Overclock.net banner
5,881 - 5,900 of 9,494 Posts
yesterday my x670e hero acted strange, suddenly got an error 00 and was after that getting memory training stuck on code 15 indefinitely, i than swapped to to my other cpu which booted fine but wanted to do a bios reset and now the motherboard is completely dead, i just hope it didn’t kill my first cpu as it was one that could do flck 2200…

new motherboard ordered for tuesday so fingers crossed my cpu survived
 
why is a chip clocked 6% higher consistently underperforming? an OCed 9950x3d, whose clocks would approach 10%+ a stock 9800x3d, somehow losing? Make it make sense. Is this all due simply to the existence of the second ccd? Which would imply that disabling it, or properly configuring the app via process lasso, would resolve this? Or is it something else...
View attachment 2701346
Parts of it is probably due to higher latency due to the extra ccd. some of it maybe due to wrong scheduling with some games.
 
Yeah, you have been writing a lot about this for a while. Now you know all details about this method, so maybe you and @digitalfrost could work as a team and make an agent, or a template, that people can use it to feed the AI with their own SVI3 TFN? That would take your efforts to a new level.
I have not used ChatGPT or Ai. I will have a look, but I like playing and tinkering myself :) .

Few days ago someone posted a issue with their CO when deviating from matching up voltages. It got me thinking about using Curve Shaper. Then someone asked about PBO CO CS eCLK combo, so that got me thinking about that also.

You can see a preliminary test of PBO CO CS eCLK here, compare my CB23 with static shown here (last 2 screen shots). I also have CB23 runs of PBO CO +200MHz BCLK 100MHz on my R7 9700X, so I know scaling for CB23 be ~25k.

My PBO CO CS +200MHz eCLK 102.15MHz has passed so far ~3.5hrs of AIDA64 CPU FPU CACHE :) .
 
Hello everyone,
First of all, thank you to Overclock.net and everyone who participates in this forum.
It's been over thirty years since I last touched an AMD CPU; the last one was an AMD Athlon cartridge :)

To start, here's my SP for my 9800X3D with Asus Crosshair X870E Hero Bios 1203:

View attachment 2702686


The RAM is a G.Skill Trident Z5 Royal Neo 48GB (2 x 24GB) DDR5 8000 MHz 40-48-48-128 M-DIE kit.

First, I tried the EXPO1, EXPO2, and EXPO Tweaked profiles, but none of them were stable.

Second, I tried configuring my kit from the AMD Overclocking menu and noticed that on ProcODT, I could switch from Hi-Z to 60 Ohms and from Hi-Z to 60 Ohms on ProcODT, and the kit was stable.

I relied on this post and Buildzoid's timings to achieve this with the following BIOS configuration:

Core Tuning => Legacy
SVM Mode => Disabled
Memory Context Restore => Disabled
PowerDown => Disabled
DDR5 Nitro Mode => Disabled
PBO => enable ( default )

For tPHYRDL, it's impossible to get it to sync. On ARdPtrInitVal P0 Control, I tried the entire range from 0 to 15 without success.

Here are my timings, stress test

View attachment 2702687

HWiNFO64
View attachment 2702688


I think RAM @8000 with FCLK @2000 is stable, but on Karhu the speeds seem a bit low, any idea to improve the timings please?

Thank you all again.
Kahru will gain test speed with FCLK, so bump that. Setting tRCDWR about 20 or so, Kahru will gain test speed. Tighter tRCDWR may pass RAM tests, but may show performance loss in benchmarks vs a little looser.

Stability test with Core Tuning Auto, as then advanced prefetchers and cache retention polices are on. Legacy use to get benchmarks. You'll find some info and links in section DDR5 Tuning Notes & Observations of this post.

Also try the other profiles of TM5, each profile test RAM differently and uses CPU differently. You'll see from CPU clocks, temps and DRAM read/write bandwidth in HWINFO. Double click a sensor and a graph appears, sometimes it's better representation to look at of data.
 
Do you wanna use CPU VDDCR_VDD Voltage (SVI3 TFN) or Vcore? Latter one is what's measured on cores, if I understand it correctlty
Usually VCORE reading is from VRM side. Each board how it shows VCORE in HWINFO may differ, due to board implementation. CPU VDDCR_VDD Voltage (SVI3 TFN) won't differ on implementation, as it's being read from CPU, it's own telemetry. You will find this information in OCN HWINFO thread and HWINFO forum by author.
 
I have been dialing in a curve shaper negative offset to see if I can squeeze anything else from the CPU. Managed to hit a CB R23 score of 46,730. Running y-cruncher BKT and VT3 for four hours to make sure it is stable.
 
yesterday my x670e hero acted strange, suddenly got an error 00 and was after that getting memory training stuck on code 15 indefinitely, i than swapped to to my other cpu which booted fine but wanted to do a bios reset and now the motherboard is completely dead, i just hope it didn’t kill my first cpu as it was one that could do flck 2200…

new motherboard ordered for tuesday so fingers crossed my cpu survived
Out of interest, which motherboard BIOS was in use when issue happened?
 
Hi there.
I have a 7950X3D with a 4x16GB kit with SK Hinyx B-DIE.
I can easily do 6000mt/s C30 with that 4x16GB kit on my 7950X3D,
is it possible that my upcoming 9950X3D will not be able to handle 6000mt/s C30 on a 4x16GB kit?

Is 6000mt/s still difficult even for zen5?
 
Anyone else finding the 9950x3d is a challenge for stability testing? I used to use occt core cycler and it would quickly fire out WHEAs on unstable cores; but now I'm getting pretty deep CO without any such errors, seems to bench well but finding idle system freezes... Like step away from the system for several hours and come back to a frozen desktop (not an idle reboot). Tried other tests like y-cruncher, but the only thing I can "fail" is the AIDA system stability test. Lowering CO seems to help it pass the test but not getting meaningful feedback as to which cores are failing.

Maybe its memory?

I had a bad first chip that would fail to post maybe 1/4 of the time and gave some weird issues, the second chip doesn't do any of that but I've never had something sail through OCCT tests and still be quite unstable.
 
Anyone else finding the 9950x3d is a challenge for stability testing? I used to use occt core cycler and it would quickly fire out WHEAs on unstable cores; but now I'm getting pretty deep CO without any such errors, seems to bench well but finding idle system freezes... Like step away from the system for several hours and come back to a frozen desktop (not an idle reboot). Tried other tests like y-cruncher, but the only thing I can "fail" is the AIDA system stability test. Lowering CO seems to help it pass the test but not getting meaningful feedback as to which cores are failing.

Maybe its memory?

I had a bad first chip that would fail to post maybe 1/4 of the time and gave some weird issues, the second chip doesn't do any of that but I've never had something sail through OCCT tests and still be quite unstable.
I keep seeing crazy high CO numbers, 40+ and people are claiming it is fully stable. With the previous 7950x3d I had to dial some cores back to as low as 2. Either these CPU’s are far better or there is something not right.
 
5,881 - 5,900 of 9,494 Posts