Overclock.net banner

221 - 240 of 292 Posts

·
Registered
Joined
·
4,317 Posts
amd or asus mess up my power table with the new agesa... now my tweaked power table my main 2 active cores on idle are 0/1 instead of by design which are 1/5 on the first ccx which affect latency sensitive apps as is trying to use the 1st core instead of the fastest one.

Power plan would park and turn on cores as needed including 2nd ccx best cores which are core 9&11 which will be turned first than the rest of the the 2nd ccx in that order

it will do 1/5 then all of 1st ccx then it jumps to 9&11 as needed then if need all it would use the cores in order of "performance quality of the core" now it just go in order as cores...
Edit:
btw i just rolled back the bios to B550-I 1803 and yes its back to how i have it.. and how it should be.. 1&5 active on idle and all the other parked. This are the cores and cores usage according to be used 1st
 

·
Registered
Joined
·
31 Posts
Hey guys.

So i now have two fully tweaked settings. 1. PBO(0mhz offset) with agressive curve and -0.05v offset on top 2. per CCX overclock 4,6/4,5 with 1.225vcore (1.206 after drop)

both fully stable. While PBO gives me better SC of course, CCX gives me better performance in modern Games while consumin way less power beauce of the lack of unnecessary boost with high vcore. Ok nothing new here...we all know that.

But how can i get the same low idle power draw(and light task like watching a youtube video) with CCX OC? somehow PBO is always a bit less in idle draw
 

·
Registered
Joined
·
101 Posts
Good day all,

Just received my 5950x and I am looking to maximize the chip as best I can.

When looking at PBO I noticed some odd behavior.

If I left the PBO settings at stock, both CCXs would boost the same (effective clock)

If I changed the EDC value I would see that CCX0 would boost up but CCX1 would lag behind by a few hundred Mhz. The CCX effective clock would sync back up if I put the EDC to 200

Is this know behavior or is there something strange going on?

Cheers,

Chris
 

·
Registered
Joined
·
275 Posts
Good day all,

Just received my 5950x and I am looking to maximize the chip as best I can.

When looking at PBO I noticed some odd behavior.

If I left the PBO settings at stock, both CCXs would boost the same (effective clock)

If I changed the EDC value I would see that CCX0 would boost up but CCX1 would lag behind by a few hundred Mhz. The CCX effective clock would sync back up if I put the EDC to 200

Is this know behavior or is there something strange going on?

Cheers,

Chris
If you have a Gigabyte board that is typical behavior. I tested my 5900x in an MSI board and clocks were uniform between CCX's. Try EDC at motherboard limit and they clocks should be the same.
 

·
SAY AMBIENT AGAIN! SAY IT
Joined
·
195 Posts
Finished prepping my 5950x for my optimus foundation block. It started out pretty warped.
2485260

And now it looks like this:
2485261

It took about 3 hours total progressing from 400 to 2000 grit sand paper. I wet sanded the entire time alternating method between 4 sets of 10 strokes vertically with a 90 degree rotation then 4 sets of 4 figure-eights each with a 90 degree rotation.
 

·
Registered
Joined
·
199 Posts
Finished prepping my 5950x for my optimus foundation block. It started out pretty warped.
View attachment 2485260
And now it looks like this:
View attachment 2485261
It took about 3 hours total progressing from 400 to 2000 grit sand paper. I wet sanded the entire time alternating method between 4 sets of 10 strokes vertically with a 90 degree rotation then 4 sets of 4 figure-eights each with a 90 degree rotation.
That's courage, goodbye warranty. :)

Please keep us updated, I'm deciding on my my loop setup as well.
 

·
SAY AMBIENT AGAIN! SAY IT
Joined
·
195 Posts
That's courage, goodbye warranty. :)

Please keep us updated, I'm deciding on my my loop setup as well.
I've never had to warranty a CPU so I'm not too concerned. I de-lidded my 6700k and that was always fine. I pencil modded my old AMD chips back in the day too. And if something happens I can just move my 3950x from my server back to the gaming PC and put the 3700x back in the server.
 

·
#sneakattack
Joined
·
452 Posts
I'm curious to know if anyone else has experienced an all-core frequency "wall" with their CPUs. My 5950X seems like a pretty solid sample, as I've been running 4.7GHz all cores at 1.22V using Realbench 2.56, OCCT with AVX2, CBR23 loops, and my own workloads to test stability. The voltage steps for each 100MHz seem quite reasonable up to 4.7GHz. However, 4.8GHz is extremely problematic for me. I can sometimes run CBR23 at 4.8GHz all cores at 1.26V, but it's definitely not stable. Even as I keep increasing the voltage, it doesn't become stable. The behavior at 1.31V is identical to the behavior at 1.26V: I can run CBR23 a few times, but it'll eventually crash after enough runs. At 1.26V, peak temps on my hottest CCD is about 83-85C with the fluid temp of my CPU loop at 24-26C. At 1.31V the hottest CCD peaks at 91C and the CPU is incredibly inefficient at this point.

So basically, 4.8GHz seems to be a wall for my CPU. I can throw plenty of voltage at it, but that doesn't matter. Most of the time when it crashes, I get a "CPU over temperature error," which seems to be a false positive. So has anyone else experienced such behavior at a particular frequency? I've messed around with just about every relevant BIOS setting I can to see if something else could be the culprit, but zero luck. I've basically concluded that my CPU simply can't manage a stable 4.8GHz all cores, which is disappointing since it can manage 4.7GHz at pretty low voltages.
 

·
Registered
Joined
·
83 Posts
Finished prepping my 5950x for my optimus foundation block. It started out pretty warped.
View attachment 2485260
And now it looks like this:
View attachment 2485261
It took about 3 hours total progressing from 400 to 2000 grit sand paper. I wet sanded the entire time alternating method between 4 sets of 10 strokes vertically with a 90 degree rotation then 4 sets of 4 figure-eights each with a 90 degree rotation.
have same WB and 5950x too. Would be great to know difference it temps) Please let me know)
 

·
Registered
Joined
·
1,178 Posts
you are pretty happy because of what exactly? by seeing random numbers in CTR? shouldnt you show us...maybe some benchmark results so we can relate to your happiness? lol^^

makes total sense that you get a better quality reading in CTR by setting a negative curve beforehand. CTR reads a lower VID then and doesnt "know" its not stock
Okay sorry for my late reply. Now I understand this maybe not the best, but I am pretty happy with it as is. Had a ton of issues with my setup yesterday where it will not even start up and then all of a sudden it started up. Unstalled the app I installed when it happened and now seem to be all good.

2487240
 

·
Registered
Joined
·
123 Posts
I finally got a working 5950X and started working with Curve Optimizer...

On second CCX any core will work with >= -20 offset. On first CCX most cores will do -12, the 3rd/4th best -8, the best core -6, and weirdly the 2nd best core won't work even with -1 offset!

Anybody has seen this behavior before?
 

·
Cheesebumps!!
Joined
·
2,420 Posts
I finally got a working 5950X and started working with Curve Optimizer...

On second CCX any core will work with >= -20 offset. On first CCX most cores will do -12, the 3rd/4th best -8, the best core -6, and weirdly the 2nd best core won't work even with -1 offset!

Anybody has seen this behavior before?
this was my 5900x behaviour as well..2nd CCD is crap and 1st one does great on CO..also I am limited to 3733 IF Clocks..so I sold it...too much of a pain..
 

·
Registered
Joined
·
199 Posts
I finally got a working 5950X and started working with Curve Optimizer...

On second CCX any core will work with >= -20 offset. On first CCX most cores will do -12, the 3rd/4th best -8, the best core -6, and weirdly the 2nd best core won't work even with -1 offset!

Anybody has seen this behavior before?
Yep, it's normal; not all cores are created equal. They also tend to match a slightly weaker CCD1 to improve yeilds.

Intel is the same, I can get 300MHz more at the same voltage with two cores disabled on my 8750H. Intel just doesn't give you the option to tweak by core.

The holy grail would be allowing per-core manual overclock.
 

·
#sneakattack
Joined
·
452 Posts
Yep, it's normal; not all cores are created equal. They also tend to match a slightly weaker CCD1 to improve yeilds.

Intel is the same, I can get 300MHz more at the same voltage with two cores disabled on my 8750H. Intel just doesn't give you the option to tweak by core.

The holy grail would be allowing per-core manual overclock.
Per-core overclocking is supported on X299. I honestly never bothered with it outside of a little experimentation since it's a lot of work tuning 18 cores separately for a very small performance improvement over an all-core OC. The largest gains would be in single-threaded applications.
 

·
Registered
Joined
·
199 Posts
Per-core overclocking is supported on X299. I honestly never bothered with it outside of a little experimentation since it's a lot of work tuning 18 cores separately for a very small performance improvement over an all-core OC. The largest gains would be in single-threaded applications.
There's a 450MHz delta between my best and worse core at a given fixed voltage. I would welcome it.
 

·
Robotic Chemist
Joined
·
3,592 Posts
I have been running -22 all core on the Curve Optimizer and a -62.5mV voltage offset for the last two weeks. It seems stable, zero issues at these settings so far, but it will be a lot longer before I really trust it since stability testing the curve optimizer is tricky unless it is very unstable (-25 crashes fast enough to be sure it doesn't work). -75mV causes lower performance in Cinebench r20. A non-zero core offset also seems to hurt performance, I am not sure why. This is with a 200A EDC. Peak core effective clock reported in HWiNFO was 4999.6 MHz during Cinebench single core.

Any suggestions or comments? Thanks. :)

2487930

2487931
2487938
2487941


Time Spy CPU score: 16662

Time Spy Extreme CPU score: 9886
 

·
Registered
Joined
·
199 Posts
I have been running -22 all core on the Curve Optimizer and a -62.5mV voltage offset for the last two weeks. It seems stable, zero issues at these settings so far, but it will be a lot longer before I really trust it since stability testing the curve optimizer is tricky unless it is very unstable (-25 crashes fast enough to be sure it doesn't work). -75mV causes lower performance in Cinebench r20. A non-zero core offset also seems to hurt performance, I am not sure why. This is with a 200A EDC. Peak core effective clock reported in HWiNFO was 4999.6 MHz during Cinebench single core.
If you're using the offset, you're opening yourself up to clock stretching, which is why it's hurting performance. Stick to just core optimizer and take your time. If you think it's stable, run core cycler overnight, or for 24 hours if you can.

Whatever best setting passes that overnight, reduce all cores by 1 and that's your 24/7.
 
  • Rep+
Reactions: lmfodor

·
Premium Member
Joined
·
6,012 Posts
If you're using the offset, you're opening yourself up to clock stretching, which is why it's hurting performance. Stick to just core optimizer and take your time. If you think it's stable, run core cycler overnight, or for 24 hours if you can.

Whatever best setting passes that overnight, reduce all cores by 1 and that's your 24/7.
Their single-core and multicore is quite good for -22. I'm not sure using an offset is actually hurting performance for them.

Let me test my multicore with and without the same offset. bbiab.

I gained close to 100 multicore R20 points using a .065 offset.
 

·
Premium Member
Joined
·
6,012 Posts
If you're using the offset, you're opening yourself up to clock stretching, which is why it's hurting performance. Stick to just core optimizer and take your time. If you think it's stable, run core cycler overnight, or for 24 hours if you can.

Whatever best setting passes that overnight, reduce all cores by 1 and that's your 24/7.
Actually, I think you're right. With an offset getting unexplained lag in games and Windows. :(
 
221 - 240 of 292 Posts
Top