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

[cellar door]

My kit is also the 3000cl15 Usually the avg rev E can do 300ns, mine does 280. You can probably do anywhere from 530 to 580 depending on bin. Do you have agesa 1.2.0.3b? Before that I was stuck at 3733 getting random reboots, after 3800 worked fine.


Talking to cellar door or me? He's on ryzen 3000 which rarely can do more than 3800. I can do 4133, I might be able to do 4200+, but it would require way to much soc voltage for my comfort. I need 1.12V for 4000, 1.18V for 4066 and 1.25V for 4133. I bet 4200 would need atleast 1.32V. 1.2V is by many considered upper safe limit.

My MB is cheap, but it has only 2 dimms (helps a lot with ram oc compared to 4) and is rated for 5100MHz ram on Ryzen 3000, max 4600 on Ryzen 5000 which makes little sense...

AAs for ram oc I'm limited by bad bin ram, it has a 94% rating on dram calc, making it sub avg compared to other B-die. 1T and liw tRCDRD is very hard, trfc won't go below 141ns at 1.46V. The rev E I had was 102% rated so it beat the majority of rev E.

Where you are you reading that ram rating in dram calc? I'm using Ryzen calc 1.7.3v

 

[Taraquin]

Where you are you reading that ram rating in dram calc? I'm using Ryzen calc 1.7.3v

Import profile made from thaipoon burner using manual, then there is a qualityrating somewhere.

 

[cellar door]

Thanks, I forgot about that - thought it was a gimmick back when I tried it a year ago. 101% for my kit.

 

[Taraquin]

Thanks, I forgot about that - thought it was a gimmick back when I tried it a year ago. 101% for my kit.

You should be able to get close to most my timings then since my kit was 102% (if the rating can be trusted).

 

[cellar door]

You should be able to get close to most my timings then since my kit was 102% (if the rating can be trusted).

Just did some preliminary testing and yeah - they work, along with my other previously set subtimmings.

The latency hasn't really improved and I'm getting pretty much similar readings for Read/Write/Copy in Aida 64.

What is up with tRCDWR being able to run at 8, and tRP at 11, and tRAS at only 21? Although I find that below tRAS 26, I have noticed an increase in Membench (ryzen calc).

These are some wild timings and way lower then what the calculator suggests or any other screenshot of e-die kits or even any other threads on the net I've seen. Where these long term stable for you, no whea errors or crashes?

To add, I can do cl14 at 1.49V with all these same timings.

 

[Chomiq]

You should be able to get close to most my timings then since my kit was 102% (if the rating can be trusted).

Have you ever saw Rev E with lower rating? Because as far as I'm aware that "rating" is just based on a lookup table for memory chips. So any Rev E will be rated as 102%.

 

[cellar door]

Have you ever saw Rev E with lower rating? Because as far as I'm aware that "rating" is just based on a lookup table for memory chips. So any Rev E will be rated as 102%.

That's what I was wondering as well - how is that rating calculated?

 

[Chomiq]

That's what I was wondering as well - how is that rating calculated?

It looks like it is actually calculating something, whatever that means:
Micron Rev E Dual Rank 2x16 GB 3600 CL16:

Micron Rev E Single Rank 2x8 GB 3200 CL16:

 

[Taraquin]

Just did some preliminary testing and yeah - they work, along with my other previously set subtimmings.

The latency hasn't really improved and I'm getting pretty much similar readings for Read/Write/Copy in Aida 64.

What is up with tRCDWR being able to run at 8, and tRP at 11, and tRAS at only 21? Although I find that below tRAS 26, I have noticed an increase in Membench (ryzen calc).

These are some wild timings and way lower then what the calculator suggests or any other screenshot of e-die kits or even any other threads on the net I've seen. Where these long term stable for you, no whea errors or crashes?

To add, I can do cl14 at 1.49V with all these same timings.

I consider aida a poor benchmark, use dram calc test og see how long it takes. Only thing I use aida for is estabilshing if soc/iod volt is high enough, as too low volt gives 3-5ns higher latency.

It looks like it is actually calculating something, whatever that means:
Micron Rev E Dual Rank 2x16 GB 3600 CL16:
View attachment 217692
Micron Rev E Single Rank 2x8 GB 3200 CL16:
View attachment 217693

No one except Usmus knows what they test for, but it seems to matter as my B-die which has 94% and generally needing 0.05V more than comparable sets is on spot, while my rev E do better tRP and tRFC than most.

How do the dual rank kit hold up? I bet if the rating is valid you could expect quite low tRCDRD, tRC and tRFC comoared to most rev E

 

[AVATARAT]

4000 IF stable, I can go up but with Whea errors.
In AIDA 52 ns.
VSOC auto under 1.10V
Up to 5150MHz boost on the fastest core.

 

[Taraquin]

4000 IF stable, I can go up but with Whea errors.
In AIDA 52 ns.
VSOC auto under 1.10V
Up to 5150MHz boost on the fastest core.

Nice, how do you get +400MHz pbo boost? Use curve optimizer?

As for timings you could try 4wtrs, 12 wtrl, 12 wr, 6 rtp and probably lower trfc sub 300.

 

[AVATARAT]

For frequency above 4.85GHz, you need BIOS with AGESA before 1.1.9.0 (that's for my mobo) and, I am not sure, but probably 5600x with one CCX. (And yes Curve Optimizer).
After this AGESA, AMD locked the boost to be max up to 4.85GHz for 5600x.
I don't know why.

I set it up from this menu:


I will try this timings, thank you.

 

[cellar door]

I consider aida a poor benchmark, use dram calc test og see how long it takes. Only thing I use aida for is estabilshing if soc/iod volt is high enough, as too low volt gives 3-5ns higher latency.


No one except Usmus knows what they test for, but it seems to matter as my B-die which has 94% and generally needing 0.05V more than comparable sets is on spot, while my rev E do better tRP and tRFC than most.

How do the dual rank kit hold up? I bet if the rating is valid you could expect quite low tRCDRD, tRC and tRFC comoared to most rev E

Aida64 is consistent and has been for a very long time, it's reliable, at least in my experience - that makes it good in my eyes.

With some more testing going as low as the timings you posted, has actually produced detrimental results. The system remained stable but latency was up and write/read/copy will down. Membench in Ryzen calc didn't improve either. Maybe it's my board or my kit. Increasing voltage didn't help.

But I did manage to drop couple timings a small bit, getting me down to 64ns.

 

[freeagent]

That's pretty neat. So a 5600X used to boost like a 5900X? I would flash back just to be able to do that.. Does that +400 work on all Zen 3?

 

[AVATARAT]

That's pretty neat. So a 5600X used to boost like a 5900X? I would flash back just to be able to do that.. Does that +400 work on all Zen 3?

I know only for 5600x.

 

[Taraquin]

Aida64 is consistent and has been for a very long time, it's reliable, at least in my experience - that makes it good in my eyes.

With some more testing going as low as the timings you posted, has actually produced detrimental results. The system remained stable but latency was up and write/read/copy will down. Membench in Ryzen calc didn't improve either. Maybe it's my board or my kit. Increasing voltage didn't help.

But I did manage to drop couple timings a small bit, getting me down to 64ns.

I get no improvement in aida lowering trrdl by 2, tRP by 1, tRC by 1, tWR by 2, tRTP by 1, and tRFC by 6. In dram calc that gives me 2 sec lower on the dram calc test. Could be that timings you try are mismatched? TRFC apoarently likes to be divideable by tRC and/or tRTP. If you use tRC 56, then 560 tRFC might work better than 550, or if you use 6 tRTP 540 tRFC might be perfect.

Some timings like to be configured in pairs, tRTP likes to be half tWR, tRDWR likes to be half tRCDRD on SR-sticks or a bit higher on DR. Trrdl likes to be equal to or 2 higher than trrds.

 

[AVATARAT]

As for timings you could try 4wtrs, 12 wtrl, 12 wr, 6 rtp and probably lower trfc sub 300.

I can't go 6 RTP but other timings are fine, so I trying to achieve 2033 IF now
But I am a bit worry about SoC above 1.10V and for IF 2033 I need 1.131V.

 

[Taraquin]

I can't go 6 RTP but other timings are fine, so I trying to achieve 2033 IF now
But I am a bit worry about SoC above 1.10V and for IF 2033 I need 1.131V.

1.15V is technically considered safe, above is a bit more unsafe.

 

[freeagent]

Over 1.16 on my 5600X gets me no where, same with on my 5900. I see people recommending 1.25 soc and it just makes me cringe.

 

[evilhf]

Muito bom !

 

[Mussels]

Muito bom !

This forum is english only, btw

 

[Taraquin]

Over 1.16 on my 5600X gets me no where, same with on my 5900. I see people recommending 1.25 soc and it just makes me cringe.

I need 1.18V to run 2033 IF without performance drop, 2066 needs 1.24V. 2100 I have not been able to boot. 1.3V maybe? This eats into the powerbudget so performance is the same as 2000 which gets away with 1.12V (technically 1.11V works fine too, but 1.1V and lower gives me perf penalty). My soc scaling is good up to 2000 IF. 1900 needs 1.06, 1933 1.08, 1966 1.1 etc.

 

[AlwaysHope]

I find the Zen 3 chips unimpressive with IF & Ram oc. Is it just me or is the general consensus?
I mean Intel with Z590 boards get up to & over 5000Mhz with ram. For DDR4, that's very good imo.
Such a dilemma going on with upgrading atm. Alder lake & Win 11 with DDR5 bring too much change too soon with new tech. In any case early next month we all will soon know...

 

[Mussels]

I find the Zen 3 chips unimpressive with IF & Ram oc. Is it just me or is the general consensus?
I mean Intel with Z590 boards get up to & over 5000Mhz with ram. For DDR4, that's very good imo.
Such a dilemma going on with upgrading atm. Alder lake & Win 11 with DDR5 bring too much change too soon with new tech. In any case early next month we all will soon know...

Yeah, Zen isn't great with RAM OC.

But on the Intel side, almost nothing actually benefits from the faster RAM (i saw plenty of reviews showing that while they can go faster, outside synthetic benches the gains were very small)

 

[freeagent]

Yeah they don't clock worth a dam, that's why we get so excited over the smallest of gains