Overclock.net banner
361 - 380 of 410 Posts

·
Registered
Joined
·
116 Posts
That is very odd. I get 17733 Timespy CPU score with only a PBO+curve optimizer overclock (SMT disabled) running 1800 fclk, and it is pretty hot in here (water just over 28°C). I have the same motherboard and I even have a -0.05V offset on the CPU voltage.

Maybe compare scores with an 1800 fclk?

I assume you have this power management stuff disabled too?

Advanced\AMD CBS\NBIO Common Options\SMU Common Options
APBDIS: 1
DF Cstates: Disabled
Fixed SOC Pstate: P0
CPPC: Enabled
CPPC Preferred Cores: Enabled

Edit: I don't think it has anything to do with APBDIS, setting it back to auto didn't reduce my scores at all. Very confusing!

Tried it, my score moved a little, but not much. Still in the 16 000 pts.
 

·
Registered
Joined
·
1,187 Posts
Did a fresh install and started back from scratch: I scored 0 in Time Spy Custom
18350 points.
I don't know what it was, but it was severely holding back my score.

View attachment 2515410
I seriously do not understand how you able to get this score with PBO. I barely get 16k cpu score when using PBO with a -24 all core curve optimizer. I have tried various means, I reach just over 17k cpu score when I manually overclock my CPU to 4.7ghz.

I just realised one thing, if I disable SMT my cpu score jumped up by about 1000 points
 

·
Registered
Joined
·
116 Posts
I seriously do not understand how you able to get this score with PBO. I barely get 16k cpu score when using PBO with a -24 all core curve optimizer. I have tried various means, I reach just over 17k cpu score when I manually overclock my CPU to 4.7ghz.

I just realised one thing, if I disable SMT my cpu score jumped up by about 1000 points
Yeah, as I had previously stated, specifically for Timespy, disabling SMT on a 5950X can boost your score by roughly a 1000 points.

However, in my case, it seems like my windows install had hella troubles with that bench in particular. Might be due to the log or registry filled with previous mem OC errors... Who knows?
 

·
Registered
Joined
·
1,187 Posts
Yeah, as I had previously stated, specifically for Timespy, disabling SMT on a 5950X can boost your score by roughly a 1000 points.

However, in my case, it seems like my windows install had hella troubles with that bench in particular. Might be due to the log or registry filled with previous mem OC errors... Who knows?
Well for me to get 18000 to 18200 cpu score my cpu need to run at manual 4750mhz. I'll try PBO now to see the difference.
 

·
Registered
Joined
·
1,187 Posts
Yeah, as I had previously stated, specifically for Timespy, disabling SMT on a 5950X can boost your score by roughly a 1000 points.

However, in my case, it seems like my windows install had hella troubles with that bench in particular. Might be due to the log or registry filled with previous mem OC errors... Who knows?
Just tried PBO and highest I am getting is around 17800 to 17900 CPU score.
 

·
Registered
Joined
·
204 Posts
I seriously do not understand how you able to get this score with PBO. I barely get 16k cpu score when using PBO with a -24 all core curve optimizer. I have tried various means, I reach just over 17k cpu score when I manually overclock my CPU to 4.7ghz.

I just realised one thing, if I disable SMT my cpu score jumped up by about 1000 points
I'd recommend giving Corecycler a go if you haven't to 100% make sure that is stable CoreCycler - tool for testing Curve Optimizer settings

When I first got my 5950x I fired in -20, then -25 all core and was booting fine and even benching. On closer inspection, out of my top 4 cores, I'm currently at -15 on 3 and -5 on one. Yes, -5. It was a 2nd best core as well. The rest are on -25 still.

I've still to optimise the other best 3 cores, I haven't tried them higher than -15 with Corecycler and I might be able to push some of the rest on -25 higher. Point is though you could have an unstable core pulling your clocks down without even necessarily knowing it (no crashing). That one core for me was failing until I dropped it right down to -5.

My PBO settings are 270 PPT, 150 TDC, 190 EDC. I have +100 on mhz boost.

Sorry for jumping in if you're 100% stable and all that, just thought I'd throw some potentials up for lower performance when I seen you say -24 all core.
 

·
Registered
Joined
·
1,529 Posts
Negative CO offset on fast cores does not improve your multi-core performance, but only single-core and low-core situations. Multi-core clock-rate is limited by the slowest currently active core.
 

·
Registered
Joined
·
1,529 Posts
Here are some CB23 comparison numbers to demonstrate that diminishing returns result in increasing wattage being of rather limited use:

Stock: 21640 (134 W)
CO (PBO limits stock!): 22220 (132 W)
4600 fixed: 23450 (174 W)

5.5% more performance at the cost of 32% more power consumption. Ouch!

To reach the "4600 fixed" score via PBO limits I have to increase Scalar, performance and load are the same then (+2W for PBO).
 

·
Registered
Joined
·
116 Posts
Here are some CB23 comparison numbers to demonstrate that diminishing returns result in increasing wattage being of rather limited use:

Stock: 21640 (134 W)
CO (PBO limits stock!): 22220 (132 W)
4600 fixed: 23450 (174 W)

5.5% more performance at the cost of 32% more power consumption. Ouch!

To reach the "4600 fixed" score via PBO limits I have to increase Scalar, performance and load are the same then (+2W for PBO).
It's definitely not 4.6GHz fixed... Unless you have a 5900X...

For measure, 4.65GHz fixed on Cinebench R23 results in this (my result), I had to fetch it back from a Reddit post from a while:

2515696
 

Attachments

·
Registered
Joined
·
1,529 Posts
Yes, 5900X here. The main point remains, diminishing returns make overclocking Zen 3 hardly worth the power consumption and stability efforts, unless you are just trying to break some records.
 

·
Old dog, likes new tricks
Joined
·
2,598 Posts
I seriously do not understand how you able to get this score with PBO. I barely get 16k cpu score when using PBO with a -24 all core curve optimizer. I have tried various means, I reach just over 17k cpu score when I manually overclock my CPU to 4.7ghz.
Ditto bro, I resemble this remark, LOL! I can't break 16K CPU score in Time Spy, but I'll get over 30,100 in Cinebench R23, and that's repeated loops to heat soak the thing as much as possible. I have no idea what to believe as far as how good (or not good) my 5950X is performing.
 

·
Premium Member
Joined
·
20,450 Posts
Know im late to the party but i bought a 5900 for 645$ new. Looking forward to share my results paired with a AurosB550M Pro.
 
  • Rep+
Reactions: Veii

·
Robotic Chemist
Joined
·
3,886 Posts
Has anyone else tested this tool? I tried it, but running the prime 95 stock test results in my best core failing all the time, even at stock settings, which is a little bizarre.
I have used it a lot, it works really well for me. I was able to tune all my cores' curves to within +/- 1 and it is super reliable, if I subtract 1 on the curve for all my cores they all fail the test a some point in an overnight run but if I use my current settings I never see an error. With a -0.05V offset I can use slightly lower curve offsets (higher negative), but I also lose the top one or two frequency steps. My best core can only run -9, but one core can do -29 (the best core still hits higher frequencies).

Some CPUs do need a positive offset on their best core to pass the default Prime95 test, it is not that bizarre because AMD's PBO pushes the silicon near its limits by default, but Prime95 is a pretty extreme stability test too.
 

·
Registered
Joined
·
76 Posts
I used CTR 2.1 RC5 to tune my 5950x to 5 Ghz on the 2 fastest cores at 1.375V for light threaded applications and 4.65 Ghz for multithreaded applications at 1.25V and called it a day. No point in pushing it to 5.1 - 5.2 Ghz and push the voltage to the extreme 1.5V or higher for some unnoticeable gain if any. Temperature stays in the 60s as well at 5Ghz 1.375V in a 30s ambient temp room.
 

·
Premium Member
Joined
·
20,450 Posts
Got my 5900 installed and ran Cinebench 20 to see how high the cpu temp goes at optimized default cooled by Fuma2. Gonna run a loop see where it ends up.
 

Attachments

·
Premium Member
Joined
·
4,391 Posts
Is there a consensus for CO as to whether best or worst cores should be lowest offset? Someone mentioned awhile back in this thread and it makes perfect sense to me that best cores should actually be highest offset, because a good core should be able to clock higher with less voltage. Why do so many people seem to be doing it the opposite way?
 

·
Robotic Chemist
Joined
·
3,886 Posts
Is there a consensus for CO as to whether best or worst cores should be lowest offset? Someone mentioned awhile back in this thread and it makes perfect sense to me that best cores should actually be highest offset, because a good core should be able to clock higher with less voltage. Why do so many people seem to be doing it the opposite way?
While that does make sense, the cores all have custom curves already which we are modifying with the CO offsets. The best cores have more aggressive curves at stock so you usually cannot offset them as much as the lower quality cores.
 
361 - 380 of 410 Posts
Top