[donnieyeen]

MLB enabled, +200FMAX PBO CO-25 CCD0 -30 CCD1 thats it

what is your cb23 score?

 

[NeilBrit]

You have some awesome cores it looks like, and pretty good cooling. i am unable to get such a score on the 5B atm. Not sure if doing something wrong but its around 66-67. But I do have sauna here, and I am very limited by hitting the temp limit quickly.

@chewonthis seems the 5B is not a problem to run on gene with mlb on, but not very good scores honestly.

Busy playing with 8300 I usually tune with CO-20 all core as I know its stable for my CPU. Just did a run with these settings. I'm sure I can better when I tune in CPU

 

[NeilBrit]

what is your cb23 score?

Not very good, 47100/200

 

[donnieyeen]

Not very good, 47100/200

That's pretty strange, isn't 5B alot depended on the cores? Yeah I am really unsure why some test are so good and others not so good on these cpu's. Trying to find some common notes here.

 

[NeilBrit]

That's pretty strange, isn't 5B alot depended on the cores? Yeah I am really unsure why some test are so good and others not so good on these cpu's. Trying to find some common notes here.

5B is dependent on CPU yes. I myself still figuring out this CPU

BTW here's my Aida run reads are 104k and writes 111k and crappy latency??? this run with CCD0 -25 and CCD1 -30

 

[donnieyeen]

5B is dependent on CPU yes. I myself still figuring out this CPU

BTW here's my Aida run reads are 104k and writes 111k and crappy latency??? this run with CCD0 -25 and CCD1 -30

View attachment 2672355

View attachment 2672357

How does your writes show 113 GB on one acreen and then 111k on the other that doesn't seem to match, 109GB roughly translates to the 111k? Seems like Aida bug? It's very strange this Aida on the CPU.

 

[Blameless]

If it's only the coating that interferes and not the caps, will be easier to remove the coating than cut the direct-die frame It comes off easily

This way I was able to use TG direct-die frame V1 with newer batches of 7950X3D that have more coating applied

I want the coating there as there will be liquid metal involved.

Anyway, I just filed some notches in the frame, took about ten minutes and fits flush now. Will probably seal around the edge with more silicone conformal coating to protect the frame itself and keep anything from seeping under it.

 

[NeilBrit]

How does your writes show 113 GB on one acreen and then 111k on the other that doesn't seem to match, 109GB roughly translates to the 111k? Seems like Aida bug? It's very strange this Aida on the CPU.

Def a bug, Aida reads 101 and 104 on other, yeah Def a bug

 

[ZhenyaUsenko]

I want the coating there as there will be liquid metal involved.

Anyway, I just filed some notches in the frame, took about ten minutes and fits flush now. Will probably seal around the edge with more silicone conformal coating to protect the frame itself and keep anything from seeping under it.

I once had the caps shorted through the coating with LM (probably were not fully covered). I don't trust it since then and apply TG Shield additionally on top of the coating.

 

[chewonthis]

I want the coating there as there will be liquid metal involved.

Anyway, I just filed some notches in the frame, took about ten minutes and fits flush now. Will probably seal around the edge with more silicone conformal coating to protect the frame itself and keep anything from seeping under it.

I'm just happy to see someone tested the cpu and gathered what is needed to get the most out of it and then executed on what it needs. It would have stayed lidded and collected dust here except for occasional memory compatibility testing.

 

[donnieyeen]

Here's my 5b best so far ....

My best run so far.

 

[NeilBrit]

Very Good Donnie, that's close to Dom

 

[Blameless]

I once had the caps shorted through the coating with LM (probably were not fully covered). I don't trust it since then and apply TG Shield additionally on top of the coating.

The coating on the 9000 series is much thicker than it was on the 7000s (one of the reasons why it was getting in the way) and it fully covers all potential shorts. I might put another coat on when I put it under water, but for now it's fine with the stock stuff.

I'm just happy to see someone tested the cpu and gathered what is needed to get the most out of it and then executed on what it needs. It would have stayed lidded and collected dust here except for occasional memory compatibility testing.

Have it up and running again. looking at the effects of the delid + LM now. My testbench cooler is a lapped Noctua U12A with a single Thermalright TL-C12R in pull.

With the stock IHS and Thermalright TFX anything past 140w was causing serious issues (hard locks in LINPACK, crashes in BKT, etc). With everything identical, but bare die, using cheap Phobya LM, it looks like at least 160w will be usable. At 140w heat load temps are 10-15C cooler. It's performing nearly at the level of my other samples now.

Moving to a direct die waterblock and hooking it up to my top 240 rad should be enough for 180-200w, I suspect.

On a side note, the stock AMD heatsink retention mechanism is the exact right height for most of my Noctua and Thermalright heatsinks to apply good pressure to bare die CPUs. Just needed to file off a little excess plastic so I could attach the Noctua retention bars to them.

 

[Blameless]

160W post-delid and not even hitting thermal limits on crappy air:

BTW here's my Aida run reads are 104k and writes 111k and crappy latency??? this run with CCD0 -25 and CCD1 -30

View attachment 2672355

Granite Ridge does seem to land a few ns above Raphael at the same settings. Not sure there is much more you can do about latency at 8000.

Have you tried reducing AdRdPtrIntVal and/or tightening up Nitro settings?

Best I can manage, latency wise, on this 9700X:

 

[LtMatt]

160W post-delid and not even hitting thermal limits on crappy air:

Granite Ridge does seem to land a few ns above Raphael at the same settings. Not sure there is much more you can do about latency at 8000.

Have you tried reducing AdRdPtrIntVal and/or tightening up Nitro settings?

Best I can manage, latency wise, on this 9700X:

What is AdRfPtrIntVal?

 

[donnieyeen]

What is AdRfPtrIntVal?

it's the setting inside DDR PMU options.

 

[Bal3Wolf]

any tips to get my 9950x fclk to run higher i seem to be limited to 2000ish maybe less still messing with it but looks like if i wanna run 2000 i gota run 1.3 SOC and 1.45 vddio.

 

[Blameless]

What is AdRfPtrIntVal?

https://www.overclock.net/posts/29358306/

On my ASRock boards it's under AMD Overclocking in the DDR PMU training area (which is a strange place to put it).

any tips to get my 9950x fclk to run higher i seem to be limited to 2000ish maybe less still messing with it but looks like if i wanna run 2000 i gota run 1.3 SOC and 1.45 vddio.

Almost unheard of to see a Granite Ridge CPU that can't do at least 2100MHz FCLK.

SoC and VDDIO also won't do much...only voltage that usually a major impact are the VDDGs (and by extension VDD_MISC).

 

[Bal3Wolf]

https://www.overclock.net/posts/29358306/

On my ASRock boards it's under AMD Overclocking in the DDR PMU training area (which is a strange place to put it).



Almost unheard of to see a Granite Ridge CPU that can't do at least 2100MHz FCLK.

SoC and VDDIO also won't do much...only voltage that usually a major impact are the VDDGs (and by extension VDD_MISC).

tried 1050 on both vddgs and 1.15 on vdd_misc.

 

[Blameless]

tried 1050 on both vddgs and 1.15 on vdd_misc.

1050mV set on VDDG will push 1100mV or a bit more. You want at least 1200mV VDD_MISC for that, if not a bit more.

Regardless, you shouldn't need anywhere near this much voltage and it might be detrimental. Try 1.15 VDD_MISC, 900mV VDDG CCD and 925mV VDDG IOD.

Also verify that it's actually FCLK that's the issue and not the memory settings.