Overclock.net banner
481 - 500 of 2,522 Posts
Double checked the batch on my 9700X when I swapped to my new board, it's a 2429SUY, as was my second sample. Both were from Newegg on launch day.

Haven't seen any reviewers with the SU parts (assembled in the Chinese plant, rather than Malaysia).
 
Discussion starter · #482 ·

Pretty funny, i have been running Win11 24H2 for the last two months.. Maybe thats why my scores are much better than many other Zen5 users :unsure:

The money shot:
Colorfulness Azure Font Parallel Slope


Other test here:
 
Miss my VDH already. This B650M Aorus Elite has some advantages, but it's training times are more than twice as long with the same settings and it's doesn't allow tFAWs below 20, which is costing a full percent of memory performance. On the plus side, it's not broken and unlocks the PMIC.


Pretty funny, i have been running Win11 24H2 for the last two months.. Maybe thats why my scores are much better than many other Zen5 users :unsure:

The money shot:
View attachment 2670109

Other test here:
Seems strange that the OS can even influence how branch prediction functions. I would have though this would be totally transparent to software, unless there is some active mitigation that a new OS build includes a better workaround for (or that Zen 5 no longer needs), or something like that.
 
  • Rep+
Reactions: blu3dragon

Pretty funny, i have been running Win11 24H2 for the last two months.. Maybe thats why my scores are much better than many other Zen5 users :unsure:

The money shot:
View attachment 2670109

Other test here:
 
The Win 11 24H2 is a huge update with main focus on support for AI, MS Copilot+ and Qualcomm Snapdragon X series chipsets, but it has a lot of other new futures and updates too.

The biggest interesting part for us, is the support for CUDIMMS in Widows. But we have to wait and see the offiscial support, these beta software can have bugs.

Do we know what's really affecting the performnace of 9000-series in Win 11 24H2 now?
 
Sorry link are in german...


It means "Sorry Windows-User but with 24H2 we will increase the performance for Ryzen 9."

Great news and it seems domdtxdissar are right.
 

Pretty funny, i have been running Win11 24H2 for the last two months.. Maybe thats why my scores are much better than many other Zen5 users :unsure:

The money shot:
View attachment 2670109

Other test here:
Doesn't seem like any of them ran the 7700x on the new Win11 build though?

I have a hard time understanding this one other than Win11 sucks I guess. Outlier, but still...

Colorfulness Azure Font Slope Parallel
 
Miss my VDH already. This B650M Aorus Elite has some advantages, but it's training times are more than twice as long with the same settings and it's doesn't allow tFAWs below 20, which is costing a full percent of memory performance. On the plus side, it's not broken and unlocks the PMIC.



Seems strange that the OS can even influence how branch prediction functions. I would have though this would be totally transparent to software, unless there is some active mitigation that a new OS build includes a better workaround for (or that Zen 5 no longer needs), or something like that.
same, i had a MSI 2 dimmer and its so bad versus HDV for RAM OC its night and day. Hope your gigabyte is better. tFAW 20 apparently is the registry limit.

im running W11 LTSC IOT 24H2 90 day trial and it its the fastest OS for 9000 series in most benchamrks. Gaming i dont know. Its updated
 
  • Sad
Reactions: OCmember
It's dual ccd you need to push the memory.

Quick ambient tests on a 9900x i may not keep.

static core clocks @ 1.2v to test ccds quality......that looks actually perfect for setting up +200 pbo boost/co tune ;)
View attachment 2670098

6600/2200 super easy.
View attachment 2670099

Then when you give it the bandwidth it wants....
View attachment 2670100
My 9900x won't boot with 6400MHz CL30 memory and SoC voltage at 1,200v

I tested the quality of the silicon and at 1,100v it only does 5125MHz on CCD1 and 5.1GHz on CCD2

Image


If I increase the CCD by 25Mhz more, it gives OCCT errors in the first few seconds.

To test the memories I use MemTestPro, I have managed to pass other Ryzen RamTest without problems and when I put MemTestPro it fails in less than 5 hours

I have also tried to raise the FCLK to 2233Mhz and it has not started again, I have had to reset the bios :ROFLMAO:

And what I like least is that in HWinfo the "CPU (Tctl/Tdie)" sensor is always very high, 48-50 degrees even if I do nothing in the system.

Image


I use this as a heatsink:


Greetings
 
My 9900x won't boot with 6400MHz CL30 memory and SoC voltage at 1,200v

I tested the quality of the silicon and at 1,100v it only does 5125MHz on CCD1 and 5.1GHz on CCD2

Image


If I increase the CCD by 25Mhz more, it gives OCCT errors in the first few seconds.

To test the memories I use MemTestPro, I have managed to pass other Ryzen RamTest without problems and when I put MemTestPro it fails in less than 5 hours

I have also tried to raise the FCLK to 2233Mhz and it has not started again, I have had to reset the bios :ROFLMAO:

And what I like least is that in HWinfo the "CPU (Tctl/Tdie)" sensor is always very high, 48-50 degrees even if I do nothing in the system.

Image


I use this as a heatsink:


Greetings
Well I wouldn't take this CPU that I have as a reference to all 9900x.

It's an alien...I just saw a weird ass batch and said hold my beer I'm gonna try something.

Amber Font Rectangle Material property Tints and shades


When I left this morning for work.

Azure Rectangle Font Parallel Number
 
Discussion starter · #496 · (Edited)
Took some pictures from my Zen5 delid, if your well versed in overclocking these will probably not be so interesting, but maybe they can help some of the new users around here if they are thinking about delidding (y)

How the ES looked right out of the delidder: (7950X with cracked die in the background, more on it later)
Passive circuit component Circuit component Hardware programmer Computer data storage Electronic component


Side by side after i have used the 7950X to scrape the glue off, around the dies
Circuit component Passive circuit component Hardware programmer Electronic component Font


First application of liquid metal to eat the indium
Circuit component Rectangle Electronic component Font Electric blue


Let it eat for ~15min
Circuit component Rectangle Hardware programmer Font Electronic component


How the indium looked after first LM removal
Circuit component Hardware programmer Computer data storage Electronic component Font


New LM application which consume the rest of the indium while i sleep
Passive circuit component Circuit component Hardware programmer Font Microcontroller


~6 hours later
Brown Circuit component Rectangle Font Hardware programmer


Repeated the prosess one more time while i went to work
Passive circuit component Circuit component Computer data storage Hardware programmer Electronic component


Final picture after cleaning with isopropanol
(its real easy to see the cracked IO die on the 7950X here)
Brown Circuit component Hardware programmer Rectangle Font


Only thing left now is to install the naked beauty before the abuse starts :devilish: :ROFLMAO:
 
Something really weird is going on with my sample. I thought it was down to the board that was failing, but even in an entirely different board I'm still not getting PMIC readings and the core ordering doesn't show correctly in HWiNFO or the Event Log. I saw memory sensor information a few times intermittently on my prior board (the HDV), but then it vanished and has never shown up on this Gigabyte board.

Starting to think the CPU itself is damaged.
 
  • Wow
Reactions: OCmember
Thats actually good news because then there is a chance they're gonna fix some of the issues. Looking forward to B2 stepping
If there is any hardware issue, they would say it by now. Because if they know and keep quiet, it will be a big blow to their credibility and people will wait with purchase in future.


I'm still hoping for micro-code, AGESA, BIOS and such issues that can be fixed by software updates. But this uncertainty and rumers are destructive for AMD. They have to make an official statement and clear it up.
 
481 - 500 of 2,522 Posts