Question Zen 4 builders thread

Page 57 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Justinus

Diamond Member
Oct 10, 2005
3,167
1,509
136
My 7950X reports about 25-30C when I'm simply watching a video, for example. NH-D15.
It's winter here so the ambient might be a bit low.

What thermal figure is this?

My stock 7950X idles at 35-40C (CCD temps are 25-30C) on an Optimus AM4 block.

From HWINFO:
1674757228034.png

Granted my idle package power is ~25W with ram set to 6200/3100UCLK and 2067 FCLK, all auto voltages.
 

gdansk

Golden Member
Feb 8, 2011
1,992
2,364
136
What thermal figure is this?

My stock 7950X idles at 35-40C (CCD temps are 25-30C) on an Optimus AM4 block.

From HWINFO:
View attachment 75374

Granted my idle package power is ~25W with ram set to 6200/3100UCLK and 2067 FCLK, all auto voltages.
1674758979530.png
CPU Die average. It went lower earlier but I only opened HWInfo again for the screenshot. 30-35C seems to be a more accurate statement. Package power is reported around 20W when idle. So I think it's mainly the difference in our ambient temperatures.
 
  • Like
Reactions: lightmanek

In2Photos

Golden Member
Mar 21, 2007
1,606
1,640
136
My 7700 runs around 43C. The screenshot below is a few minutes after some CB R23 runs so I don't think the AIO fluid has cooled all the way back down. But this thing idles at 30W. When I ran a 30 min CB R23 Multi-Core it peaked at 66C and 90W and just under 5GHZ all core. Just weird to see such high idle temps and power usage.

Idle temps.png
 
  • Like
Reactions: lightmanek

Det0x

Golden Member
Sep 11, 2014
1,027
2,953
136
On the last day of having m-die installed, 6500MT/s GDM disabled completed! :cool:

7950x 16/32 threads @ large custom cooling
Hynix m-die @ 6500MT/s CL28 real T1 (GDM disable)
MicrobenchmarksGui Large pages = 60.73ns memory latency
Aida64 Extreme = 50.8ns memory latency

10k+ karhu @ 299.84mb/s
25 cycles testmem5 1usmus cfg
2 full y-cruncher cycles all tests enabled
6500 GDM OFF-12.png
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
How do you change gear down mode? I did not know you could change that.

HWiNFO64‘s Temp report is a little weird indeed because it reports the package temperature higher than individual cores’. As long as it does not report wildly unexpected temperature I think it should be taken as an approximation.

And I am also done with Samsung. Crappy phones, crappy TVs, deceptive practice with SSDs, I’ve had enough.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,485
14,437
136
How do you change gear down mode? I did not know you could change that.

HWiNFO64‘s Temp report is a little weird indeed because it reports the package temperature higher than individual cores’. As long as it does not report wildly unexpected temperature I think it should be taken as an approximation.

And I am also done with Samsung. Crappy phones, crappy TVs, deceptive practice with SSDs, I’ve had enough.
I am not sure you can change gear down mode. If you select too high of a speed it will change it for you.

As far as temps, I would use Ryzen master when in doubt
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
Oh BTW I found out what proggy was affecting the boost frequency. It was Windows Sandbox. I first suspected Hyper-V, but with Hyper-V alone the max boost is the same. I uninstalled the Sandbox during troubleshooting something else and realized my old boost frequency was back.
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
No, really it’s immaculate. I think I barely make out the cable(s) from water block going into the board‘s fan connectors. Is that right?

and where is the power supply?!

lol I am in awe seeing 9 fans (RGB!) running without a trace of wires. It’s like magic!
 

In2Photos

Golden Member
Mar 21, 2007
1,606
1,640
136
No, really it’s immaculate. I think I barely make out the cable(s) from water block going into the board‘s fan connectors. Is that right?

and where is the power supply?!

lol I am in awe seeing 9 fans (RGB!) running without a trace of wires. It’s like magic!
I was extremely happy with how the cables turned out! The 2 wires on top of the motherboard are the AIO pump and RGB connector. I wrapped as much as I could around the pump so only the bare minimum was visible. All other motherboard connectors attach to the bottom behind the GPU.

The Lian Li uni fans latch onto each other so there's only one cable from each set of fans. I oriented each set to hide the connector.

The PSU in this case is behind the motherboard.

Here's the back of the case with and without the SSD drive bracket.

PXL_20230126_135835068.MP~2.jpg
PXL_20230126_134032269.MP~2.jpg
 
  • Like
Reactions: lightmanek

Det0x

Golden Member
Sep 11, 2014
1,027
2,953
136
How do you change gear down mode? I did not know you could change that.
Depend on what motherboard vendor you have.. On Asus the option is called "ADDR_CMD_MODE -> Buf"
Friendly advice, make sure your timings/voltages are really stable before you try this, if you want to avoid having to reset bios multiple times :)
(GDM is autocorrecting alot of settings in the background, so i recommend to keep it on if your not prepared to spend hours in the bios)
On the last day of having m-die installed, 6500MT/s GDM disabled completed! :cool:

7950x 16/32 threads @ large custom cooling
Hynix m-die @ 6500MT/s CL28 real T1 (GDM disable)
MicrobenchmarksGui Large pages = 60.73ns memory latency
Aida64 Extreme = 50.8ns memory latency

10k+ karhu @ 299.84mb/s
25 cycles testmem5 1usmus cfg
2 full y-cruncher cycles all tests enabled
View attachment 75385
After playing around with some new a-die memory for about 24 hours:
1674996265425.png
More screenshots can be found here:
 
Last edited:

SK10H

Member
Jun 18, 2015
113
42
101
I am testing a 7900x on X670E board and want to try the cppc preferred cores option but notice that it's missing on the AM5 platform. I am using an Asrock board but notice Asus does not have this option in the BIOS as well. On AM4, we can find the cppc preferred cores option in BIOS-> Advanced -> AMD CBS -> NBIO Common Options -> SMU Common Options to enable/disable it.
We can verify it's enabled when going to Event Viewer->Windows Logs->System->Filter using Event ID 55 Kernel-Processor-Power.
On my AM4, if cppc preferred core is enabled, it shows Performance state type: ACPI Collaborative Processor Performance Control
If it's disabled, it shows Performance state type: ACPI Performance (P) / Throttle (T) States
Opinion and experience varies from the discussion from a small sample but there does not seem to be much discussion on this for Ryzen 7000. I tried to set the Power option->processor min/max processor state to 100% and no changes from the event viewer log.

Ryzen Master marked my ccd0 core 4 as the best core, probably align with what Windows scheduler choose as well these days. But my ccd0 is very bad, every core can only do -5 while core 4 is the worst at -0 base on a few hours of corecycler p95 small AVX result.
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
Depend on what motherboard vendor you have.. On Asus the option is called "ADDR_CMD_MODE -> Buf"
Friendly advice, make sure your timings/voltages are really stable before you try this, if you want to avoid having to reset bios multiple times :)
(GDM is autocorrecting alot of settings in the background, so i recommend to keep it on if your not prepared to spend hours in the bios)

After playing around with some new a-die memory for about 24 hours:
View attachment 75530
More screenshots can be found here:
Thx I can't wait to try out!
 

SK10H

Member
Jun 18, 2015
113
42
101
I am testing a 7900x on X670E board and want to try the cppc preferred cores option but notice that it's missing on the AM5 platform. I am using an Asrock board but notice Asus does not have this option in the BIOS as well. On AM4, we can find the cppc preferred cores option in BIOS-> Advanced -> AMD CBS -> NBIO Common Options -> SMU Common Options to enable/disable it.
We can verify it's enabled when going to Event Viewer->Windows Logs->System->Filter using Event ID 55 Kernel-Processor-Power.
On my AM4, if cppc preferred core is enabled, it shows Performance state type: ACPI Collaborative Processor Performance Control
If it's disabled, it shows Performance state type: ACPI Performance (P) / Throttle (T) States
Opinion and experience varies from the discussion from a small sample but there does not seem to be much discussion on this for Ryzen 7000. I tried to set the Power option->processor min/max processor state to 100% and no changes from the event viewer log.

Ryzen Master marked my ccd0 core 4 as the best core, probably align with what Windows scheduler choose as well these days. But my ccd0 is very bad, every core can only do -5 while core 4 is the worst at -0 base on a few hours of corecycler p95 small AVX result.

It looks like I can disable PSS Support in BIOS impacting C States as a workaround. Idle power consumption doesn't seem to be impacted that much at quick glance.
cppc.png

On a 7900x and 7950x, AMD clearly know they put in 1 good 1 bad ccd on it. As the screenshot shows, my hardware-fused order is ccd 0. My ccd 0 barely make the cut at 0 and -5 in CO, ccd 1 can probably avg around -20-25, still only testing p95 small avx and may need to up the curve a bit for full stability later.
So why set my bad ccd0 as the optimal cores to use? Or is it a simply way to always set ccd0 as the optimal one regardless of quality?
 

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
I'm interested to see what memory presets ASUS comes out with for their motherboards in the future.

I saw my board had a Hynix SR (2x16GB) DDR5-6200 preset so I loaded it... and it seems okay? Not setting world records at 59ns latency per AIDA64, but considering it's a "free" upgrade I'll take it. Still checking stability but good so far. There was also a 6400 preset but that actually had worse latency so I assume it set UCLK out of sync or something.
 

Justinus

Diamond Member
Oct 10, 2005
3,167
1,509
136
I just upgraded from a 2x16GB SR Hynix M die kit, the Trident Z5 Neo 6000CL30 kit to a 2x32GB DR Hynix A die kit, the Trident Z5 Neo 6000CL30 kit.

Both at identical settings except a couple timings slightly different (30-38-38-96 for M die, 30-40-40-96 for A die), all other timings identical. The results are slightly perplexing. I only ran aida and maxxmem, but take a look.

SR M die kit:
1675187014954.png1675187030737.png

DR A die kit:
1675187051729.png1675187073481.png

Aida bandwidth went down quite a bit, but latency is better. Maxxmem bandwidth is identical, but latency is again better.
 
  • Like
Reactions: igor_kavinski

SK10H

Member
Jun 18, 2015
113
42
101
Using just a bare motherboard with the cpu and RAM, no GPU on open bench. Look like
1) I got 2 bad 7900x in a row - really unlucky but hard to believe since the cpu are not the same batch
2) My 4 x 16GB Ram has issue, testing 650w psu is bad, motherboard is bad.
3) some 7900x cannot do p95 small avx at stock speed at stock volt.

To rule out #2, I was able to test 2 instances of y-cruncher at 30GB each for 10 hrs without issue at the oc timings.
34.JPG

In addition, p95 small/smallest fft is cpu bound anyway and I am testing 1 thread only with default EXPOS/3600MHz Ram speed.

1st 7900x, week 2233. On this one, cpu core0 was able to do -30 for 1/2/4/6/8 combined threads on p95 small SSE for hours at various speed, but fail 1T avx/avx2 right away. So I don't believe in corecycler comment about SSE default being the better option to test. CCD1 however was able to pass avx2 around -20 if I skip all the ccd0 cores.
defective.PNG

2nd 7900x, week 2239. This time, I even bump the vsoc to 1.4v from 1.35v, VDDCR_CPU load-line to level 1 which is the highest, in addition to the other usual settings vdd misc 1.3v, vddg 1.2/1.2, vddp 1.15, dram vpp 1.9. ccd0 and ccd1 all cores fail at default settings.
defective1.JPG

So I gave up on 7900x and refund, waiting to see how's the 3D version turn out. Lost quite a bit of confidence on the stability of these latest AMD CPU.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,485
14,437
136
Using just a bare motherboard with the cpu and RAM, no GPU on open bench. Look like
1) I got 2 bad 7900x in a row - really unlucky but hard to believe since the cpu are not the same batch
2) My 4 x 16GB Ram has issue, testing 650w psu is bad, motherboard is bad.
3) some 7900x cannot do p95 small avx at stock speed at stock volt.

To rule out #2, I was able to test 2 instances of y-cruncher at 30GB each for 10 hrs without issue at the oc timings.
View attachment 75703

In addition, p95 small/smallest fft is cpu bound anyway and I am testing 1 thread only with default EXPOS/3600MHz Ram speed.

1st 7900x, week 2233. On this one, cpu core0 was able to do -30 for 1/2/4/6/8 combined threads on p95 small SSE for hours at various speed, but fail 1T avx/avx2 right away. So I don't believe in corecycler comment about SSE default being the better option to test. CCD1 however was able to pass avx2 around -20 if I skip all the ccd0 cores.
View attachment 75705

2nd 7900x, week 2239. This time, I even bump the vsoc to 1.4v from 1.35v, VDDCR_CPU load-line to level 1 which is the highest, in addition to the other usual settings vdd misc 1.3v, vddg 1.2/1.2, vddp 1.15, dram vpp 1.9. ccd0 and ccd1 all cores fail at default settings.
View attachment 75706

So I gave up on 7900x and refund, waiting to see how's the 3D version turn out. Lost quite a bit of confidence on the stability of these latest AMD CPU.
Questions... Was the memory on the QVL ? Was it EXPO memory ? Was it at stock ? Was the bios ever at stock ? -30 is very hard to achieve.

From what I have seen, is that the correct memory at the correct settings is the biggest fail in general.
 

SK10H

Member
Jun 18, 2015
113
42
101
Questions... Was the memory on the QVL ? Was it EXPO memory ? Was it at stock ? Was the bios ever at stock ? -30 is very hard to achieve.

From what I have seen, is that the correct memory at the correct settings is the biggest fail in general.

It's GSkill Flare X5 EXPOS memory. When I test CPU 1T in my latest test, I load BIOS default back to 3600MHz Ram speed from the EXPOS 6000, auto fclk etc just to rule things out. :confused: Then also try bumping oc voltage back while maintaining these default speed.
 

AdamK47

Lifer
Oct 9, 1999
15,171
2,830
126
Using just a bare motherboard with the cpu and RAM, no GPU on open bench. Look like
1) I got 2 bad 7900x in a row - really unlucky but hard to believe since the cpu are not the same batch
2) My 4 x 16GB Ram has issue, testing 650w psu is bad, motherboard is bad.
3) some 7900x cannot do p95 small avx at stock speed at stock volt.

To rule out #2, I was able to test 2 instances of y-cruncher at 30GB each for 10 hrs without issue at the oc timings.
View attachment 75703

In addition, p95 small/smallest fft is cpu bound anyway and I am testing 1 thread only with default EXPOS/3600MHz Ram speed.

1st 7900x, week 2233. On this one, cpu core0 was able to do -30 for 1/2/4/6/8 combined threads on p95 small SSE for hours at various speed, but fail 1T avx/avx2 right away. So I don't believe in corecycler comment about SSE default being the better option to test. CCD1 however was able to pass avx2 around -20 if I skip all the ccd0 cores.
View attachment 75705

2nd 7900x, week 2239. This time, I even bump the vsoc to 1.4v from 1.35v, VDDCR_CPU load-line to level 1 which is the highest, in addition to the other usual settings vdd misc 1.3v, vddg 1.2/1.2, vddp 1.15, dram vpp 1.9. ccd0 and ccd1 all cores fail at default settings.
View attachment 75706

So I gave up on 7900x and refund, waiting to see how's the 3D version turn out. Lost quite a bit of confidence on the stability of these latest AMD CPU.
It doesn't seem like your first 7900X was bad if it was only the negative CO offset causing it to fail. SSE is a good test since it will allow higher boosts. YCruncher AVX512 is another good test as it more strenuous per core. It won't find the unstable edge cases of higher boosts an SSE test performs.
 

SK10H

Member
Jun 18, 2015
113
42
101
It doesn't seem like your first 7900X was bad if it was only the negative CO offset causing it to fail. SSE is a good test since it will allow higher boosts. YCruncher AVX512 is another good test as it more strenuous per core. It won't find the unstable edge cases of higher boosts an SSE test performs.

I believe the 1st 7900x is kinda bad as well, I need positive CO for avx2 to pass, my guess is probably about +5 or +10 would do. CCD0 core 0 was the best for p95 SSE test but fail right away at AVX.
CoreCycler p95 small sse
No error meaning like a quick 20 mins loop to several hrs depending if I am around to change the setting.
cpu0 - no error at -25, -30 good for 1/2/4/6/8 threads p95 small but fail avx 1min
cpu1 - error at -25, try -20
cpu2 - no error at -20, try -25
cpu3 - no error at -20, -25 fail, back to -20
cpu4 - error at -20, try -15
cpu5 - error at -20, -15, stable -10?

CoreCycler p95 small avx2
cpu0 - error at -5, 0, try +5
cpu1 - error at -5, 0, try +5
cpu2 - error at -5, 0, try +5
cpu3 - error at -5, 0, try +5
cpu4 - error at +5, try +10
cpu5 - error at 0, try +5
Then I stop testing so don't know how far I would have to go.