Overclock.net banner

121 - 140 of 185 Posts

·
Registered
Joined
·
236 Posts
At this stage im tempted to try a boot on 4.5ghz and see where it goes i just wanna see this cpu hit that mark more often since i got the xt version
Below are unload gaming numbers

gaming push.PNG
gamingvoltage.PNG
 

·
Registered
Joined
·
236 Posts
FINAL NOTES
Amd cool and quite caused performance loss? And thermal paste

I bought 3900xt a couple of weeks ago and after poor pbo, llc and undervolting performance in a number of ways. I almost gave up never really saw my cpu pushing past 4.2ghz while gaming and rarely hitting 4.6ghz. I saw a few threads mentioning that AMDs cool and quite function killed their cpu performance so I turned it off. And i instantly noticed in hw monitor that my first 5 cores were hitting 4.7ghz mark and the others 4.3ghz.
So my question is are there any other power saving settings i should disable in my gigabyte bios? And just incase ive always used the windows high performance setting for amd if anyone is wondering.
Second question is about thermal paste appilcation as you will see my intial dot method was a big fail. I then watched a number of guides and applied it differently and this time underload i hit 71c using normal cpu power setting not auto in bios compared 89c on prime 95 testing with a undervolt. System idle went from 40c to 35c. Any other tips on thermal paste?

Im done pbo disabled, llc left on auto and ryzen master not used only thing ill do is buy proper cl14/16 ram in the future with an xmp of 3600mhz.

Bios settings image: Old thermal paste: Old thermal paste water cooler: New thermal paste : New thermal paste finish: Sytem specs
Ryzen 3900xt
Gigabyte x570 gaming x
32gb of g.skill [email protected] 16-19-19-36
MSI Vega 64
 

·
Registered
Joined
·
22 Posts
FINAL NOTES
Amd cool and quite caused performance loss? And thermal paste

I bought 3900xt a couple of weeks ago and after poor pbo, llc and undervolting performance in a number of ways. I almost gave up never really saw my cpu pushing past 4.2ghz while gaming and rarely hitting 4.6ghz. I saw a few threads mentioning that AMDs cool and quite function killed their cpu performance so I turned it off. And i instantly noticed in hw monitor that my first 5 cores were hitting 4.7ghz mark and the others 4.3ghz.
So my question is are there any other power saving settings i should disable in my gigabyte bios? And just incase ive always used the windows high performance setting for amd if anyone is wondering.
Second question is about thermal paste appilcation as you will see my intial dot method was a big fail. I then watched a number of guides and applied it differently and this time underload i hit 71c using normal cpu power setting not auto in bios compared 89c on prime 95 testing with a undervolt. System idle went from 40c to 35c. Any other tips on thermal paste?

Im done pbo disabled, llc left on auto and ryzen master not used only thing ill do is buy proper cl14/16 ram in the future with an xmp of 3600mhz.

Bios settings image: Old thermal paste: Old thermal paste water cooler: New thermal paste : New thermal paste finish: Sytem specs
Ryzen 3900xt
Gigabyte x570 gaming x
32gb of g.skill [email protected] 16-19-19-36
MSI Vega 64
Same boat as you, I gave up on PBO its piss poor on my 3900xt. If anything, left to its own devices it performs on par (if even that) as a stock 3900x.
I just settled on an old fashioned OC. I experimented with LLC and ironically the system is MORE stable with LLC on Auto than set to anything else. Weird.

I can get CCD1 pretty stable at 44.5 and CCD2 at 43.25 running 1.244v which keeps my temps reasonable. By reasonable, I mean better heat/performance than anything else I tried.

PBO my temps never go past 75C which (thanks AMD/ASUS for looking out for my CPU) is pretty weaksauce... like if I'm pounding this thing for all its worth, I don't need water wings let if FLY!

Re: thermal compound, this is a pretty big ****-off IHS so you need to squirt a fair bit. I spent a ton of time repasting laptop dies with god-awful heatsinks that seat less well than grandpa's hemorrhoid backside, this is a different animal. A decent cooler's contact surface is often slightly convex and will flex inward as it pushes on to the IHS. The question is how much torque is needed to get optimal coverage without becoming TOO tight or unbalanced. If you are looking to possibly repaste again, first give each screw a quarter-to-half turn and see if that improves temps a bit under load.

I suggest you do NOT spread the paste about like you are trying to smooth out joint compound on drywall, you will end up with a less than perfect seal.
The X method ensures your corners are covered, no airgaps
The pea method guarantees the centre but corners maybe less. (Use a bigger pea than you did in the pics above)
The line method is only effective if you know where the bits lie under the IHS, you want your line to cross your CCDs, not between them. Still can end up with dry corners.

Too much compound is better than not enough.
 

·
Registered
Joined
·
236 Posts
Same boat as you, I gave up on PBO its piss poor on my 3900xt. If anything, left to its own devices it performs on par (if even that) as a stock 3900x.
I just settled on an old fashioned OC. I experimented with LLC and ironically the system is MORE stable with LLC on Auto than set to anything else. Weird.

I can get CCD1 pretty stable at 44.5 and CCD2 at 43.25 running 1.244v which keeps my temps reasonable. By reasonable, I mean better heat/performance than anything else I tried.

PBO my temps never go past 75C which (thanks AMD/ASUS for looking out for my CPU) is pretty weaksauce... like if I'm pounding this thing for all its worth, I don't need water wings let if FLY!

Re: thermal compound, this is a pretty big ****-off IHS so you need to squirt a fair bit. I spent a ton of time repasting laptop dies with god-awful heatsinks that seat less well than grandpa's hemorrhoid backside, this is a different animal. A decent cooler's contact surface is often slightly convex and will flex inward as it pushes on to the IHS. The question is how much torque is needed to get optimal coverage without becoming TOO tight or unbalanced. If you are looking to possibly repaste again, first give each screw a quarter-to-half turn and see if that improves temps a bit under load.

I suggest you do NOT spread the paste about like you are trying to smooth out joint compound on drywall, you will end up with a less than perfect seal.
The X method ensures your corners are covered, no airgaps
The pea method guarantees the centre but corners maybe less. (Use a bigger pea than you did in the pics above)
The line method is only effective if you know where the bits lie under the IHS, you want your line to cross your CCDs, not between them. Still can end up with dry corners.

Too much compound is better than not enough.

For fun i decided to try this ccx/ccd bios overclocking so i went 1.28v ccd0 @4.55ghz and ccd 1 @4.3ghz manage to boot windows lasted 5mins on its own and cinebench caused a crash. If it wasnt for the windows boot error messages after overclocking, I would try more but it turns me off. WIndows tries to repair nothing comes of it then the next boot is fine.

I imagine if i set ccd0 @4.55ghz and the ccd1 on auto that would be fine if it booted and worked? As I understand thats how most overclocks just a few cores overclocked? Its been a long and learning journey with amd's overclocking style. One day ill try more things
 

·
Registered
Joined
·
236 Posts
I know i said i was done but after finally finding finding the CCD settings on my motherboard I cant help but try again. Every guide was either a different brand of motherboard or a core overclock. I kept wondering why i couldnt find these settings and I finaly did.

The first image shows my bios and a old settings, I tried [email protected]/[email protected] at 1.28v/1.35v both booted to windows when using cinebench the system crashed. Second attempt was [email protected]/[email protected] 1.28v windows booted gotta further in cinebench before a crash.

First question can i overclock ccd0 for instance and leave ccd1 on auto?

Second question If i want my cpu to last at least 5yrs on overclock whats the max voltage i could use? I previously ran a overclocked [email protected] for over 4years so I want the same lifespan.

Third question after a failed overclock windows will try system repair, I either decline it or do a another restart and then everything is fine is this normal? Might it be that the moment i crash regardless of a bluescreen post or not i instantly press the reset button? This turns me off trying to overclock.

 

·
Registered
Joined
·
590 Posts
@Ark-07 after all the thinkering with your 3900xt, do you mind posting your cb15/20 single and multicore scores?
 
  • Rep+
Reactions: Ark-07

·
Registered
Joined
·
236 Posts
@Ark-07 after all the thinkering with your 3900xt, do you mind posting your cb15/20 single and multicore scores?
Will do im gonna try the pbo glitch first current base scores are 7080 multi c20 and 501 single core. Undervolt 7180 multi c20 495 single.
 

·
Registered
Joined
·
590 Posts
@Ark-07 that's bellow the average for a full stock cpu, when well cooled. I do here 522/7250-7325 but it is very temperature sensitive, if is around 27C outside I will do less multicore, around 7150. I've been reading your posts and never said anything cause it's clear you are trying stuff, as we all did, the pbo bug as well. And most of us ended up with the same conclusion, stock cpu is always better and will give the right amount of performance single and multicore. Maybe there will be some more performance to squeeze out with 1usumus tool, but fixed clock, ccx overclock, pbo bug, all of them, you will lose single core performance. But you can and should try them all, for science. Hopefully you will end up leaving at stock.
Oh, one thing to add. I was also annoyed cause my 3900x would never reach 4.6ghz. That's VERY bios, temperature, and whateverisrunninginthebackground dependant. I can tell the difference in boosting (numerically speaking only) in between bios 6.20A and 6.40 of the taichi x370.
6.20A used to boost way more, HWiNFO64 would catch even 4625mhz in some cores when under light load. But 4.60 does not. Maximum is 4550Mhz. But the scores are the same. I can also reproduce same good boosting of 6.20A in 6.40, I just need to use 1usmus profile, enable cppc etc as per instructions and it will boost. But the scores are the same, cb15, cb20, superpi, cpuZ, etc. So sometimes it's just a virtual thing, that stretch clock to 4.6-4.625 does not translate to performance. It's visual only.
Try closing all background apps and run BoostTester along hwinfo64 and check your boost. Of course, cpu full auto.
Another thing, gaming. Max clock is 4325 sometimes. I see people saying they game around 4200-4500 most of the time, whereas I play around 4100-4325. Did I find a fix? No. But checking a lot of cpu scores in the 3dmark page I get same thing... Just my two cents about it.
"leave auto, enjoy more".
 

·
Registered
Joined
·
236 Posts
@thomasck Your right i got an extra 200 point in cinebench for higher voltages and temps. I bet single core performance was probably worse i didnt test it. But it was nice seeing 4.5ghz while gaming so I tried a voltage offset off -0.0720 on next. Cinebench took a huge hit 6746 as a result. This pbo glitch is nice I wanna leave it on with a more stable voltage offset, but I know myself too well the idea im using a glitch to get more performance is gonna really bother me. My only thing to do now is to decide whether auto cpu core voltage is better on auto or normal. I definately notice far smoother performance with it on auto for gaming but voltages tend to be a tad higher. Ive yet to try normal voltage without a offset.

This all started because i noticed my cinebench scores are lower then most it and i was hoping to see a boost of 4.5ghz while gaming not 4.2-4.4. I thought maybe it was because of my ram been hynix cjr that i got lower scores in cinebench? But i noticed all those 7800+ scores are overclocked also and have bdie ram with better timings. I was then led down a long and confusing spiral of information it seemed when the 3900x first came out people freaked out about the voltages going above 1.35v. Then later information showed that amd said anything up to 1.47v is safe while idle and it drops much lower during heavy workloads like prime95. Spent a week stressing about voltages because the oldest information was misinformed as I understand it.

Ill wait another year, maybe by then there will be some really awesome watercooler for ryzen cpu's and I'll see how things ended up for people using the pbo glitch and fixed overclocks. One video scared me where the person said because of a high voltage overclock their 3900x degraded and barely pushes 3.9ghz anymore. One more question of the settings in this image i disabled amd cool and quite function and now global c state control to disabled. I want none of my cores to park as they, i prefer max performance so what should i choose? Ive disabled pbo again. Thank you for your input and reassurance.
 

·
Registered
Joined
·
22 Posts
@thomasck Your right i got an extra 200 point in cinebench for higher voltages and temps. I bet single core performance was probably worse i didnt test it. But it was nice seeing 4.5ghz while gaming so I tried a voltage offset off -0.0720 on next. Cinebench took a huge hit 6746 as a result. This pbo glitch is nice I wanna leave it on with a more stable voltage offset, but I know myself too well the idea im using a glitch to get more performance is gonna really bother me. My only thing to do now is to decide whether auto cpu core voltage is better on auto or normal. I definately notice far smoother performance with it on auto for gaming but voltages tend to be a tad higher. Ive yet to try normal voltage without a offset.

This all started because i noticed my cinebench scores are lower then most it and i was hoping to see a boost of 4.5ghz while gaming not 4.2-4.4. I thought maybe it was because of my ram been hynix cjr that i got lower scores in cinebench? But i noticed all those 7800+ scores are overclocked also and have bdie ram with better timings. I was then led down a long and confusing spiral of information it seemed when the 3900x first came out people freaked out about the voltages going above 1.35v. Then later information showed that amd said anything up to 1.47v is safe while idle and it drops much lower during heavy workloads like prime95. Spent a week stressing about voltages because the oldest information was misinformed as I understand it.

Ill wait another year, maybe by then there will be some really awesome watercooler for ryzen cpu's and I'll see how things ended up for people using the pbo glitch and fixed overclocks. One video scared me where the person said because of a high voltage overclock their 3900x degraded and barely pushes 3.9ghz anymore. One more question of the settings in this image i disabled amd cool and quite function and now global c state control to disabled. I want none of my cores to park as they, i prefer max performance so what should i choose? Ive disabled pbo again. Thank you for your input and reassurance.
2458870



2458871


7744/508 is the score from my current setting, the 7908/520 was I think 45-45-43.5-43.5 at about 1.25v which did crash after 3-4 blender runs

The first image being about 50mhz lower can run for weeks. Some absolute torture tests can cause the CPU to get too warm but really, things like AIDA64 and Blender can run indefinitely and temps top out in the 80s

EDC bug did absolutely nothing, I think it's ineffective for 3900xt.
PBO may or may not have done anything but in my experience running this chip at stock settings/PBO/anything except an old fashioned OC I found results underwhelming at best. The chip DOES perform pretty well as can be seen in the images above, just needs a little coaxing.
 

·
Registered
Joined
·
590 Posts
@Ark-07 In my tests I failed in not losing single core performance, every scenario lead to that. What do you mean auto vs normal voltage "auto cpu core voltage is better on auto or normal"? We can say normal voltage is auto, cause then is controlled by the cpu. The normal behaviour of a 3900x cpu when on Auto is,
  • all core intense load = high heat, lower vcore, around 4050mhz when at 75C+. As voltage does up current goes down. The opposite as well.
  • idle = always high vcore, however the current is way lower, and here is where people go crazy thinking the cpu is defective. And there's a lot more around it, background programs, and what power plan is being used at the time. High performance most cores will park around 3750 when doing nothing, and will kick to 4600, 4550, 4500 in some cores when I start using the pc with vcore always around 1.488V, but again, very low current. Balanced I see cores bouncing less and a little lower vcore around 1.450V. And power saver all cores go to 2.2GHz and vcore is 0.9v, which is I use all the time when not gaming. Is it possible to get the cores to park at a lower speed when using balanced or high performance? In some cases changing "power supply idle control" helps with that. In my case, a taichi x370 it did not. I also don't bother about. With or without it the rig pulls the same from the wall, around 65W when just browsing etc.
  • light loads = high boost, high vcore as this boost is singlecore so that's how the zen2 were designed to work, nothing to worry about.
Auto with a negative offset is good, but you gotta bench the cpu before appling the offset cause if the negative offset is too great you might lose performance. Until 0.0625 I don't lose performance, I lose the boost as I don't see 4.6ghz and I see less 4.55+-. However the scores are the same for any singlecore bench and multicore are better due to better thermals. But it depends on the motherboard as well, and what LLC is being used.

You ram has nothing to do with cinebench, any of them. RAM does not have any sort of major positive or negative impact over any cinebench. Run it with ram at 2133/3200/3800mhz and you will see most none difference, maybe some extra points due to higher fclk, that's it. But nothing wow.

If your cinebench is lower, or you have too high temperature, or not enough vcore with if is on auto is very unlikely. Most likely temperature. As I said before, the to be achieved clocks while under heavy load are very temperature sensitive. In cool days I do 7350, in chill days 7250 and summer around 7150.. Right now is about 20C outside and I'm doing 7274 in high performance and 6887 in power saver - so all cores clocked to 2.2ghz.

About water cooler well, I don't know what else can be done. Maybe another block built having in mind the positioning of the chiplets, I don't know. But feels strange seeing my 3900x breach 75C mark when is under a loop with 3x360mm rads. Well, it can be better, was I have a lian li o11d and airflow is not the best thing. If a remove just one panel I get around minus 2 or 3C straight away. 3900X are hot, 3900XT seems to be less hot, but stills a hot chip.

To do not send cores do park/sleep you got to use always high performance plan, so they will be around 3700 -> boost when light load. Forget monitoring it from RM, use HWiNFO64. And you should mess with cool n quiet and cstates. When off the cpu should send less cores to sleep/park. That is board dependant so I can not talk much about it. In the asrock bios there's no more cool n quiet, just cstates and dfstates which should allow or not some cores do sleep/park when not in use but I see no difference when it's on or off. Also power supply idle control might play a role in this too, you gotta test. Even "oc/uncore soc on or off" might be related as at least in my bios says "force components fclk etc to work always in their maximum frequency", but again I see no difference. LLC might influence too.

Have you tried closing all background apps and run BoostTester?

@alasdairvfr 7908/520 it's a really great multicore score, maintaining a normal single core too. I was not able to acheive that when testing ccx overclock, singlecore was always around 500. My voltage ceiling for it was 1.25V, I think I was stable at 4.4-4.4-4.3-4.3, well, my 3900x it's from launch so that's pretty much it hehe. 3900XT do better. EDC/pbo bug did nothing here too, actually I got worse clocks a lower voltage, but I did not want to mess/interfere with FIT so I just let it go. Let's see what's this 1usumus tool is about, if I get extra 7% more performance in singlecore for free I'm happy.

What is your current setting for 7744/508? I could to 7718/507 with 4.35ghz allcore 1.25v. 4.4ghz 1.25v was not stable.
 

·
Registered
Joined
·
22 Posts
@alasdairvfr 7908/520 it's a really great multicore score, maintaining a normal single core too. I was not able to acheive that when testing ccx overclock, singlecore was always around 500. My voltage ceiling for it was 1.25V, I think I was stable at 4.4-4.4-4.3-4.3, well, my 3900x it's from launch so that's pretty much it hehe. 3900XT do better. EDC/pbo bug did nothing here too, actually I got worse clocks a lower voltage, but I did not want to mess/interfere with FIT so I just let it go. Let's see what's this 1usumus tool is about, if I get extra 7% more performance in singlecore for free I'm happy.

What is your current setting for 7744/508? I could to 7718/507 with 4.35ghz allcore 1.25v. 4.4ghz 1.25v was not stable.
Right now i'm 44.5/43.25 with a vcore set to 1.244. It's happy where it is. I tried testing for stability at 45 on ccd1 but even at 1.28vcore still didn't get there. I feel my chip would need more voltage than my cooler can cool for beyond a simple CB run. I tried even one or the other CCX at 45 leaving the rest as my current stable config so at least my single core count could stay high but no luck. Not that I'm complaining, I am pretty happy with my binned chip. that's really all the 3900XT is - a 3900X that promises better IPC for the power.

When the 1usmus tuner comes out i'll wipe the bios and give it a whirl, might as well see if his tool does better than PBO. It's going to be interesting to see what it finds as a sweet spot. I was disappointed that Anthony on LTT didn't try to push it (he went for undervolt instead which is fair, just not exciting) or try it on a chip that had a wide CCD performance gap. Something that would take a user some time to discover, would like to see how/if the tool handles that.
 

·
Registered
Joined
·
236 Posts
There is something in my bios thats holding me back i feel I got 7632 on cinebench on a oc [email protected] and [email protected] @1.35v. When watching hwinfo64 the actual voltage was 1.324v max. Like before for some reason my scores are 200 points behind everyone else. Im water cooled and cinebench hits 4.03ghz underload @65c, prime95 hits 4.2ghz [email protected]

I dont know whats holding back cinebench from using higher clock speeds. Maybe my bios version I really dont know its strange if i use negative offset of 0.075 or higher i hit 7180 mark. But single core performance and gaming stability is worse.

Just to clarify everyone is using cinebench release 20 right?

added update: shutting down extra software i always run like speccy, icue, razor software and hwinfo64 i hit 7120.
 

·
Registered
Joined
·
22 Posts
There is something in my bios thats holding me back i feel I got 7632 on cinebench on a oc [email protected] and [email protected] @1.35v. When watching hwinfo64 the actual voltage was 1.324v max. Like before for some reason my scores are 200 points behind everyone else. Im water cooled and cinebench hits 4.03ghz underload @65c, prime95 hits 4.2ghz [email protected]

I dont know whats holding back cinebench from using higher clock speeds. Maybe my bios version I really dont know its strange if i use negative offset of 0.075 or higher i hit 7180 mark. But single core performance and gaming stability is worse.

Just to clarify everyone is using cinebench release 20 right?

added update: shutting down extra software i always run like speccy, icue, razor software and hwinfo64 i hit 7120.
So there is voltage being read at the CPU and again at the socket on the board. Look for the other value on hwinfo64. Most people find that the board runs 20-30 mv higher than the chip itself. There has been discussion (and I believe the jury is still out) whether the 'max safe' would be at the socket or chip level since we OC'ers like to fly fairly close to the sun. Also, different boards behave differently, some boards pump more power, some boards stick to the voltage where the chip is lower, my board tends to put my chip at the 1.244 as is set in bios but the board itself is reading 1.264 (20mv higher).

Since I'm happy with my clock speed and higher clock requires MUCH more voltage and resultant heat, I'm able to sustain what I believe to be a safe longterm voltage. Also, like everything else the true safe voltage is a variable number based on your ambient temp/cooling, the peculiarities of the chip, amount of load, etc.

Whether at 1.35v or 1.325v, I feel you are close to if not slightly beyond a safe longterm static voltage. A few CB20 runs won't be a problem here but if you are wondering why you are missing 200points, a couple of things. 1. make sure you set your priority to high or realtime in taskmgr.exe that is worth 1-200 points right there. It pauses all daemons and even freezes your screen while the test runs. It helps mitigate the need to manually turn stuff off for benching as well.

The other thing, for all-core your 45 on ccd0 is great but if you could sacrifice 50mhz there to add 100-125 on ccd1, you would net gain. I played around with all this including nerfing my weaker ccd and really your single core gain is at the cost of avg core performance.
 

·
Registered
Joined
·
951 Posts
So there is voltage being read at the CPU and again at the socket on the board. Look for the other value on hwinfo64. Most people find that the board runs 20-30 mv higher than the chip itself. There has been discussion (and I believe the jury is still out) whether the 'max safe' would be at the socket or chip level since we OC'ers like to fly fairly close to the sun. Also, different boards behave differently, some boards pump more power, some boards stick to the voltage where the chip is lower, my board tends to put my chip at the 1.244 as is set in bios but the board itself is reading 1.264 (20mv higher).

Since I'm happy with my clock speed and higher clock requires MUCH more voltage and resultant heat, I'm able to sustain what I believe to be a safe longterm voltage. Also, like everything else the true safe voltage is a variable number based on your ambient temp/cooling, the peculiarities of the chip, amount of load, etc.

Whether at 1.35v or 1.325v, I feel you are close to if not slightly beyond a safe longterm static voltage. A few CB20 runs won't be a problem here but if you are wondering why you are missing 200points, a couple of things. 1. make sure you set your priority to high or realtime in taskmgr.exe that is worth 1-200 points right there. It pauses all daemons and even freezes your screen while the test runs. It helps mitigate the need to manually turn stuff off for benching as well.

The other thing, for all-core your 45 on ccd0 is great but if you could sacrifice 50mhz there to add 100-125 on ccd1, you would net gain. I played around with all this including nerfing my weaker ccd and really your single core gain is at the cost of avg core performance.
Interesting, one thing I have noticed with my MSI Meg X570 Ace and Ryzen 9 3950X is when I set for example my vcore in the bios at 1.1875 for 4.2ghz all core clock it would read 1.169 sometimes even 1.150vcore in Hwinfo64's CPU Core Voltage (SV12 TFN). Now I have learned not to even touch the LLC level and my previous C7H taught me well having measured actual vs read vcore and it was exact when using AUTO on LLC and with it set high it would fluctuate lower and higher all the time, not constant.

I have also learned where the vdroop I mentioned before is between 0.0185 and 0.025 this would drop even more when using a lower vcore such as 1.175 vcore set in bios as here the drop would be to 1.144 and 0.031v or there about. Now this would vary with most motherboards but I am really happy with the board I have. It was hard for me dropping Asus but I'll recommend this board to any person.
 

·
Registered
Joined
·
236 Posts
So there is voltage being read at the CPU and again at the socket on the board. Look for the other value on hwinfo64. Most people find that the board runs 20-30 mv higher than the chip itself. There has been discussion (and I believe the jury is still out) whether the 'max safe' would be at the socket or chip level since we OC'ers like to fly fairly close to the sun. Also, different boards behave differently, some boards pump more power, some boards stick to the voltage where the chip is lower, my board tends to put my chip at the 1.244 as is set in bios but the board itself is reading 1.264 (20mv higher).

Since I'm happy with my clock speed and higher clock requires MUCH more voltage and resultant heat, I'm able to sustain what I believe to be a safe longterm voltage. Also, like everything else the true safe voltage is a variable number based on your ambient temp/cooling, the peculiarities of the chip, amount of load, etc.

Whether at 1.35v or 1.325v, I feel you are close to if not slightly beyond a safe longterm static voltage. A few CB20 runs won't be a problem here but if you are wondering why you are missing 200points, a couple of things. 1. make sure you set your priority to high or realtime in taskmgr.exe that is worth 1-200 points right there. It pauses all daemons and even freezes your screen while the test runs. It helps mitigate the need to manually turn stuff off for benching as well.

The other thing, for all-core your 45 on ccd0 is great but if you could sacrifice 50mhz there to add 100-125 on ccd1, you would net gain. I played around with all this including nerfing my weaker ccd and really your single core gain is at the cost of avg core performance.
I would keep the overclock if it was stable at no higher then 1.28v would even settle for 4.4ghz on all cores at that voltage but the amount of conflicting information out there is huge. As i can tell some go off whats safe for intel cpu's and amd has said 1.47v idle spikes are ok etc. I imagine i would have to settle at 4.3ghz for 1.28v if possible but then i wonder if would that extend my cpu life with the added heat? Or do i trust amd to manage my cpu for me with it going from 1.0v-1.475?
 

·
Registered
Joined
·
590 Posts
There is something in my bios thats holding me back i feel I got 7632 on cinebench on a oc [email protected] and [email protected] @1.35v. When watching hwinfo64 the actual voltage was 1.324v max. Like before for some reason my scores are 200 points behind everyone else. Im water cooled and cinebench hits 4.03ghz underload @65c, prime95 hits 4.2ghz [email protected]

I dont know whats holding back cinebench from using higher clock speeds. Maybe my bios version I really dont know its strange if i use negative offset of 0.075 or higher i hit 7180 mark. But single core performance and gaming stability is worse.

Just to clarify everyone is using cinebench release 20 right?

added update: shutting down extra software i always run like speccy, icue, razor software and hwinfo64 i hit 7120.
7120 is about right, could be a bit better but it's good. As you can see what I've said about background process proceeds. By the fact you mentioned icue, razor etc you might be running tons of other crap services in the background too and don't realise, that will lower your score even more. As mentioned by another user setting it to a higher priority helps in scoring higher, so does killing explorer.exe and couple other windows related .exe too. You might have more stuff triggering the cpu in the background.

I also don't get are your temperatures, too low. I've just ran cb20 without background stuff did not change priority (I never do) and scored 7277 reaching 79.6C. It's really warm outside, around 28C. What is your vcore range (SVI2 TFN) while running prime95 small fft? I'd fresh install windows in another smaller driver or whatever and run cb20 with nothing else installed. There's no much maginc when the cpu is on auto, or it's a lot of background services, or is temperature, or is a too strong negative offset, or too much vcore. They all should score around 7100-7300 when properly cooled. What's your SoC, VDDG and VDDP set in bios? What board?

edit

trust amd, leave on full auto. maybe a negative offset later but do your stuff on auto for a while and then try a offset.
 

·
Registered
Joined
·
22 Posts
I would keep the overclock if it was stable at no higher then 1.28v would even settle for 4.4ghz on all cores at that voltage but the amount of conflicting information out there is huge. As i can tell some go off whats safe for intel cpu's and amd has said 1.47v idle spikes are ok etc. I imagine i would have to settle at 4.3ghz for 1.28v if possible but then i wonder if would that extend my cpu life with the added heat? Or do i trust amd to manage my cpu for me with it going from 1.0v-1.475?
On Auto it will park cores showing a higher voltage or run closer to 1.5 when the current is very low. It shouldn't/won't stay at that voltage under any real load. It freaks everyone out a bit at first especially the wild swings in vcore. A negative offset will improve performance for most people the score to a point until you start seeing clock stretching which is the gap between max effective clock and max clock (perf).

Clock stretching really threw me for a loop at the beginning since I typically do stock clocks lean back vcore till it crashes to see what my minimum stable voltage is at stock. I like that as an indicator of headroom. Needless to say this thing appeared to undervolt forever till I realized what's happening.
 

·
Registered
Joined
·
236 Posts
7120 is about right, could be a bit better but it's good. As you can see what I've said about background process proceeds. By the fact you mentioned icue, razor etc you might be running tons of other crap services in the background too and don't realise, that will lower your score even more. As mentioned by another user setting it to a higher priority helps in scoring higher, so does killing explorer.exe and couple other windows related .exe too. You might have more stuff triggering the cpu in the background.

I also don't get are your temperatures, too low. I've just ran cb20 without background stuff did not change priority (I never do) and scored 7277 reaching 79.6C. It's really warm outside, around 28C. What is your vcore range (SVI2 TFN) while running prime95 small fft? I'd fresh install windows in another smaller driver or whatever and run cb20 with nothing else installed. There's no much maginc when the cpu is on auto, or it's a lot of background services, or is temperature, or is a too strong negative offset, or too much vcore. They all should score around 7100-7300 when properly cooled. What's your SoC, VDDG and VDDP set in bios? What board?

edit

trust amd, leave on full auto. maybe a negative offset later but do your stuff on auto for a while and then try a offset.
I did use a negative offset lost single core performance as a result and in the end i reverted back to auto, I would of liked to disable llc but on this board its strangly worded such as normal, standard, low, meduim llc and so on. As for the motherboard its the Gigabyte x570 gaming x X570 GAMING X (rev. 1.0) | Motherboard - GIGABYTE Australia I wish i had paid an extra $100 for the better version thats the one most people seem to have and there are more guides for it. But its becoming more easy to cross reference overclocking information across different motherboards as I learn more. After trying to limit all background programs and setting cinebench on high in task manager I hit 7190 willing to bet the missing score is because of my security software running. As for voltages i left everything on auto expect my ram voltage. Also wanted to ask, the adia64 memory benchmark will it show better readings as a result of better ram timings or is it something that fluctuates?

Forgot to mention my temps, I dont know what to really say expect ryzen masters readings for my temps are a alot lower then hwinfo64 and speccy. I messed around with the amount of pressure I applied by hand on the water cooler screws till i saw lower temps. Basically made it hand tight without over applying pressure. At first I suspected I applied to much pressure and saw crazy spikes and that stopped after loosening the pressure i applied. I then reapplied thermal paste in a square fashion, I also keep both water cooler radiator fans at about 2100rpm. Which is surprisingly quite and I see a max of 71c underload heavy load in prime 95 compared to 89c when i applied to much pressure on the watercooler.
 

Attachments

·
Registered
Joined
·
236 Posts
On Auto it will park cores showing a higher voltage or run closer to 1.5 when the current is very low. It shouldn't/won't stay at that voltage under any real load. It freaks everyone out a bit at first especially the wild swings in vcore. A negative offset will improve performance for most people the score to a point until you start seeing clock stretching which is the gap between max effective clock and max clock (perf).

Clock stretching really threw me for a loop at the beginning since I typically do stock clocks lean back vcore till it crashes to see what my minimum stable voltage is at stock. I like that as an indicator of headroom. Needless to say this thing appeared to undervolt forever till I realized what's happening.
Yeah im in a strange spot i found a negative offset to cause instability while gaming but then also it could of been because of my old ram tmings which i significantly improved in the last 2 days and i saw even greater fps gains while gaming. I would have to test a negative offset again.
 
121 - 140 of 185 Posts
Top