3900x oc to 4.3 question

K

kandor

Gawd
Joined
Aug 24, 2005
Messages
632
Hi, I have my 3900x running at 4.3 all cores. So that's fine, improves the cinebench r20 score (7500), but by manually forcing all cores to 4.3 it means they are allways there, and I'm missing out on the boost to 4.6 on single threaded apps.

I'm new to the amd bios environment so I'll keep looking and see if I can accomplish both. How do you deal with this?

Thanks,
 
Were you even getting 4.6 confirmed? At lot of folks where getting lackluster single core boost on the 3900.
 
I would see 4.6 on single core just watching the load on Core Temp though so can't speak about the accuracy of that. I'm currently running 1.35v for the 4.3 manual with LLC at 3, so I see a droop to 1.34 under load. Just now though I'm reading that 1.35 may be a little too high? Aida64 peaked just over 90c only for a bit but was average high '70s to mid '80s most of the time.
 
kandor said:
Hi, I have my 3900x running at 4.3 all cores. So that's fine, improves the cinebench r20 score (7500), but by manually forcing all cores to 4.3 it means they are allways there, and I'm missing out on the boost to 4.6 on single threaded apps.

I'm new to the amd bios environment so I'll keep looking and see if I can accomplish both. How do you deal with this?

Thanks,
Click to expand...

Unfortunately, it's either or when you go for an all-core overclock. Do you actually do anything in the realm of multi-threaded workloads or are you only concerned with Cinebench scores? If its the latter, I'd just run PB2 and be done with it.
 
OFaceSIG said:
Were you even getting 4.6 confirmed? At lot of folks where getting lackluster single core boost on the 3900.
Click to expand...

I can get 4.6GHz boost clocks (single-core) out of our 3900X sample after the 1.0.0.4 patch B AGESA code update. Many earlier AGESA updates claimed to address that, but this is the only one that truly did it.
 
kandor said:
I would see 4.6 on single core just watching the load on Core Temp though so can't speak about the accuracy of that. I'm currently running 1.35v for the 4.3 manual with LLC at 3, so I see a droop to 1.34 under load. Just now though I'm reading that 1.35 may be a little too high? Aida64 peaked just over 90c only for a bit but was average high '70s to mid '80s most of the time.
Click to expand...

1.35v is kind of a lot for 4.3ghz all core. Average voltage for 4.3ghz from the dozen 3900x I've seen posted required 1.32v give or take. What sort of cooling do you have?
 
Dan_D said:
Unfortunately, it's either or when you go for an all-core overclock. Do you actually do anything in the realm of multi-threaded workloads or are you only concerned with Cinebench scores? If its the latter, I'd just run PB2 and be done with it.
Click to expand...

Unfortunately I do both, solidworks and FEA are strongly single threaded and rendering / point cloud software is incredibly multi-threaded (like 100% utilization for 8 hours importing data on my previous 8700k). I knew about PB and PBO but not PB2? More research required.
 
Last edited:
kandor said:
Unfortunately I do both, solidworks and FEA are strongly single threaded and rendering / point cloud software is incredibly multi-threaded (like 100% utilization for 8 hours importing data on my previous 8700k). I new about PB and PBO but not PB2? More research required.
Click to expand...

You can do a per core ratio overclock to achieve a best compromise for both single and multi thread performance.

https://hardforum.com/threads/3950x-overclocking-guide-needed.1989231/#post-1044407437
 
  • Like
Reactions: Dan_D
like this
thesmokingman said:
1.35v is kind of a lot for 4.3ghz all core. Average voltage for 4.3ghz from the dozen 3900x I've seen posted required 1.32v give or take. What sort of cooling do you have?
Click to expand...

I have the thermaltake 360 in a Fractal R6 case. I'll try the voltage a little lower, thanks for the link - I'll try it and see how it goes!
 
You can not boost if you manually overclock all cores, as far as I can tell. I ran it up to 4.4Ghz but only could get it when I ran all core to 4.3 in the BIOS and used RM to get to 4.4Ghz. Running PBO I get a 4.55 according to HWInfo64 but only about a 4.45 according to Ryzen Master with the Peak bar enabled to see what cores peak to what.
 
kandor said:
Unfortunately I do both, solidworks and FEA are strongly single threaded and rendering / point cloud software is incredibly multi-threaded (like 100% utilization for 8 hours importing data on my previous 8700k). I knew about PB and PBO but not PB2? More research required.
Click to expand...

What is your all core boost at stock? Mine is usually around 4.2GHz when doing something like encoding. Doesn't really drop below that unless i'm stress testing. Assuming you're getting the same, a 4.3GHz manual all core overclock seems pretty pointless and even counter productive considering you no longer get the single core boost.
 
Dan_D said:
Unfortunately, it's either or when you go for an all-core overclock. Do you actually do anything in the realm of multi-threaded workloads or are you only concerned with Cinebench scores? If its the latter, I'd just run PB2 and be done with it.
Click to expand...

I'd run PB2 no matter what and be done with it. These chips are designed best to use the algorithms AMD designed them around. Overclocking has been dead for a decade. It's just a marketing term now.

Overclocking was back when a 3.2ghz chip was a 3.2ghz chip with no boost or any of that shit. You had to push it to 3.8 or 4.0 or whatever outside of spec and violating warranty. That was overclocking.

Overclocking was an art of base clock, front side bus, and multiplier, and voltages. Not like today where the chip sells out of the box fully capable of auto clocking to the highest stable frequency.
 
tangoseal said:
I'd run PB2 no matter what and be done with it. These chips are designed best to use the algorithms AMD designed them around. Overclocking has been dead for a decade. It's just a marketing term now.

Overclocking was back when a 3.2ghz chip was a 3.2ghz chip with no boost or any of that shit. You had to push it to 3.8 or 4.0 or whatever outside of spec and violating warranty. That was overclocking.

Overclocking was an art of base clock, front side bus, and multiplier, and voltages. Not like today where the chip sells out of the box fully capable of auto clocking to the highest stable frequency.
Click to expand...

Except this isn't true. At least not on the Intel side.
 
RamonGTP said:
What is your all core boost at stock? Mine is usually around 4.2GHz when doing something like encoding. Doesn't really drop below that unless i'm stress testing. Assuming you're getting the same, a 4.3GHz manual all core overclock seems pretty pointless and even counter productive considering you no longer get the single core boost.
Click to expand...

If I'm loading the chip all cores, it's running at stock 3.8, so manual of 4.3 makes a big difference: eg. 7500 cinebenchr20 up from 6950
 
kandor said:
If I'm loading the chip all cores, it's running at stock 3.8, so manual of 4.3 makes a big difference: eg. 7500 cinebenchr20 up from 6950
Click to expand...

What kind of real-world load are you putting on it that it goes all the way down to 3.8? Do you have PBO enabled? Even your CB20 scores is only a 7% difference
 
RamonGTP said:
What kind of real-world load are you putting on it that it goes all the way down to 3.8? Do you have PBO enabled? Even your CB20 scores is only a 7% difference
Click to expand...

It shouldn't drop that low. These should do 4.0-4.1GHz with a heavily multi-threaded workload.
 
Hmm, ok I'll check, still learning my way around amd bios. I tested p95 small fft at 4.3ghz manual and 1.325v and get instant errors. So I'm resetting to stock and keep testing. Thanks for all the info.
 
stop doing an all core OC. OC these multi ccd chips (3900x/3950x) on a per ccd basis. you will get much greater performance and use lower voltage.
 
Ok, I reset to stock and then just changed PBO from auto to enabled. Core voltage on auto runs around 1.43v which seems crazy, but that drops to 1.32 under load. With PBO enabled during cb20 load I'm seeing 4.05 exactly as you guys predicted. So next up I'll test the per core oc as per the above link. cb20 hovers around 7000 now, whereas 4.3 manual was giving 7500+. It makes a difference on 80 hour renders etc!

Thanks again,
 
You also need to factor in that your 4.3 wasn't stable so you'd probably need to push that down to 4.2 for stability and then your 7% performance increase drops to even less than that.
 
True, it was stable though at 1.35, instability in p95 showed at 1.325 so pretty narrow margin I agree.
 
You must log in or register to reply here.
Back
Top