Welcome Guest ( Log In | Register )

30 Pages « < 6 7 8 9 10 > » Bottom

Outline · [ Standard ] · Linear+

 AMD Ryzen /Threadripper OverClocking V1.1, big or small, share your results here...

views
     
owikh84
post Sep 9 2019, 11:51 PM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



Holiday benching with 3600X + C8H that I recently purchased from shopee at a good deal biggrin.gif

user posted image

Starting off with CPU OC, so far I've tested 3 types of coolers at 4.2GHz OC @ 1.375v on all cores.
Thermal results based on Prime95 v29.8 b6 custom blend 1344K min/max for 30mins as follows:
Cooler Min (°C) Max (°C)
AMD Wraith Spire 50.0 92.3
Raijintek Orcus 240mm 47.1 79.8
Custom WC 40.5 74.4

Max OC under 1.40v should be 4.25GHz, thermal results under custom WC as below:
CPU Clock Min (°C) Max (°C)
Stock @ 4.1-4.125GHz @ 1.28v 39.6 66.6
4.2GHz @ 1.375v* 40.5 74.4
4.25GHz @ 1.375v 40.3 76.3
* Can still be reduced a bit

Some benchmarks at 4.25GHz OC:

Cinebench R15:
user posted image

Cinebench R20:
user posted image

Hexacore CPU comparisons:
CPU CB R15 MT CB R15 ST CB R20 MT CB R20 ST
3600X @ 4.25GHz 1695 199 3866 490
8700K @ 5.00GHz 1650 217 3962 517

4.3GHz is bootable but sometimes CB R20 crashed at 1.425v so I think more juices is required to get it stable.
Will revisit this later and update in this thread...
nrw
post Sep 10 2019, 12:02 AM

Casual
***
Junior Member
398 posts

Joined: Jul 2014


QUOTE(owikh84 @ Sep 9 2019, 10:03 PM)
B-die is 2666 CL19, pretty much the same true latency but less bandwidth.

2666 CL19: 2000/2666*19 = 14.25ns
2933 CL21: 2000/2933*21 = 14.32ns

Ya, curious about A-die's OCbility vs B-die.
We gonna wait and see later...  biggrin.gif
*
can't judge from absolute latency to oc capabilities tongue.gif

the m-dies for that series was kind of underwhelming (however that one was 1Y(15nm) and not 1Z(12nm)

~RM650-700 fo 32GB stick btw.

This post has been edited by nrw: Sep 10 2019, 12:04 AM
owikh84
post Sep 10 2019, 12:43 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



Continuing with RAM OC, so far hasn't been so impressive mainly due to high latency issue.
It could be that the latest DRAM Calculator v1.6.2 by 1usmus has decided to set the subtimings too loose that eventually hitting the RAM performance. Maybe need further tweaking to optimize the bandwidths and latency.

RAM OC with SK-Hynix AFR:

3466 16-19-19-40-1T @ 1.41v:
user posted image

3533 16-19-19-40-1T @ 1.45v (not stable):
user posted image

RAM OC with Samsung B-die:

user posted image

3600 14-15-15-30-1T @ 1.45v:
user posted image

3800 16-17-16-32-1T @ 1.41v:
user posted image

4000 16-17-16-32-1T @ 1.45v:
user posted image

4133 16-17-16-32-1T @ 1.45v:
user posted image

Again, still need to figure out how to reduce improve the efficiency.
Aim for 3800 CL14 stable but I don't think possible with the current ADATA D80 sticks.
Later I swap to my TridentZ RGB 4266 kit and see how much I can push. notworthy.gif

TSah_khoo
post Sep 10 2019, 09:06 AM

- No Action Talk Only -
Group Icon
Elite
8,103 posts

Joined: Jan 2003
From: Tanah Lot @ Dungun, Terengganu.


@owikh84 many thanks for the efforts bro... tests like these are time & efforts consuming. i hope AMD MY see d efforts here & should there be new model IMHO they should at least send u one unit for a spin. it's not always about marketing bells & whistles (not that it's not important) but efforts like these shouldn't gone unnoticed. thumbsup.gif

very nice timing on the BDie for 3800 & above. glad to see most of d user can do 3800 @ 1:1 while mine max out @ 3733 only. sad.gif

recently acquired a pair of ven pro rgb 3600 (cl18) with BDie but i think it's not gonna hold CL16 @ 4k & above. still in d midst of testing but hopefully i can post something within d week. sweat.gif

btw nice to see memtweak it works on Ryzen 3K, hopefully it's compatible with non ROG/asus boards as well. tongue.gif
nrw
post Sep 10 2019, 10:44 AM

Casual
***
Junior Member
398 posts

Joined: Jul 2014


QUOTE(owikh84 @ Sep 10 2019, 12:43 AM)
Continuing with RAM OC, so far hasn't been so impressive mainly due to high latency issue.
It could be that the latest DRAM Calculator v1.6.2 by 1usmus has decided to set the subtimings too loose that eventually hitting the RAM performance. Maybe need further tweaking to optimize the bandwidths and latency.

RAM OC with SK-Hynix AFR:

3466 16-19-19-40-1T @ 1.41v:
user posted image

3533 16-19-19-40-1T @ 1.45v (not stable):
user posted image

RAM OC with Samsung B-die:

user posted image

3600 14-15-15-30-1T @ 1.45v:
user posted image

3800 16-17-16-32-1T @ 1.41v:
user posted image

4000 16-17-16-32-1T @ 1.45v:
user posted image

4133 16-17-16-32-1T @ 1.45v:
user posted image

Again, still need to figure out how to reduce improve the efficiency.
Aim for 3800 CL14 stable but I don't think possible with the current ADATA D80 sticks.
Later I swap to my TridentZ RGB 4266 kit and see how much I can push.  notworthy.gif
*
Gj, looking at your latency I am fairly sure your power down mode is enabled.
owikh84
post Sep 10 2019, 11:48 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



QUOTE(nrw @ Sep 10 2019, 10:44 AM)
Gj, looking at your latency I am fairly sure your power down mode is enabled.
*
I need to recheck this later as the GDM and PDM are mixes of enabled and disabled as per recommendation by DRAM Calculator.

Edit: Okay I just rechecked and found that my Power Down Mode has been disabled right from 3466 all the way up to 4133. So, the source of high latency could be from other settings.

This post has been edited by owikh84: Sep 10 2019, 09:59 PM
nrw
post Sep 10 2019, 12:49 PM

Casual
***
Junior Member
398 posts

Joined: Jul 2014


QUOTE(owikh84 @ Sep 10 2019, 11:48 AM)
I need to recheck this later as the GDM and PDM are mixes of enabled and disabled as per recommendation by DRAM Calculator.
*
Since few versions those recommendations are quite off, power down disabled is hardly a source for instabiliy, whereas the disabled gear down mode from the last version is not really advisable in my opinion for long term stability.

I made the observation with my c7h that despite disabling power down mode it remains active with certain configurations.
owikh84
post Sep 14 2019, 12:13 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



3600X 4.3GHz @ 1.45v, I think I'm pretty much hitting the ceiling already. Yes, note the voltage is exponentially increased. Thermal results based around P95 28.9 b6 custom blend 1344K min/max FFT for 30mins:

CPU Clock Min (°C) Max (°C)
Stock @ 4.1-4.125GHz @ 1.28v 39.6 66.6
4.2GHz @ 1.375v* 40.5 74.4
4.25GHz @ 1.375v 40.3 76.3
4.30GHz @ 1.450v 40.0 82.6

* Can still be reduced a bit

user posted image

Cinebench benchmark scores also improved a bit:
CPU CB R15 MT CB R15 ST CB R20 MT CB R20 ST
3600X @ 4.30GHz 1715 201 3893 493
3600X @ 4.25GHz 1695 199 3866 490
8700K @ 5.00GHz 1650 217 3962 517

CB R15:
user posted image

CB R20:
user posted image
owikh84
post Sep 14 2019, 12:57 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



Latest AGESA 1.0.0.3 ABBA seems to have fixed my 3600X's boost bug that I had earlier, now I can see one core is boosting up to 4.4GHz as per AMD's original specs. CB R20 runs with everything default:

BIOS 0803:
user posted image

BIOS 1001:
user posted image
nrw
post Sep 14 2019, 08:09 PM

Casual
***
Junior Member
398 posts

Joined: Jul 2014


QUOTE(owikh84 @ Sep 14 2019, 12:57 AM)
Latest AGESA 1.0.0.3 ABBA seems to have fixed my 3600X's boost bug that I had earlier, now I can see one core is boosting up to 4.4GHz as per AMD's original specs. CB R20 runs with everything default:

BIOS 0803:
user posted image

BIOS 1001:
user posted image
*
good work! thumbup.gif

My 3700x made similar improvements and is within specification now.
The 3600 I got still doesn't want to reach 4.2 single core.
It comfortably does 4.25 all core with 1.3675V though hmm.gif

Guess there will be more improvements to come as 3900x' didn't seem as positively effected.

The core rating in Ryzen master however is still off.

What you haven't pointed out from your results is that (unfortunately in hwinfo64 the polling period is not visible) voltages under load and hence temperatures dropped for stock settings.
owikh84
post Sep 14 2019, 10:19 PM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



QUOTE(nrw @ Sep 14 2019, 08:09 PM)
What you haven't pointed out from your results is that (unfortunately in hwinfo64 the polling period is not visible) voltages under load and hence temperatures dropped for stock settings.
*

Voltage is set by ASUS BIOS so nothing to do with AGESA, temperature is stock cooler vs custom loop LOL.

nrw
post Sep 14 2019, 10:58 PM

Casual
***
Junior Member
398 posts

Joined: Jul 2014


QUOTE(owikh84 @ Sep 14 2019, 10:19 PM)
Voltage is set by ASUS BIOS so nothing to do with AGESA, temperature is stock cooler vs custom loop LOL.
*
then the testing method is quite off blink.gif as thermal throttling starts in low to mid 70s already.
and ofc the smu in your new agesa controls voltage blush.gif
why you think its called system management unit?
with its sensory sampling its controlling circuits on the chip. that includes voltage level, voltage supply to psms, thermal management, c-s boost.

owikh84
post Sep 15 2019, 08:29 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



QUOTE(nrw @ Sep 14 2019, 10:58 PM)
then the testing method is quite off  blink.gif  as thermal throttling starts in low to mid 70s already.
and ofc the smu in your new agesa controls voltage  blush.gif
why you think its called system management unit?
with its sensory sampling its controlling circuits on the chip. that includes voltage level, voltage supply to psms, thermal management, c-s boost.
*

Good point, but I was referring to the load voltage behavior being controlled by BIOS LLC settings. smile.gif

BTW, I'm still having high memory latency (70ns+) and low bandwidth (50k+ MB/s) seen in AIDA64 benchmark.
Tried everything such as reducing tRFC down to 160ns, PDM/GHM on/off, tRC 42-44, tFAW 16-20, new ABBA BIOS etc none of them seems to solve the issue. sad.gif
khiunngiap
post Sep 15 2019, 07:39 PM

Getting Started
**
Junior Member
196 posts

Joined: Sep 2010
Hi guys , anyone know how to lock 3700x cores speed across all ?

I managed to do it for my 1700 but not this 3700x.

Ps: I know many has been saying no real different with pbo on . But I want to justify my self and I had searched info from the net but no luck.
khiunngiap
post Sep 15 2019, 07:44 PM

Getting Started
**
Junior Member
196 posts

Joined: Sep 2010
Hi guys , anyone know how to lock 3700x cores speed across all ?

I managed to do it for my 1700 but not this 3700x.

Ps: I know many has been saying no real different with pbo on . But I want to justify my self and I had searched info from the net but no luck.
TSah_khoo
post Sep 16 2019, 11:49 AM

- No Action Talk Only -
Group Icon
Elite
8,103 posts

Joined: Jan 2003
From: Tanah Lot @ Dungun, Terengganu.


QUOTE(khiunngiap @ Sep 15 2019, 07:39 PM)
Hi guys , anyone know how to lock 3700x cores speed across all ?

I managed to do it for my 1700 but not this 3700x.

Ps: I know many has been saying no real different with pbo on . But I want to justify my self and I had searched info from the net but no luck.
*
for a start i will disable pbo/core performance boost (might diff from manufactures to manufacturers) & set manual cpu multi. given how lemon some Ryzen 3K are i'd suggest u start from stcok clock & slowly up (100MHz a notch mebe) & see how far it goes. wouldn't run above 1.4v for 24/7 so i'll take d highest stable clock w/ 1.4v & my cooling can afford (max 75'c with avx programs). my 3600 can't even p95 stable @ 3.9GHz with 1.45v. so i give up on cpu oc, just play with ram oc a lil here n there. dont feel discouraged thou, i saw quite a few did 4.2/4.25GHz for 24/7 with very decent volt so just try it out. nod.gif

This post has been edited by ah_khoo: Sep 16 2019, 11:51 AM
khiunngiap
post Sep 16 2019, 02:19 PM

Getting Started
**
Junior Member
196 posts

Joined: Sep 2010
I can hit 4.3.however, it doesn't run constantly 4.3 for all cores. It does only depends on usages.
TSah_khoo
post Sep 20 2019, 09:50 PM

- No Action Talk Only -
Group Icon
Elite
8,103 posts

Joined: Jan 2003
From: Tanah Lot @ Dungun, Terengganu.


slowly burnin d ram... laugh.gif

Attached Image
owikh84
post Sep 21 2019, 09:44 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



QUOTE(ah_khoo @ Sep 20 2019, 09:50 PM)
slowly burnin d ram...  laugh.gif
*

Well done on the 2000% HCl. thumbup.gif
How's your AIDA64 benchmark score look like?
Good to see MemTweakIt is working on a none-ASUS board as well.

BTW, did you notice that HWMonitor is giving you vCore up to 2v max?
I'd suggest try to avoid using this software, or it could be that the GIGA boards are still using iTE Super I/O chip, which reads voltages not as accurate as the Nuvoton chip.


This post has been edited by owikh84: Sep 22 2019, 09:22 AM
owikh84
post Sep 22 2019, 11:43 AM

i7 Clan
Group Icon
Elite
8,711 posts

Joined: Nov 2007
From: Butterworth, PG / Machang, Kelantan



3800 14-16-14-28-1T @ 1.480v

user posted image

30 Pages « < 6 7 8 9 10 > » Top
 

Change to:
| Lo-Fi Version
0.0175sec    0.26    6 queries    GZIP Disabled
Time is now: 27th November 2025 - 09:37 PM