Overclock.net banner

1 - 19 of 19 Posts

·
Registered
Joined
·
17 Posts
Discussion Starter · #1 ·
I would like to get the most performance out of my 5950x. I noticed that CCD0 draws much higher power, at the same frequency, than CCD1. At idle, CCD0 seems to draw about 10W, and CCD1 draws about 5W. Under load, though, CCD0 draws something like 90W, and CCD1 only draws 70W, again with the same multiplier. Consequently, CCD0 gets 5-6C higher than CCD1.

Initially I thought this meant CCD1 was the "good" chiplet, since it was more efficient. But actually, CCD0 can hit higher frequencies (vcore is the same for both with this chip) than CCD1 without errors. So, CCD1 is more power efficient, but it also has a lower frequency ceiling.

Is it possible to adjust power consumption per-CCD with a fixed vcore? I am using a B550 Aorus Master. I am not sure how to best optimize both chiplets with the tools available on this chipset.
 

·
Totally Tubular
Joined
·
1,644 Posts
As far as I know, there are no controls for that on any CPU. The cores draw what they draw. You can limit the total power draw, but not per-CCD or even per-core. At least, not that I've seen anyway.

Even on monolithic CPUs, temperature varies. My 10850K often has 7°C differences in maximum temperature between the coolest and hottest cores during stress tests. When there's less load - like gaming - there's less variation. It makes sense that the same thing would happen in a chiplet design.
 

·
Registered
Joined
·
17 Posts
Discussion Starter · #5 ·
You can control the voltages for each core under Curve Optimizer in the BIOS
If I enable PBO and CO, does this allow me to set per-core voltage offsets without changing the set frequency? Or will this also shift my frequencies to the AMD set ones?
 

·
Registered
Joined
·
199 Posts
If I enable PBO and CO, does this allow me to set per-core voltage offsets without changing the set frequency? Or will this also shift my frequencies to the AMD set ones?
No, stick with how you're doing it and focus on finding the efficiency sweet spot.
 

·
Iconoclast
Joined
·
30,997 Posts
But actually, CCD0 can hit higher frequencies (vcore is the same for both with this chip) than CCD1 without errors. So, CCD1 is more power efficient, but it also has a lower frequency ceiling.
This is typical. Higher leakage parts tend to clock higher, with the same voltage, but need more power and produce more heat. CCD0 is usually the faster binned, less efficient, chip on dual CCD AMD parts.
 
  • Rep+
Reactions: blu3dragon

·
Totally Tubular
Joined
·
1,644 Posts
CCD0 is usually the faster binned, less efficient, chip on dual CCD AMD parts.
Does anybody know why that is? Unless they're doing it on purpose, shouldn't it be random?
 

·
Registered
Joined
·
145 Posts
Does anybody know why that is? Unless they're doing it on purpose, shouldn't it be random?
It seems common enough to be intentional. Could be yield related (only so many perfect die around) or that it actually gives better all round performance (small number of threads clock high on CCD0, when you have a large number of threads the average clock is better due to the lower power draw of CCD1).
 

·
Registered
Joined
·
145 Posts
Best way to optimize would be to enable PBO and the tune the core offsets.

Also tune your flck, memory clock & timings.
 

·
Registered
Joined
·
4,317 Posts
You can control the voltages for each core under Curve Optimizer in the BIOS
you can but in the end the worst ccx is the one that owns the whole hierarchy as this voltages are not independent they affect the whole ccx and eventually both ccx's


Does anybody know why that is? Unless they're doing it on purpose, shouldn't it be random?

the way pbo behaves is very inefficient. pbo is good for ppl who dont want to hazzle with overclocks

i prefer a static overclock on this system perse.. i can boost 6x48x cores and 6x47x cores instead of 1 core a 49x-51.50x dancing around the macarena and my max pbo tweaked all cores is between 46-47x mostly 46.50x using more voltage & upto 20c hotter than my 48x ccx1 y 47x ccx2 its a no brainer man lol

static perf mhz per mhz in 99% of cases is better on static overclock as pbo have some perf impact on the same multi
cb single core tests im just 3% behind @ 48x vs pbo boosting.. thats best case scenario for pbo 1x boost lol
 

·
Cheesebumps!!
Joined
·
2,420 Posts
this also happens on my previous 5900x..nothing you can do..lmao..

going further below just gives instability..my goto solution was to sell the chip off to RGB boys..and call it a day..let them enjoy the bad binned chip on DDR4 3200..
 

·
Registered
Joined
·
199 Posts
I've run real-world gaming tests with steam, origin, vms and firefox open in the background and 4300 all-core beat PBO w/CO. Delta is even bigger running an HEVC encode as well.

I've even backed off to a per-CCX clock of 4150/4100 because it requires a full 0.05v more under load for 4300, which gives off quite a bit more heat.
 

·
Registered
Joined
·
17 Posts
Discussion Starter · #15 ·
I've even backed off to a per-CCX clock of 4150/4100 because it requires a full 0.05v more under load for 4300, which gives off quite a bit more heat.
Mine is set to around 4.3/4.15 at 1.19v. Under full load it uses 150-155W.
 

·
Registered
Joined
·
4,009 Posts
I've run real-world gaming tests with steam, origin, vms and firefox open in the background and 4300 all-core beat PBO w/CO. Delta is even bigger running an HEVC encode as well.

I've even backed off to a per-CCX clock of 4150/4100 because it requires a full 0.05v more under load for 4300, which gives off quite a bit more heat.
I think it's important to talk about what kind of frequencies are you getting with PBO enabled. I see a solid 4.5Ghz to 4.55Ghz with PBO enabled on all 16 cores of my 5950X. It goes higher if there's less cores involved. There's no way an all core overclock of 4.3Ghz will be better.

Isn't 4.15/4.1 below stock boosting profile on all cores? I think my 5950X did about 4.3 to 4.35 on all cores right out of the box.
 

·
Registered
Joined
·
199 Posts
I think it's important to talk about what kind of frequencies are you getting with PBO enabled. I see a solid 4.5Ghz to 4.55Ghz with PBO enabled on all 16 cores of my 5950X. It goes higher if there's less cores involved. There's no way an all core overclock of 4.3Ghz will be better.

Isn't 4.15/4.1 below stock boosting profile on all cores? I think my 5950X did about 4.3 to 4.35 on all cores right out of the box.
It depends entirely on workload and your use case. What kind of clockspeed and temps do you see under Prime95 AXV2 small fft?

I'm using 16 cores 24/7, so for me lower core boosts are irrelevant. If you are using the full CPU regularly, especially with AVX loads, you're going to need some hefty cooling to sustain 4.5GHz for hours.
 

·
Registered
Joined
·
4,009 Posts
It depends entirely on workload and your use case. What kind of clockspeed and temps do you see under Prime95 AXV2 small fft?

I'm using 16 cores 24/7, so for me lower core boosts are irrelevant. If you are using the full CPU regularly, especially with AVX loads, you're going to need some hefty cooling to sustain 4.5GHz for hours.
I see 4.55Ghz in Cinebench R20/R23 all core with just PBO.
 

·
Registered
Joined
·
199 Posts
I see 4.55Ghz in Cinebench R20/R23 all core with just PBO.
Right, but Cinebench doesn't stress as hard as some other items can. The temp delta between cinebench and AVX2 small fft for me is 20C.

What do you get running prime?
 
1 - 19 of 19 Posts
Top