What's new
Carbonite

South Africa's Top Online Tech Classifieds!
Register a free account today to become a member! (No Under 18's)
Home of C.U.D.

AMD Ryzen 3000 Overclocking Thread

Kloppies

Epic Member
Rating - 100%
45   0   0
Joined
Oct 30, 2010
Messages
441
Reaction score
76
Points
3,815
Location
Cape Town
I thought I'd start a overclocking thread for Zen 2. I know there isn't much to gain from manual overclocking these CPUs, but at least memory overclocking is more of a thing now.

I got my 3600 on Tuesday, so haven't had a lot of time to play with it yet. It's paired with the popular G.Skill F4-3200C14D-16GTZR B-die set on a Asus X370-I (dual-slot DIMM).

The SAFE timings suggested by 1usmus's DRAM calculator v1.6 is not stable for me above 3200MHz, but I wouldn't expect 14-14-15-14 to be a given for 3600MHz and up on any B-die kit?

What I have tested as stable with Memtest is 3733MHz (1:1 fclk) at 1.45V, SOC 1.1V. Manual timings of 16-16-16-34-48, Trfc 298, Trdrd/Twrwr SCL both 4, and most of the rest Auto.

I only have PBO enabled, not auto-OC, and a negative core voltage offset of 0.0125V, which seems to help lower temps a little bit without losing performance.

Screenshot of Ryzen Master and AIDA64 mem benchmark:
 
I thought I'd start a overclocking thread for Zen 2. I know there isn't much to gain from manual overclocking these CPUs, but at least memory overclocking is more of a thing now.

I got my 3600 on Tuesday, so haven't had a lot of time to play with it yet. It's paired with the popular G.Skill F4-3200C14D-16GTZR B-die set on a Asus X370-I (dual-slot DIMM).

The SAFE timings suggested by 1usmus's DRAM calculator v1.6 is not stable for me above 3200MHz, but I wouldn't expect 14-14-15-14 to be a given for 3600MHz and up on any B-die kit?

What I have tested as stable with Memtest is 3733MHz (1:1 fclk) at 1.45V, SOC 1.1V. Manual timings of 16-16-16-34-48, Trfc 298, Trdrd/Twrwr SCL both 4, and most of the rest Auto.

I only have PBO enabled, not auto-OC, and a negative core voltage offset of 0.0125V, which seems to help lower temps a little bit without losing performance.

Screenshot of Ryzen Master and AIDA64 mem benchmark:

I saw GN say its best to test performance by benchmark to confirm the higher clocks mean actual higher performance. Because you could reduce vcore a lot and PBO will allow higher clocks but worse performance.
 
I thought I'd start a overclocking thread for Zen 2. I know there isn't much to gain from manual overclocking these CPUs, but at least memory overclocking is more of a thing now.

I got my 3600 on Tuesday, so haven't had a lot of time to play with it yet. It's paired with the popular G.Skill F4-3200C14D-16GTZR B-die set on a Asus X370-I (dual-slot DIMM).

The SAFE timings suggested by 1usmus's DRAM calculator v1.6 is not stable for me above 3200MHz, but I wouldn't expect 14-14-15-14 to be a given for 3600MHz and up on any B-die kit?

What I have tested as stable with Memtest is 3733MHz (1:1 fclk) at 1.45V, SOC 1.1V. Manual timings of 16-16-16-34-48, Trfc 298, Trdrd/Twrwr SCL both 4, and most of the rest Auto.

I only have PBO enabled, not auto-OC, and a negative core voltage offset of 0.0125V, which seems to help lower temps a little bit without losing performance.

Screenshot of Ryzen Master and AIDA64 mem benchmark:
Cannot seem to view you screenshot. With my Ryzen 9 3900x at first I was running 4.3ghz all core using 1.35vcore which is actually 1.33v when measuring with multimeter.

I decided to drop vcore to 1.275v which is 1.25vcore actual for 4.28ghz. My system only bench stable. Memory is 3600 cl16 with 1800mhz set for infinity fabric.

Ill do some dry ice runs this weekend to see if I can get between 4.8 and 5ghz but the IF cannot go past 1600mhz under extreme cooling so memory Ill play with 3200.

Sent from my SM-G960F using Tapatalk
 
I saw GN say its best to test performance by benchmark to confirm the higher clocks mean actual higher performance. Because you could reduce vcore a lot and PBO will allow higher clocks but worse performance.

Yeah, that is what I'm trying to test, 0.0125 is very little though. Would need to do a few runs of Cinebench or something to tell if it's really making a difference. The all core boost during R20 bench settles at 3925 MHz and scores 3367.
 
Yeah, that is what I'm trying to test, 0.0125 is very little though. Would need to do a few runs of Cinebench or something to tell if it's really making a difference. The all core boost during R20 bench settles at 3925 MHz and scores 3367.

Your 3925 all core does not sound right. The 3600 I had before the 3900x was running 4200 all core at stock, when enabling pbo it was the same.

Sent from my SM-G960F using Tapatalk
 
Your 3925 all core does not sound right. The 3600 I had before the 3900x was running 4200 all core at stock, when enabling pbo it was the same.

Sent from my SM-G960F using Tapatalk

Mmm, interesting, in any workloads? The few Cinebench runs I've done, I haven't seen higher than 3950, in lighter all-core loads it boost up to about 4100. I see my R20 score is a bit low though compared to what some reviewers got.

The only bios available for my board is Agesa 1.0.0.2, which supposedly allows higher boost clocks.
 
Cannot seem to view you screenshot. With my Ryzen 9 3900x at first I was running 4.3ghz all core using 1.35vcore which is actually 1.33v when measuring with multimeter.

I decided to drop vcore to 1.275v which is 1.25vcore actual for 4.28ghz. My system only bench stable. Memory is 3600 cl16 with 1800mhz set for infinity fabric.

Ill do some dry ice runs this weekend to see if I can get between 4.8 and 5ghz but the IF cannot go past 1600mhz under extreme cooling so memory Ill play with 3200.

Sent from my SM-G960F using Tapatalk
You always have to beat me to the XOC stuff dont you lols :) .. should be getting one as well .. was yours sponsored or did you fork out the coin as these chips are darn pricey ..
 
You always have to beat me to the XOC stuff dont you lols :) .. should be getting one as well .. was yours sponsored or did you fork out the coin as these chips are darn pricey ..
Nope I forked out the coin. Unfortunately Im not privileged or good enough to be sponsored.

Sent from my SM-G960F using Tapatalk
 
No amd partner in SA and no samples sent to SA so im in the same boat buddy :(
They pricey neh haha
Not as pricey as... well Ill leave it there.

Sent from my SM-G960F using Tapatalk
 
It is a bit annoying that putting images onto the site needs to be done from outside, but hell, let me try. So I got the Ryzen 9 3900X at first running it in my Asus Crosshair VII Hero until the board said good bey, just as I wanted to do some Dry Ice runs. Well I decided to go with a different board and at first settled on a Gigabyte Aorus Elite, big mistake, I sent the board back and used the credit plus some extra to get a MSI Meg X570 Ace. This board is really great to say the least.

Well with this board I was able to go up to 4.39Ghz all core manual overclock which needed 1.38Vcore.... yup a little too high, but hell I'm just benching so it is not for everyday use. I'll show some results, but look at the difference between the X570 and X470 board..... clocks not exactly the same as on the X470 I used per ccx clocking which I cannot do with the X570 as it goes crazy on Vcore.

Sorry I cannot even upload images so I'll explain. With the Asus C7H board I got 7844 in CB20 with CCX0 at 4.435ghz, CCX1 at 4.41Ghz, CCX2 at 4.385Ghz and CCX3 at 4.334ghz. With the X570 board, the new board the highest I managed was 7738 but that is 4.39Ghz all core.

Please explain to me which image upload works with Carb.
 
Why no love for this board?


Only 4 fan headers, seriously not what I expected. Secondly the Qpost LCD on the board for me is a must, it makes fault finding so much easier. With the Aorus Elite I could not even clock my memory higher and clocking the cpu was also pretty difficult. With the Msi Meg X570 Ace I am able to clock it higher than I did with the C7H and with less vcore.
 
Only 4 fan headers, seriously not what I expected. Secondly the Qpost LCD on the board for me is a must, it makes fault finding so much easier. With the Aorus Elite I could not even clock my memory higher and clocking the cpu was also pretty difficult. With the Msi Meg X570 Ace I am able to clock it higher than I did with the C7H and with less vcore.
Not disagreeing with you, but you are comparing a R7000 MSI board with a R4000 Gigabyte board.
Shouldn't you have tried the Master in place of the Elite? Wouldn't they perform about the same
 
Not disagreeing with you, but you are comparing a R7000 MSI board with a R4000 Gigabyte board.
Shouldn't you have tried the Master in place of the Elite? Wouldn't they perform about the same
The primary reason I took the Msi Meg x570 Ace was 7 fan headers and the qpost lcd. I just prefer to use the qpost lcd as it shows if I went a little overboard with the memory settings. Cpu clocks not an issue and I really enjoy the MSI bios.

Sent from my SM-G960F using Tapatalk
 
I've got my 3700x to run rock solid stable at an all core OC of 4.25Ghz @ 1.30 Vcore . This is with LLC level 5, SOC on auto and Ram at 3733MHz 15-17-17-36 @ 1.5V (Samsung B-Die). Infinity Fabric (FCLK) is at 1866.5Mhz. This is on my MSI B450 Gaming Pro Carbon.

With the current gimped bios (1.0.0.3ab on my MB) I find that with both Stock and PBO the single core boosts are few and far between and not reaching the 4.4Ghz anyway. One or two cores will hit 4.375Ghz under very light load but not often and the rest will jump to around 4.25Ghz to 4.3Ghz. PBO gives better all core clocks under load than the stock setting but pushes more voltage than necessary (Around 1.38V to 1.44V) creating excess heat and current draw which then seems to limit the max loaded all core frequency to around 4.15Ghz to 4.2Ghz (Less than my all core manual OC). Because of these auto features currently being less than ideal I found the manual overclock gives about a 10 percent performance increase over stock in both benchmarks and gaming. Compared to PBO the increase is about 5 to 7 percent faster in gaming and 3 or 4 percent in benchmarks.

This increase in performance also comes with a much safer loaded voltage and cooler and more stable temperatures. I have a Corsair H150I 360mm AIO and my max temps during stress tests are 66Deg and idle is around 28 to 30 degrees. Intensive gaming sees around 45 Degrees. With PBO enabled I was seeing as high as 78 Degrees with higher and more unstable idle and light load temps. I tried a voltage offset with PBO to help with this and at -0.05V offset I did see an improvement in temps without a performance decrease but it still couldn't match the Manual OC performance and temperatures.

I experimented with higher clock speeds but needed more voltage for minimal gains. Could get one CCX to 4.3Ghz and the other to 4.225Ghz at 1.35V but saw negligible performance increases in games and benchmarks so not worth the extra voltage. I am sure that if I went for 1.38V to 1.4V I could get a stable 4.3Ghz all core OC but not worth risking degradation for a marginal performance increase.

So, for now, it seems I have found the sweet spot with my chip, Once the BIOS and chipset drivers mature I will try auto and PBO again. I mostly do gaming so consistent boost behaviour across a few cores to 4.4Ghz would probably be beneficial, but for now the auto boosting and voltage control features, at least with my motherboard, seem to be holding things back.

Overall I am impressed and happy with the chip. Much better performance than my previous 2600x. My Cinebench R15 score is 2232 as things stand. I think I can still neaten things up on the RAM side but doubt I will see much more here as it's already running a pretty solid speed.
 
I've got my 3700x to run rock solid stable at an all core OC of 4.25Ghz @ 1.30 Vcore . This is with LLC level 5, SOC on auto and Ram at 3733MHz 15-17-17-36 @ 1.5V (Samsung B-Die). Infinity Fabric (FCLK) is at 1866.5Mhz. This is on my MSI B450 Gaming Pro Carbon.

With the current gimped bios (1.0.0.3ab on my MB) I find that with both Stock and PBO the single core boosts are few and far between and not reaching the 4.4Ghz anyway. One or two cores will hit 4.375Ghz under very light load but not often and the rest will jump to around 4.25Ghz to 4.3Ghz. PBO gives better all core clocks under load than the stock setting but pushes more voltage than necessary (Around 1.38V to 1.44V) creating excess heat and current draw which then seems to limit the max loaded all core frequency to around 4.15Ghz to 4.2Ghz (Less than my all core manual OC). Because of these auto features currently being less than ideal I found the manual overclock gives about a 10 percent performance increase over stock in both benchmarks and gaming. Compared to PBO the increase is about 5 to 7 percent faster in gaming and 3 or 4 percent in benchmarks.

This increase in performance also comes with a much safer loaded voltage and cooler and more stable temperatures. I have a Corsair H150I 360mm AIO and my max temps during stress tests are 66Deg and idle is around 28 to 30 degrees. Intensive gaming sees around 45 Degrees. With PBO enabled I was seeing as high as 78 Degrees with higher and more unstable idle and light load temps. I tried a voltage offset with PBO to help with this and at -0.05V offset I did see an improvement in temps without a performance decrease but it still couldn't match the Manual OC performance and temperatures.

I experimented with higher clock speeds but needed more voltage for minimal gains. Could get one CCX to 4.3Ghz and the other to 4.225Ghz at 1.35V but saw negligible performance increases in games and benchmarks so not worth the extra voltage. I am sure that if I went for 1.38V to 1.4V I could get a stable 4.3Ghz all core OC but not worth risking degradation for a marginal performance increase.

So, for now, it seems I have found the sweet spot with my chip, Once the BIOS and chipset drivers mature I will try auto and PBO again. I mostly do gaming so consistent boost behaviour across a few cores to 4.4Ghz would probably be beneficial, but for now the auto boosting and voltage control features, at least with my motherboard, seem to be holding things back.

Overall I am impressed and happy with the chip. Much better performance than my previous 2600x. My Cinebench R15 score is 2232 as things stand. I think I can still neaten things up on the RAM side but doubt I will see much more here as it's already running a pretty solid speed.
Sounds really good... I had a hickup with my Msi Meg x570 Ace saying goodbye. Picked up another yesterday and as things stand I'm IBT very high mem preset stable at 4.325ghz using 1.325vcore on my Ryzen 9 3900x. Temps during IBT is 81c but running cinebench or so it is about mid 70c. The cpu can do 4.2ghz at 1.1875vcore IBT stable.

Im pretty happy with the setup. We will not see the normal advertised boost clock anymore. AMD lowered it with the new Agesa for longevity.

Sent from my SM-G960F using Tapatalk
 
B eta bioses ABBA with boost fix already available for MSI boards.


Thank you my brotheren, I will try it out with my ACE mobo. I did however find after sitting with Bios Agesa 1.0.0.3AB for the past couple of weeks that their latest official bios V1.4 1.0.0.3 ABB actually resulted in better performance but my max overclock dropped a bit, but performance even better than my previous max OC.
 

Users who are viewing this thread

Back
Top Bottom