't like P95 as they believe it "unrealistic loads" and causes a lot of heat and power draw

 

[lorry]

If it stresses all 24 threads heavily, then I guess it would. Gets kinda difficult to tell with voltage on these bigger SKUs because Ryzen boost will call for different Vcore depending on the load (even if they all display "100% usage") and the amount of cores being utilized. Highest Vcore will be seen under light, almost "idle" load, and lowest will be the heaviest load across the most cores.

The voltage offset you can apply is applied all the time, so regardless of what Vcore the chip is calling for at that instant, the same offset will be applied.

P95 Small or smallest? But some people don't like P95 as they believe it "unrealistic loads" and causes a lot of heat and power draw

I ran the Membench within Dram Calc for 4 mins and it dropped no lower than 1.081V

No idea quite what these memtest results show, but here they are





I understand that 74 is my current latency, or I think it is, and that 260 is my current time and the best is 230. Would that then mean that using this memtest my PC of ram needs to be as close to 230 as I can get it?
And the graph? Any ideas on that?

 

[Zach_01]

The important thing is to finish it without errors, as you tweak the memory of course. Raising speed and lower timings.
Yes the 230sec is probably the best 3200 ram that the calc has in its data base for this 24threaded CPU.
If you change speed (ex. 3600) the "best" result would be something else (lower)

 

[lorry]

The important thing is to finish it without errors, as you tweak the memory of course. Raising speed and lower timings.
Yes the 230sec is probably the best 3200 ram that the calc has in its data base for this 24threaded CPU.
If you change speed (ex. 3600) the "best" result would be something else (lower)

How good is this compared to memtest86?
As I presume that once stable at 3600 I do this test and if it passes is there still a need to do the full memtest86?
I presume it's best to ensure complete stability?

 

[Deleted member 185158]

Hey Lorry,

Does your motherboard have a setting called SenseMi Skew and SenseMi offset by any chance?
If it does, I can help you with an alternative to boosting affects for the cpu.... (After you get your ram sorted out of course)

 

[lorry]

Hey Lorry,

Does your motherboard have a setting called SenseMi Skew and SenseMi offset by any chance?
If it does, I can help you with an alternative to boosting affects for the cpu.... (After you get your ram sorted out of course)

Offhand I don't know, but I don't recall seeing either as an option or heading anywhere. Where might it be under?

 

[Deleted member 185158]

Offhand I don't know, but I don't recall seeing either as an option or heading anywhere. Where might it be under?

On my Asus ROG board, it's in the tweaker's paradise section. I'm not familiar with your board, but would be in perhaps advanced section?? (not really sure, you'll have to look)

Give you some nifty tips and tricks. These processors do amazing things when altering these settings. You'll see a lot less idle time, higher P-states maintained.
I'll explain more if you have these settings, I hope you do. So much tweaking Ryzen chips.... just gotta have the right board. (not saying you don't just not familiar with yours)

 

[lorry]

On my Asus ROG board, it's in the tweaker's paradise section. I'm not familiar with your board, but would be in perhaps advanced section?? (not really sure, you'll have to look)

Give you some nifty tips and tricks. These processors do amazing things when altering these settings. You'll see a lot less idle time, higher P-states maintained.
I'll explain more if you have these settings, I hope you do. So much tweaking Ryzen chips.... just gotta have the right board. (not saying you don't just not familiar with yours)

I'll check tomorrow.
I intend to boot into the bios first thing tomorrow and just familiarise myself with where all the various ram settings are, so looking for a couple more would be best done then

Just did a search online for it And gigabyte board but nothing showed. I'll find out properly tomorrow

 

[Deleted member 185158]

No hurries.

Don't be terribly picky with the memory timings. Tighter is better, but often times sacrifice speed. If you can do a clean 16-16-16-16-x-x 1 or 2T 1.40v or less, 3600mhz that's a great place to be.
Performance and latency happens from Raw speed, tighter timings only enhance this.

Memtest is really good for testing memory. But like with anything, it's not the "tell all" just like P95 or OCCT by themselves. Do test with multiple testers and for sure take the time to enjoy the rig and do some gaming. Testing does require some gaming testing you know. Benchmarks/stress tests (synthetic or not) are pretty straight forward. But doesn't simulate actual usage, like pause game Alt/Tab, open browser, throw on some music and so forth.

Unless you intend to make some scores, don't over think all this k?

 

[lorry]

No hurries.

Don't be terribly picky with the memory timings. Tighter is better, but often times sacrifice speed. If you can do a clean 16-16-16-16-x-x 1 or 2T 1.40v or less, 3600mhz that's a great place to be.
Performance and latency happens from Raw speed, tighter timings only enhance this.

Memtest is really good for testing memory. But like with anything, it's not the "tell all" just like P95 or OCCT by themselves. Do test with multiple testers and for sure take the time to enjoy the rig and do some gaming. Testing does require some gaming testing you know. Benchmarks/stress tests (synthetic or not) are pretty straight forward. But doesn't simulate actual usage, like pause game Alt/Tab, open browser, throw on some music and so forth.

Unless you intend to make some scores, don't over think all this k?

Good advice
Really I'm just learning atm, see how it all goes

 

[Deleted member 185158]

Good advice
Really I'm just learning atm, see how it all goes

And that's why you don't over think it.. You just get the main chunk of performance from some enhancements and enjoy.

However... I'll advertise to help any one on the seek for no stability (or very little) and smoke everyone else with a similar processor lol. (I'm, far from the best, but have had my fair share of fun )

There is no sweet spot between daily use and bench-marking purposes. period. Once I give you the red pill, we go down the rabbit hole. You'll never return the same lol. Turn into a hw junkie. I've seen it many times in the last couple of decades. This hobby can become very addictive.

Oh and by the way, you can slap a fan on the memory if it feels warm to the back of your finger. (this helps stability just like a cpu )

 

[Zach_01]

Does your motherboard have a setting called SenseMi Skew and SenseMi offset by any chance?

Mine doesnt have anything like that in UEFI... Gigabyte but X570

 

[Deleted member 185158]

Mine doesnt have anything like that in UEFI... Gigabyte but X570

Oh the SenseMi offset??
Yea Im not sure if its an exclusive thing to certain boards.


Perhaps it's labeled differently? Might be ROG only? Not sure.