Overclock.net banner
Cancel
Create thread New Forums

[Official] Zen 4 X3D Owner's Club (7800x3D / 7900x3D / 7950x3D)

Tags
3d v-cache 7800x3d 7900x3d 7950x3d amd
5 reading
Jump to Latest
6,661 - 6,680 of 9,076 Posts
#6,662 ·
Audioboxer said:
Are you on the GENE? If not I guess it's about what Blameless said, does it effectively double?

Only benching I've checked so far is LinpackExtreme, results still come in around 800~805mb.

5 mins and I'll check AIDA.

Edit - AIDA doesn't seem to show much change outside of MoE. I'm up super early tomorrow for work so I'll do more testing when I'm home and see what variables all have an impact, like bumping VSOC.
Click to expand...
I am. And I highly doubt there is a fault in switching frequency adjustment. You are getting lower core temperature by adjusting SOC switching frequency. It doesn't make sense. There shouldn't be a direct relationship. Especially considering the fact this temperature difference is selective and only affects your frequency CCD (as far as I can see from screenshots).

But I do know it can set suboptimal CPU/memory parameters while training if it doesn't like some of the voltages it gets. And I really don't think this temperature reduction you are getting is a positive change. You couldn't get this reduction without either frequency or voltage dropping. And those are firmly tied by the curve.
 
#6,663 ·
ZhenyaUsenko said:
I am. And I highly doubt there is a fault in switching frequency adjustment. You are getting lower core temperature by adjusting SOC switching frequency. It doesn't make sense. There shouldn't be a direct relationship. Especially considering the fact this temperature difference is selective and only affects your frequency CCD (as far as I can see from screenshots).

But I do know it can set suboptimal CPU/memory parameters while training if it doesn't like some of the voltages it gets. And I really don't think this temperature reduction you are getting is a positive change. You couldn't get this reduction without either frequency or voltage dropping. And those are firmly tied by the curve.
Click to expand...
I'll do more testing when I'm home.

I wasn't testing CCD0, only CCD1. The temps where they were on AUTO made more "sense" though for direct die/LM. I know I've only really run Linpack, but that is a CPU benchmark at least. Still getting 800+ with the cooler temps. Curve values remaining higher as well, it wasn't just temperature.

Can't explain it yet myself, but something in the chain was not happy and preliminary testing was singling out VDDSOC switching frequency. I'll try it back on with a higher VSOC tonight. Or when I say "back on", with higher numbers like 700. It's still going to be on at AUTO.

Are you on FW 2007? I could try going back to 1905 as well and see if anything changes.
 
#6,664 ·
Font Screenshot Number Parallel Rectangle


Based on the recently released Geekbench 5 results of Zen 5 (Strix Point), a Chinese user named MebiuW predicted that there may be a 60% IPC performance improvement.

Rectangle Slope Font Plot Parallel


Geekbench reflects AVX512-based AES encryption performance, so whether Zen 5 supports native AVX512 is also a variable.
We'll have to compare it to the QS's higher clock speeds to know for sure, but we expect a performance increase of over 20%.
 
#6,665 ·
Some benchmark results:

CPU: BCLK 105 - LLC7 - PBO Scaler X6 - Boost Clock (-) 150 MHz = 5150 MHz

VDD: 1.50v
Font Technology Screenshot Software Terrestrial plant

Font Screenshot Software Technology Multimedia
 
  • Rep+
Reactions: tsg and RedF
Save
Reply Quote Rep+ Rep+
#6,666 ·
I'm losing the will to live here :ROFLMAO:

Font Screenshot Multimedia Display device Computer


When OCCT wants/tries to pull large wattage spikes, very quick errors on CCD1 curve.

Screenshot Font Multimedia Software Display device


When it doesn't, it will run fine.

I'm looking at FIT Scalar and Switching Frequencies right now.

To add even more **** to the pile

Computer Font Screenshot Personal computer Technology


When the stability testing is fine and power draw is seemingly not spiking as much, SHA3 is even playing ball. I was 100% convinced with that CCD0 curve it simply wouldn't work, but it is.

But when OCCT/Corecycler is crashing, SHA3 will also BSOD.

Linpack outputting scores around 800 no matter what just adds more to the confusion :ROFLMAO: Though that might suggest no performance is being gained whenever this CPU and the combination of settings set seems to push voltage/wattage higher for whatever reason. Which is good if I'm not losing any performance, I just need to find out what the hell is causing wattage spikes, why and deal with it.

Switching frequencies all on 600 with scalar set to 1x and even 10x seems to be fine. CPU switching frequency going to 700 (its max is 800) then had power spikes again. I don't even know if the process of simply direct die cooling is what is triggering opportunistic settings at times and these then get upset with CCD1 curve offsets. I don't know enough about PBO and half of these settings, let alone have much experience with direct die cooling, to know what to look out for. Most people in here will just be IHS cooling and will have thermals a bit higher on CCD1 as a result.

But I've not got much time to test right now/can't spend time constantly rebooting PC. Will do more testing later.

One question I do have for later on, is when set to AUTO do switching frequencies move about/potentially set at different values per boot?
 
#6,669 ·
ZhenyaUsenko said:
2) At "VDDIO / VDDQ == 1.51V / 1.56V" (-0.05V) - B2 idle power is 1.000W and A2 is still 0.875W (**)
Click to expand...
I was testing VDD = VDDQ = 1.5V since I have seen someone running 8000C34 w that voltage, basically all done mismatch fix stress test done etc.. Then I started playing w nitro and quickly realize 1/3/0 does not work for me current setting. So return to the profile and accidentally set VDDIO = 1.45V instead of one notch down, and this is the lowest idle power I have seen so far on my SR, what do you think?

Font Screenshot Electronic device Computer Software
 
#6,670 ·
eighty20 said:
I was testing VDD = VDDQ = 1.5V since I have seen someone running 8000C34 w that voltage, basically all done mismatch fix stress test done etc.. Then I started playing w nitro and quickly realize 1/3/0 does not work for me current setting. So return to the profile and accidentally set VDDIO = 1.45V instead of one notch down, and this is the lowest idle power I have seen so far on my SR, what do you think?

View attachment 2657201
Click to expand...
I also haven't seen it this low at these voltages. Also 1.5 never worked for me at 34 tCL.
 
#6,671 · (Edited)
ZhenyaUsenko said:
I also haven't seen it this low at these voltages. Also 1.5 never worked for me at 34 tCL.
Click to expand...
...On my older GSkill 7200 2x24 GB kit, I have run CL34 8000 with VDDIO at 1.34V - which I leave untouched no matter what profile I use (48 GB 8000, 8200; 6400 96GB...). BTW, I have VDDQ now at 1.335V (1.34V before) but on my mobo, I can only use 1.34V for VDDIO (no 0.015v steps).
 
#6,672 · (Edited)
ZhenyaUsenko said:
Did you already check Boosting behaviour with what you call "wattage spikes" compared to "no wattage spikes"?
Click to expand...
Time to check my sanity more like :ROFLMAO:

I've only had an hour of testing things since getting home but I've discovered more weirdness. Seemingly, if my task tray apps are open, notably discord, there is a chance OCCT won't draw as much power. I normally close task tray apps before all serious stability testing, memory especially, but due to all the rebooting that was going on the past 48 hours trying to figure out what on earth was causing all of this mess I wasn't always closing everything each time.

Font Screenshot Multimedia Electronic device Software


There's one peak with everything closed and switching frequencies maxed out (obviously with a reduced CCD1 curve not to instantly error). Compare 107w to other runs peaks at like 78~80w o_O

Font Screenshot Software Technology Electronic device


This full run for example with no core on CCD1 over 55 degrees, whereas at 90~100w+ getting to 60~66 degrees.

I think I just have to trust you and others that switching frequencies really shouldn't do this, and probably neither should voltages like VDDP/VDDIO and ignore them. I'm still not 100% sure what scalar does, but I can't pin it down as a reason for wattage increases.

So I'm wondering if it's simply a combination of the randomised nature of AVX2 Small FFT testing (although the beginning of this test does seem to be bursty) and apps open on my PC/Windows jank and simply now accepting reduce the hell out of the CCD1 curve.

Because between reboots and settings changes I haven't been able to consistently nail down anything that will only run up to 75~80w when all my system apps are closed :ROFLMAO:

Going to keep testing the rest of the night but leaning towards just accepting my thermal overhead and LLC8 should likely result in more power/voltage usage and therefore an impact on any potential curve and simply ignore runs where power usage has been lower and a higher curve has passed.
 
#6,673 ·
Audioboxer said:
Time to check my sanity more like :ROFLMAO:

I've only had an hour of testing things since getting home but I've discovered more weirdness. Seemingly, if my task tray apps are open, notably discord, there is a chance OCCT won't draw as much power. I normally close task tray apps before all serious stability testing, memory especially, but due to all the rebooting that was going on the past 48 hours trying to figure out what on earth was causing all of this mess I wasn't always closing everything each time.

View attachment 2657203

There's one peak with everything closed and switching frequencies maxed out (obviously with a reduced CCD1 curve not to instantly error). Compare 107w to other runs peaks at like 78~80w o_O

View attachment 2657204

This full run for example with no core on CCD1 over 55 degrees, whereas at 90~100w+ getting to 60~66 degrees.

I think I just have to trust you and others that switching frequencies really shouldn't do this, and probably neither should voltages like VDDP/VDDIO and ignore them. I'm still not 100% sure what scalar does, but I can't pin it down as a reason for wattage increases.

So I'm wondering if it's simply a combination of the randomised nature of AVX2 Small FFT testing (although the beginning of this test does seem to be bursty) and apps open on my PC/Windows jank and simply now accepting reduce the hell out of the CCD1 curve.

Because between reboots and settings changes I haven't been able to consistently nail down anything that will only run up to 75~80w when all my system apps are closed :ROFLMAO:

Going to keep testing the rest of the night but leaning towards just accepting my thermal overhead and LLC8 should likely result in more power/voltage usage and therefore an impact on any potential curve and simply ignore runs where power usage has been lower and a higher curve has passed.
Click to expand...
You keep referring to this as "wattage spikes" while I think you should monitor core frequency in the first place. When your boost frequency is higher, CPU requests higher voltage and therefore wattage is higher
 
#6,674 ·
ZhenyaUsenko said:
You keep referring to this as "wattage spikes" while I think you should monitor core frequency in the first place. When your boost frequency is higher, CPU requests higher voltage and therefore wattage is higher
Click to expand...
I just mean spike as in the highest it goes, not that it's wrong or an error.

I know with more thermal headroom PBO may push harder and that will equal more voltage/wattage.

Only thing I'm trying to understand is the runs in OCCT/Corecycler where it has been lower and therefore a more aggressive curve has passed.

Yes, I'm monitoring effective clocks whilst running tests at the moment (y)
 
#6,676 · (Edited)
eighty20 said:
I was testing VDD = VDDQ = 1.5V since I have seen someone running 8000C34 w that voltage, basically all done mismatch fix stress test done etc.. Then I started playing w nitro and quickly realize 1/3/0 does not work for me current setting. So return to the profile and accidentally set VDDIO = 1.45V instead of one notch down, and this is the lowest idle power I have seen so far on my SR, what do you think?

View attachment 2657201
Click to expand...
I bet my dead Gene you'll get lots of reds if not an entire ocean as soon as you run TM5.đź‘˝
 
6,661 - 6,680 of 9,076 Posts
About this Discussion
9.1K Replies
540 Participants
Last post:
MarkPost
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling