Ryzen Owners Zen Garden

[Mussels]

Honestly if you're after controlling temps etc, just go with a modest all core OC. the 200Mhz single core boost i got vs my 4.3Ghz manual OC wasnt worth the extra 15C in multi threaded (which is almost everything i do, except for old DX9 games)

 

[entropy13]

Got temps down from 92C to 80C by manually setting a 4.2GHz overclock (which is the maximum Boost clock anyway) @ 1.28V. PBO and CPB are disabled. And I'm using the stock cooler lol

Temps were from the in-game benchmark in Grid: Autosport.

EDIT: Looks like I can have it run 5C cooler (from 99C to 94C) on Cinebench at a lower voltage, 1.265V. Cinebench crashes at 1.25V.

OK So I just opted to adjusting PPT. It's at 80W (running at 3.6GHz) and it no longer goes beyond 80C on Cinebench.

In-game temps (Grid: Autosport) never went beyond 73C, so that's good as well. Looks like just changing PPT already did a lot.

 

[pantherx12]

OK So I just opted to adjusting PPT. It's at 80W (running at 3.6GHz) and it no longer goes beyond 80C on Cinebench.

In-game temps (Grid: Autosport) never went beyond 73C, so that's good as well. Looks like just changing PPT already did a lot.

Still allows very high single core power draw just a heads up

@Mussels Problem is there's no wiggle room in this at all. Because I have one very hot core all the time ( 10 c higher than the rest) if I do 3.9 or 3.8ghz which I can do with very reasonable voltage it still gets thermal cut off. This is at the same v-core that gets applied for stock 3.5-3.6ghz speeds.

It can just about manage 3.6ghz . When gaming.
So annoying as when just rendering it can manage 3.9 just about for hours.


I would say it's an air bubble in the paste but I've remounted a million times. Pretty sure it's from uneven mounting pressure due to only 3 bolts holding down cooling.

 

[Mussels]

Still allows very high single core power draw just a heads up

@Mussels Problem is there's no wiggle room in this at all. Because I have one very hot core all the time ( 10 c higher than the rest) if I do 3.9 or 3.8ghz which I can do with very reasonable voltage it still gets thermal cut off. This is at the same v-core that gets applied for stock 3.5-3.6ghz speeds.

It can just about manage 3.6ghz . When gaming.
So annoying as when just rendering it can manage 3.9 just about for hours.


I would say it's an air bubble in the paste but I've remounted a million times. Pretty sure it's from uneven mounting pressure due to only 3 bolts holding down cooling.

you need to fuck that warranty off and do what i did, and sandpaper your fingertips off
i mean, lap your ryzen. yes.

 

[pantherx12]

It's tempting but but I'll leave it as is for the time being just incase the rest of the laptop goes. I could justify loosing warranty on the CPU but as it came as a package it would void my entire warranty .

Question is this level of difference typical for ryzen 3000 2 ccd processors?

If no, I could shave off so much of my max temperatures if I could rectify this issue .

Screen shots still occasionally don't work so apologies for photograph.

 

[Shou Miko]

@Mussels have your lapped your own Ryzen?

 

[pantherx12]

I found out something interesting whilst messing about. The reason one CCD is so hot is because each core draws 20% more power than the " good" CCD!

Really annoying behaviour, anyone know any AMD employees? Give them a kick please and see if this can be corrected through bios updates.

To get them to run at the same wattage I have to run ccd0 at 3ghz and ccd1 at 3.6ghz.


That's just crazy, I know they bin the chips but such a mismatch is pure madness.

 

[Mussels]

@Mussels have your lapped your own Ryzen?

the 2700x is mine
the 3700x runs cold and smooth, so its not as bad as the 27 was

 

[toilet pepper]

the 2700x is mine
the 3700x runs cold and smooth, so its not as bad as the 27 was

I guess my 3600 needs lapping as well. I tried several coolers and pastes and it sill runs around 70ish when gaming. Around 27c ambient here.

 

[AsRock]

With the bios update on my ASRock Pro4.

Got mine on the 2nd and it's lowered the performance by a little all so noticed the bus clock runs 98.8 and not 100% now too on top of that by default SoC/Uncore OC mode ( Higher idle power usage ) was enabled by default ( wasn't before).

As i have mine on all core 4200\1600 HWinfo used to report 4200 but now 4.192.4, but 99% sure that's a bug with HWinfo as RM reports 4.199\4.200.

CPU's running even cooler now, 41c idle, room temp is 33.3c

All so got 5 extra profiles YAY.

 

[tabascosauz]

With the bios update on my ASRock Pro4.

Got mine on the 2nd and it's lowered the performance by a little all so noticed the bus clock runs 98.8 and not 100% now too on top of that by default SoC/Uncore OC mode ( Higher idle power usage ) was enabled by default ( wasn't before).

As i have mine on all core 4200\1600 HWinfo used to report 4200 but now 4.192.4, but 99% sure that's a bug with HWinfo as RM reports 4.199\4.200.

CPU's running even cooler now, 41c idle, room temp is 33.3c

All so got 5 extra profiles YAY.

Is this the new v2 1.0.0.2 BIOS? I'm flashing right now to the 0805 BIOS on my board, which is the second iteration of 1.0.0.2. The 1.10 update was an initial attempt at 1.0.0.2 on my previous ASRock B550 and went horribly, so we'll see if Asus does any better.

I still can't find anyone else who's noticed a 4000MHz cap in CB R20 MT on 1.0.0.0/.1/.2 with B550. ST works just fine. It's way too oddly consistent across 2 different boards and 4 BIOS revisions to be a freak mistake.

Also, 4192MHz at 4200MHz is a 99.8MHz BCLK, completely normal reporting on absolutely every motherboard ever made when Spread Spectrum is on, as should be default settings. You're not losing performance.

 

[AsRock]

Is this the new v2 1.0.0.2 BIOS? I'm flashing right now to the 0805 BIOS on my board, which is the second iteration of 1.0.0.2. The 1.10 update was an initial attempt at 1.0.0.2 on my previous ASRock B550 and went horribly, so we'll see if Asus does any better.

I still can't find anyone else who's noticed a 4000MHz cap in CB R20 MT on 1.0.0.0/.1/.2 with B550. ST works just fine. It's way too oddly consistent across 2 different boards and 4 BIOS revisions to be a freak mistake.

Also, 4192MHz at 4200MHz is a 99.8MHz BCLK, completely normal reporting on absolutely every motherboard ever made when Spread Spectrum is on, as should be default settings. You're not losing performance.

Yes it is v2

Spread Spectrum is default has always been off.

 

[Mussels]

i cant even find spread spectrum on my board, but yeah when on it lowers the clock that tiny fraction

edit: got the new AGESA on my x370 board, too

Auto settings are getting 4GHz all core, 4.25-4.35 single (hell i dont think i even have XMP on)

 

[AsRock]

Yeah well, on or off it makes no difference in HWinfo or in RM, how ever i did notice a 50 point loss in CB.

 

[Shou Miko]

I am looking into tuning my memory better then XMP or AMP since it's on AMD

My 4 sticks which is 2xkits runs 16-17-17-35 T1 @ 3000MHz but I feel this could be better but since it's mixed Hynix M-Die (Hynix MFR) and Samsung B-Die I am not sure which timings to trust the most.

Using DRAM Calulator for Ryzen by @1usmus it doesn't make my decision easier for 3000 or 3200MHz since my Hynix M-Die based kit is only 3000MHz so the calulator won't go over 3200MHz and I need the 32GB for what I am doing and I don't want to give up these kind of memory because you cannot really buy it anymore.

3200MHz Timings for both kits


So a little close together yet still far and the Hynix kit manages 14-17-17-34 T1 and the B-Die 14-20-19-40 T1 I guess the B-Die ain't the best I could have gotten.

 

[biffzinker]

My 4 sticks which is 2xkits runs 16-17-17-35 T1 @ 3000MHz but I feel this could be better but since it's mixed Hynix M-Die (Hynix MFR) and Samsung B-Die I am not sure which timings to trust the most.

I’m mixing 2 kits of 16GB Hynix 3200 MHz with Samsung 2667 MHz for 32GB and managed 3333 MHz at the default 16-18-18-36 T1 timings.

 

[Shou Miko]

I’m mixing 2 kits of 16GB Hynix 3200 MHz with Samsung 2667 MHz for 32GB and managed 3333 MHz at the default 16-18-18-36 T1 timings.

Interesting

 

[Toothless]

Got a buddy to join the red side. 3600 that I was able to stick to 4.2 all core. Slapped my 3700x cooler on there and temps are about 10-15c cooler manually clocked than stock.

1.069v under load, 1.281v idle for core.

Peak 60.5c for the core after five runs of R20.

Gonna slap it in a case and get it finished for the dude.

 

[GoldenX]

That stock cooler looks so good. If only it was a bit better in cooling performance.

 

[tabascosauz]

Stepped the DJR up to 3733 @ 1.38V and so far stable in benches and 50% 32GB HCI. Timings are identical to my 3600 profile save for slightly higher tRFC. I'm not sure if 3800 is within reach; 3733 first for 200% HCI and a few days to make sure the IMC is okay.



My problem is still that I can't find any way to run my kit on this board without Geardown (GDM) at any timings. I've tried my usual custom 16-19-19-37 profile, 18-19-19-39 XMP (because of GDM), 16-19-21-36 DRAM calc recommendations, and XMP -1 tCL 16-19-19-39. None of them can pass HCI Memtest (DRAM calc's Membench also integrates short HCI testing) and will throw up errors almost immediately without GDM. The exact same timings and Vdram with GDM is always 100% stable through HCI. Voltage I've tested all the way up to 1.43V, makes no difference.

GDM is always costing me 0.4ns in AIDA, but more importantly 2-2.5 seconds in Membench. I was at 117.35s 3600 no GDM on my ITX board. It is nice to have the reliable boost to stability, but still.

On my ITX board, all of my profiles were error-free without GDM. I did discover a section in the guide about eliminating GDM through working with procODT and Clkdrv, might try that later. There are also board differences but the lack of info on B550 boards doesn't help; my Aorus ITX was obviously direct-DIMM and 6-layer, while my new B550 mATX is daisy-chain and I have no idea whether it's 4-layer or 6-layer. Asus' B550 mATX boards all share the same RAM validation speeds, and Asus' cheap X570-Prime is actually a 6-layer, but this segment is traditionally 4-layer......but at the same time, I've not had any issues with raw frequency as is usually the issue on 4-layer boards, only GDM.

 

[GoldenX]

I have been fighting with my RAM for months now, too. Samsung C-die is total trash. You can't even do 3266 (+66MHz) without errors after minutes.

 

[Toothless]

Should I be concerned with the sensors going a bit weird? Nothing shuts off or anything but it's the second time it's read these weird things.

CPU TCTL/TDIE and CCD1.

 

[tabascosauz]

Should I be concerned with the sensors going a bit weird? Nothing shuts off or anything but it's the second time it's read these weird things.

CPU TCTL/TDIE and CCD1.

No:

Unless it's happening all the time.

 

[Nordic]

you need to fuck that warranty off and do what i did, and sandpaper your fingertips off
i mean, lap your ryzen. yes.

Given that ryzen performance is so thermally dependent, I was considering lapping my 3900x. I had about a 14c drop on my 2500k when I did it. What results did you get?

 

[biffzinker]

What results did you get?

Found this from @Mussels.

2700x is now at 4.1GHz all core, 1.3125V - 55C load in torture tests, prior to the lapping that thing would always be 80C on that cooler.

Ryzen Owners Zen Garden

Looks like some form of clock stretching? I get higher scores than your 4.3 result just running without PBO @ 4.075-4.15GHz all core normal boost...are you using Ryzen Master for OC? RM does all sorts of weird things to clocks and scores. You have faster ram than i do, and i OC'd in BIOS only...

www.techpowerup.com