What is your max infinity fabric on ryzen 5000? September 2021 edition

[Dar1usElite]

Yeah, but sometimes thaiphoon is inaccurate. It's usually right, but I've seen it miss a few times.

Do I just check the sticker on the dimm itself as advised by Tabasco?

@tabascosauz

 

[Cowboystrekk]

Do I just check the sticker on the dimm itself as advised by Tabasco?

@tabascosauz View attachment 224903

You have CJR. 8820C is CJR. https://www.techpowerup.com/forums/threads/about-buying-ram-thats-not-listed-in-qvl.283727/

This probably explains the problems with tCL. My recommendations for tRFC is then try 496, 480 or 464.

 

[Dar1usElite]

You have CJR. 8820C is CJR. https://www.techpowerup.com/forums/threads/about-buying-ram-thats-not-listed-in-qvl.283727/

This probably explains the problems with tCL. My recommendations for tRFC is then try 496, 480 or 464.

I'll try that after. I did what you said with auto on everything and 18 22 22 22 40 and it's 6 runs in @ 1.4v and no errors yet

TRFC of like 666 on auto though
TFAW 40, tRC 89, trrds/l 8/11.. I'll post a zen timings screenshot if it passes.

 

[Valantar]

I'm starting to zero in on my final RAM settings now. Finished 20 runs of TM5 1usmus with no errors.

This is at 1.46V now, after previously running 1.48. (I seem to remember 1.45 not being stable at these or slightly looser settings, though there have been so many changes over the past weeks it's hard to keep track!) With the fan I put over the RAM and VRMs chugging along inaudibly the RAM barely passes 40 degrees under load. My AIDA64 latency readings are creeping down towards the 60ns mark as well, pretty happy with that (currently running way faster than my not-quite-stable 4000 settings.

Is it worth trying to reduce tRFC any more at 1.46V, or should I leave it at this?

 

[Cowboystrekk]

I'm starting to zero in on my final RAM settings now. Finished 20 runs of TM5 1usmus with no errors.
View attachment 224916
This is at 1.46V now, after previously running 1.48. (I seem to remember 1.45 not being stable at these or slightly looser settings, though there have been so many changes over the past weeks it's hard to keep track!) With the fan I put over the RAM and VRMs chugging along inaudibly the RAM barely passes 40 degrees under load. My AIDA64 latency readings are creeping down towards the 60ns mark as well, pretty happy with that (currently running way faster than my not-quite-stable 4000 settings.

Is it worth trying to reduce tRFC any more at 1.46V, or should I leave it at this?

You can't go much lower. Next step is 256, that needs more voltage. For 256 you probably need 1.48-1.5V. For 240 you need around 1.55V.

I'll try that after. I did what you said with auto on everything and 18 22 22 22 40 and it's 6 runs in @ 1.4v and no errors yet

TRFC of like 666 on auto though
TFAW 40, tRC 89, trrds/l 8/11.. I'll post a zen timings screenshot if it passes.

If it works then start lowering timings one by one, prioritize tRFC, tRC, tRRDS/L and tFAW in pair, tWR and tRTP in pair.

 

[Valantar]

You can't go much lower. Next step is 256, that needs more voltage. For 256 you probably need 1.48-1.5V. For 240 you need around 1.55V.

Sounds like a good place to leave things then Next up: Curve Optimizer testing!

Edit: well that crashed fast. I don't think my CPU likes my +100 PBO offset at -30 CO, as it reset almost immediately after booting. Can't find anything to identify the core responsible in the Event Log either. IIRC I wasn't running any PBO offset when I tested -30 previously. Guess I'll have to start a bit more conservatively

Edit2: So, am I right in thinking WHEA-Logger error 18 "Processor APIC ID" identifies the core that had an error? If so, it looks like my "best" cores are the least tolerant of undervolting. Kind of makes sense in a way.

 

[Cowboystrekk]

Sounds like a good place to leave things then Next up: Curve Optimizer testing!

Edit: well that crashed fast. I don't think my CPU likes my +100 PBO offset at -30 CO, as it reset almost immediately after booting. Can't find anything to identify the core responsible in the Event Log either. IIRC I wasn't running any PBO offset when I tested -30 previously. Guess I'll have to start a bit more conservatively

Edit2: So, am I right in thinking WHEA-Logger error 18 "Processor APIC ID" identifies the core that had an error? If so, it looks like my "best" cores are the least tolerant of undervolting. Kind of makes sense in a way.

Use corecycler to find which cores can`t do -30. Remember that dropping + on PBO will allow you a better negative value which increases allcore speed. On my setup at +200 only core 0 and 1 must have negative of -29, all others run -30, core 0 is the "best" core, so much for that. The cores that show up in event viewer as no 18 needs a lower CO-value.

 

[Valantar]

Use corecycler to find which cores can`t do -30. Remember that dropping + on PBO will allow you a better negative value which increases allcore speed. On my setup at +200 only core 0 and 1 must have negative of -29, all others run -30, core 0 is the "best" core, so much for that. The cores that show up in event viewer as no 18 needs a lower CO-value.

Yep, that's what I've gathered. Core 0 isn't stable even at -15, but is at -10 (hard crashes at -15 and higher). Core 1 has given me one error running Corecycler at -20, but no crashes yet. The rest seem good (so far) at -20. Next steps, in order: tune in core 1 (I'm thinking +1 at a time, easily checked as it only gives errors under load so far), test cores 2-7 at -25 and - 30, then pinpoint the threshold for core 0 between -15 and -10. After that I might try more than +100 PBO offset, but we'll see. This will take a while!

 

[Cowboystrekk]

Yep, that's what I've gathered. Core 0 isn't stable even at -15, but is at -10 (hard crashes at -15 and higher). Core 1 has given me one error running Corecycler at -20, but no crashes yet. The rest seem good (so far) at -20. Next steps, in order: tune in core 1 (I'm thinking +1 at a time, easily checked as it only gives errors under load so far), test cores 2-7 at -25 and - 30, then pinpoint the threshold for core 0 between -15 and -10. After that I might try more than +100 PBO offset, but we'll see. This will take a while!

You must decide if you want 200 pbo now, you might need to reconfigure all cores if you set 100 first and 200 later.

 

[Valantar]

You must decide if you want 200 pbo now, you might need to reconfigure all cores if you set 100 first and 200 later.

I haven't been able to boot reliably above +100 at -10 all core, so that likely tells me that my troublesome cores are starting to reach their limit (or they might need 0 or even a positive offset, which... Nah.). I'm fine with 4.95GHz really, so boosting MT through improving CO is more important to me.

 

[Deleted member 215115]

I'm starting to zero in on my final RAM settings now. Finished 20 runs of TM5 1usmus with no errors.
View attachment 224916
This is at 1.46V now, after previously running 1.48. (I seem to remember 1.45 not being stable at these or slightly looser settings, though there have been so many changes over the past weeks it's hard to keep track!) With the fan I put over the RAM and VRMs chugging along inaudibly the RAM barely passes 40 degrees under load. My AIDA64 latency readings are creeping down towards the 60ns mark as well, pretty happy with that (currently running way faster than my not-quite-stable 4000 settings.

Is it worth trying to reduce tRFC any more at 1.46V, or should I leave it at this?

Looks good. The only thing you could try is tRDWR 8 + tWRRD 1.

 

[Dar1usElite]

You can't go much lower. Next step is 256, that needs more voltage. For 256 you probably need 1.48-1.5V. For 240 you need around 1.55V.


If it works then start lowering timings one by one, prioritize tRFC, tRC, tRRDS/L and tFAW in pair, tWR and tRTP in pair.

 

[Cowboystrekk]

View attachment 224974

Try tweaking timings as I suggested, a few rare errors don't matter if it doesn't happend in games etc Use 2T and gdm off, try CL17 and work your way down the timings.

 

[freeagent]

Looks like it can use more SOC and IOD. But I am used to a dual CCD chip now. Pretty sure I used more than 1.1 on my 5600 for that speed though.

 

[Cowboystrekk]

Looks like it can use more SOC and IOD. But I am used to a dual CCD chip now. Pretty sure I used more than 1.1 on my 5600 for that speed though.

On 2000fclk I need 1.11v soc and 1.03v iod, on 1900 I need 1.08v soc and 1.00v iod, slightly below works, but performance drops. Tested on a friends 5600X and it behaves the same. It seems like dual rank or 4 dimms sometimes need more soc/iod. We are both on SR. Tuned Samsung B-die might require a bit more than others since they stress memory controller more than other ICs due to lower tRCDRD, tRC and tRFC.

 

[freeagent]

On 2000fclk I need 1.11v soc and 1.03v iod, on 1900 I need 1.08v soc and 1.00v iod, slightly below works, but performance drops.

But that's on your chip..for 2K mine needs 1.156 and I use LLC3 because it does not hold steady without it. He is only at 1900, so 1.125v is not a big deal.

uned Samsung B-die might require a bit more than others since they stress memory controller more than other ICs due to lower tRCDRD, tRC and tRFC

That could be too.. all I have used is B-Die unfortunately.

 

[Dar1usElite]

Try tweaking timings as I suggested, a few rare errors don't matter if it doesn't happend in games etc Use 2T and gdm off, try CL17 and work your way down the timings.

ok this is weird. It passed 20 runs with no errors on these timings.. I think i set the primaries, the trc and trfc and 1t with gdm enabled and the rest is on auto @ 1.4v.

 

[Cowboystrekk]

View attachment 225001
ok this is weird. It passed 20 runs with no errors on these timings.. I think i set the primaries, the trc and trfc and 1t with gdm enabled and the rest is on auto @ 1.4v.

Ok, that is good. Try trrd/faw next, 6/8 and 24 should work, 5/7 and 20 might work and 4/6 and 16 is best but a bit harder, wr/rtp after. Wr should be = cl, rtp=half wr. Wtr 4/10-12 and scl's at 4 last.

But that's on your chip..for 2K mine needs 1.156 and I use LLC3 because it does not hold steady without it. He is only at 1900, so 1.125v is not a big deal.

That could be too.. all I have used is B-Die unfortunately.

There is binning on the IOD aswell, one might need more volt than another My friends gets whea at 1933 or higher, mine does 2066 fine without wheas, at 1900 we need same voltage.

 

[Dar1usElite]

Ok, that is good. Try trrd/faw next, 6/8 and 24 should work, 5/7 and 20 might work and 4/6 and 16 is best but a bit harder, wr/rtp after. Wr should be = cl, rtp=half wr. Wtr 4/10-12 and scl's at 4 last.

Alright so try trrds = 6, trrdl = 8 and tfaw = 24? Im genuinely suprised it passed with 16 20 20 20 36 56 to be honest... i dont know if i can get the primaries much tighter than that. Ill try tightening the trfc down to one of the values you said above and then ill try tweaking the sub timings once i find the limit on the trfc

 

[Deleted member 215115]

all I have used is B-Die unfortunately.

More like fortunately. There's B-die and then there's "other memory".

 

[freeagent]

More like fortunately. There's B-die and then there's "other memory".

But!

In a blind taste test without staring at numbers I doubt anyone would notice a difference

Because full default feels fine.. 2133 and a bunch of 20s

Of course everything says it sucks, but to me it felt fine..

 

[Valantar]

But!

In a blind taste test without staring at numbers I doubt anyone would notice a difference

Because full default feels fine.. 2133 and a bunch of 20s

Of course everything says it sucks, but to me it felt fine..

Well, I mean, few of us ever get to do a side-by-side comparison of anything, and it's not like running crap memory gives you 10% lows in the 20s. A lot of spec tweaking performance increases are mainly a process of convincing ourself that our setups are now working better. It might be noticeable at times, but typically not.

 

[freeagent]

Well, I mean, few of us ever get to do a side-by-side comparison of anything, and it's not like running crap memory gives you 10% lows in the 20s. A lot of spec tweaking performance increases are mainly a process of convincing ourself that our setups are now working better. It might be noticeable at times, but typically not.

I think side by side you would notice, but I don't think anyone would intentionally run that slow unless there was instability.. or they just didn't know.. Thinking back to my first boot on a 3600XT with this board and ram at full default, I spent about an hour like that.. did my windows install and updates, poked around the toobz a bit and it felt fine.. compared to a 3770K

First big step was XMP lol

 

[Ibizadr]

It's normal past 3866 got some whea errors?

 

[Valantar]

It's normal past 3866 got some whea errors?

Pretty normal, yes.