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

[Ohnoitzbryan]

May be more stable now, mlst important is if it's gamingstable

What is the worst that can happen if the RAM is unstable?

I ran Karhu again and I keep getting 1 error after multiple hours. Maybe related to heat? 50+C

I dialed back on the Timings to 16-16-16-32, T1, GDM enabled.

1ns slower and Copy bandwith suffered a bit, rest is still around 59k Bandwith.
However it allows me to run it with 1.40v compared to 1.45v.

It was stable for hours and gaming but again karhu ran over night and spit out 1 error after around 3 hrs.

Anything else I could do to improve stability?
I might try to put a fan over the DIMMS.

 

[Taraquin]

What is the worst that can happen if the RAM is unstable?

I ran Karhu again and I keep getting 1 error after multiple hours. Maybe related to heat? 50+C

I dialed back on the Timings to 16-16-16-32, T1, GDM enabled.

1ns slower and Copy bandwith suffered a bit, rest is still around 59k Bandwith.
However it allows me to run it with 1.40v compared to 1.45v.

It was stable for hours and gaming but again karhu ran over night and spit out 1 error after around 3 hrs.

Anything else I could do to improve stability?
I might try to put a fan over the DIMMS.

Typically BSODs during load. If RFC (and REFI on Intel) is too low there is risk of data corruption. I would keep the 1.45v setting that got one rare error, may be heat related, may be a timeout etc, but remember that rare errors in memtest probably means it 100% stable in games. On my previous setup I had a pair of Micron sticks. I ran them at 3733 at rc 55 for almost a year, 100% stable. Then I tried karhu and testmem 5, got loads of errors. Turns out I need rc 56, IRL it gad nothing to say.

 

[Valantar]

What is the worst that can happen if the RAM is unstable?

Random hard system crashes/BSODs, not unlikely to include some data loss/corruption.

I ran Karhu again and I keep getting 1 error after multiple hours. Maybe related to heat? 50+C

If it's Samsung B-die, then yes, that's a heat problem. B-die doesn't like to exceed 50°C at all.

 

[Ohnoitzbryan]

Random hard system crashes/BSODs, not unlikely to include some data loss/corruption.

If it's Samsung B-die, then yes, that's a heat problem. B-die doesn't like to exceed 50°C at all.

I assume its B-die, according to the label on the kit , 3200mhz c14 XMP. Also found it listed on bdie finder.

It really got over 50C towards the 60C mark when i ran Karhu last night.
I think it has to do with the case and the overall airflow, mostly exhaust on my Radiators and only the bottom brings in some fresh air. This is probably not enough airflow for the DIMMs.
I mean idle and during gaming i reach mid 40C , its currently 25C room temperature and since the GPU is also liquid cooled it wont make the DiMMs any hotter during gaming sessions.

The only time i notice a decrease in RAM temps is when the GPU Radiator fans on top are running around 60% and they pull the heat away from the RAM. I wonder if a push pull config would improve this further as the fans would be closer to the RAM kit?

For now i installed a single Noctua Fan which blows directly towards the DIMMS and i see idle of around 32C and low 40C during Karhu. I might try karhu tonight again and let the Fan run to see if the Errors i get are heat related or not.

Not gonna lie, it looks super silly but it does its job!

 

[Valantar]

I assume its B-die, according to the label on the kit , 3200mhz c14 XMP. Also found it listed on bdie finder.

It really got over 50C towards the 60C mark when i ran Karhu last night.
I think it has to do with the case and the overall airflow, mostly exhaust on my Radiators and only the bottom brings in some fresh air. This is probably not enough airflow for the DIMMs.
I mean idle and during gaming i reach mid 40C , its currently 25C room temperature and since the GPU is also liquid cooled it wont make the DiMMs any hotter during gaming sessions.

The only time i notice a decrease in RAM temps is when the GPU Radiator fans on top are running around 60% and they pull the heat away from the RAM. I wonder if a push pull config would improve this further as the fans would be closer to the RAM kit?

For now i installed a single Noctua Fan which blows directly towards the DIMMS and i see idle of around 32C and low 40C during Karhu. I might try karhu tonight again and let the Fan run to see if the Errors i get are heat related or not.

Not gonna lie, it looks super silly but it does its job!

View attachment 252638View attachment 252639

I've got a similar setup with an NF-A12x15 directly above my RAM - though a bit cleaner behind the mesh sidepanel of my Meshlicious Then again I'm also sitting at ~30°C ambients with summer hitting southern Sweden hard right now, with idle RAM temps in the low 40s (can't remember what my RAM voltage is set to, but I'm running B-die at 3800 15-15-15-30, so it needs cooling!). When I installed it I intended it to be temporary, but I've ended up just leaving it. It works, and I can't see it, so whatever

 

[Ohnoitzbryan]

Yea it was actually the heat.
Below are my results after almost 3 hours and running that fan on only 50% speed.
Those are around -20C max under constant load.
This way ill def have some headroom for the voltage.

I went ahead afterwards and ordered a old version of the Corsair RAM cooler.
Was able to get a new one on Ebay for 15€ + 4 € shipping cause some shop closed down and they had like 3 in stock.

Just hope the CPU cooler display wont get in the way but then again im sure ill find a way to attach the thing.
Cant complain for under 20€ and still sealed.

 

[Valantar]

Yea it was actually the heat.
Below are my results after almost 3 hours and running that fan on only 50% speed.
Those are around -20C max under constant load.
This way ill def have some headroom for the voltage.

I went ahead afterwards and ordered a old version of the Corsair RAM cooler.
Was able to get a new one on Ebay for 15€ + 4 € shipping cause some shop closed down and they had like 3 in stock.

Just hope the CPU cooler display wont get in the way but then again im sure ill find a way to attach the thing.
Cant complain for under 20€ and still sealed.

View attachment 252709

Oh, I hope you're prepared for the noise from that cooler. A Noctua fan swap might be in order for that thing, unless you get lucky and it somehow has tolerable quality fans.

 

[eazen]

It’s funny the vote doesn’t even cover 3600 which is what I have, with my older 3700X 1800/3800 was possible now I’m down to 3600, didn’t bother to try 3733 or 3666, I don’t think it matters.

 

[Taraquin]

It’s funny the vote doesn’t even cover 3600 which is what I have, with my older 3700X 1800/3800 was possible now I’m down to 3600, didn’t bother to try 3733 or 3666, I don’t think it matters.

I have tuned a lot of Ryzen 5k and so far the worst I found did 3666, since you didn't bother testing 3666 or 3733 who knows if they work 3666 will give you about 1-2% more perf, 3733 1-2% on top of that if you are CPU bound. Not a lot, but some.

 

[freeagent]

t’s funny the vote doesn’t even cover 3600 which is what I have, with my older 3700X 1800/3800 was possible now I’m down to 3600, didn’t bother to try 3733 or 3666, I don’t think it matters.

If it helps I was able to nail down 1900 with 4 sticks on my 3600XT. Not a hair more though.. completely unwilling to play.

 

[eazen]

If it helps I was able to nail down 1900 with 4 sticks on my 3600XT. Not a hair more though.. completely unwilling to play.

That’s a pretty good IMC. 4 ram sticks is usually detrimental, I mean it’s really the same quality as the IMC of the 3700X I had before, it was two double ranked 16 GB sticks, so probably comparable, but I didn’t try more than 1900/3800.

If it helps I was able to nail down 1900 with 4 sticks on my 3600XT. Not a hair more though.. completely unwilling to play.

I have a 5800X3D now, so I don’t think it matters anymore.

 

[freeagent]

That’s a pretty good IMC. 4 ram sticks is usually detrimental, I mean it’s really the same quality as the IMC of the 3700X I had before, it was two double ranked 16 GB sticks, so probably comparable, but I didn’t try more than 1900/3800.


I have a 5800X3D now, so I don’t think it matters anymore.

I sold my XT shortly after I bought my 5900X.. between my 5600X and the 5900X that XT was totally out classed.

 

[eazen]

I sold my XT shortly after I bought my 5900X.. between my 5600X and the 5900X that XT was totally out classed.

So what’s your IF clock now? Must be over 4000 then or at least 4000. I’m not that lucky but to be fair my Ram maxes out at 3800 anyway.

 

[freeagent]

So what’s your IF clock now? Must be over 4000 then or at least 4000. I’m not that lucky but to be fair my Ram maxes out at 3800 anyway.

My 5600X does 2K stable with 2 single rank sticks, and it can bench at 2100. My 5900X does 1900 stable with 4 sticks, and anything more than that and it is a whea generator.

 

[eazen]

My 5600X does 2K stable with 2 single rank sticks, and it can bench at 2100. My 5900X does 1900 stable with 4 sticks, and anything more than that and it is a whea generator.

Both pretty good, comparable to my ex 3700X. It’s ridiculous how big the difference is between regular Ryzen and Ryzen APU imcs though, they can do much more.

 

[Taraquin]

That’s a pretty good IMC. 4 ram sticks is usually detrimental, I mean it’s really the same quality as the IMC of the 3700X I had before, it was two double ranked 16 GB sticks, so probably comparable, but I didn’t try more than 1900/3800.


I have a 5800X3D now, so I don’t think it matters anymore.

5800X3D is less dependent on ram oc/tuning than regular Zen 3 since the large L3 cache makes latency/BW less important, but running tuned 3733 vs 3600 xmp will net you over 10% performance in some games, and 1% lows will be affected the most.
Example Cyberpunk:

9% avg and 12% on lows. 5800X gains 16% on both so large L3 cache compensates for slower ram.

 

[QuietBob]

5800X3D is less dependent on ram oc/tuning than regular Zen 3 since the large L3 cache makes latency/BW less important, but running tuned 3733 vs 3600 xmp will net you over 10% performance in some games, and 1% lows will be affected the most.
Example Cyberpunk:
View attachment 252887
View attachment 252889
9% avg and 12% on lows. 5800X gains 16% on both so large L3 cache compensates for slower ram.

Great info, didn't expect CP2077 to be so bandwidth hungry (DDR5 results). Do you have a link to that video?

 

[Taraquin]

Great info, didn't expect CP2077 to be so bandwidth hungry (DDR5 results). Do you have a link to that video?

All in Russian unfortunately, but numbers speak for themselves. They tune quite good, but leave some potential on the table (they use gear down mode on AMD, 1-2% slower vs 1t, but easier to get stable, on Intel there is some core OC potential left out).

A few games lowes BW, Cyberpunk and Troy perform much better with DDR5 vs DDR4 due to this. Troy loves cores aswell.

 

[eazen]

5800X3D is less dependent on ram oc/tuning than regular Zen 3 since the large L3 cache makes latency/BW less important, but running tuned 3733 vs 3600 xmp will net you over 10% performance in some games, and 1% lows will be affected the most.
Example Cyberpunk:
View attachment 252887
View attachment 252889
9% avg and 12% on lows. 5800X gains 16% on both so large L3 cache compensates for slower ram.

DLSS ultra performance? No it won’t even yield me 1% I’m 99% GPU bound, the last game where this cpu helped me is CP2077 even my old Zen 2 was easily enough here. I also don’t think it will yield me 10% anywhere else to be honest. Zen 2 is far more reliant on ram bandwidth than Zen 3, on top Zen3D is even less than Zen 3.

Great info, didn't expect CP2077 to be so bandwidth hungry (DDR5 results). Do you have a link to that video?

Unless you use low graphics settings which isn’t the point in CP2077, you won’t see any benefits. This game is easily handled by a Zen 2 if you use high graphics settings.

 

[Taraquin]

DLSS ultra performance? No it won’t even yield me 1% I’m 99% GPU bound, the last game where this cpu helped me is CP2077 even my old Zen 2 was easily enough here. I also don’t think it will yield me 10% anywhere else to be honest. Zen 2 is far more reliant on ram bandwidth than Zen 3, on top Zen3D is even less than Zen 3.


Unless you use low graphics settings which isn’t the point in CP2077, you won’t see any benefits. This game is easily handled by a Zen 2 if you use high graphics settings.

My 3600 begs to differ. With my 3060ti my 5600X gets almost 50% more fps in 1080p high, both with tuned ram. If you play at 1080p ram tuning will matter on a 3060ti+ GPU, at 1440p not so much.

 

[eazen]

My 3600 begs to differ. With my 3060ti my 5600X gets almost 50% more fps in 1080p high, both with tuned ram. If you play at 1080p ram tuning will matter on a 3060ti+ GPU, at 1440p not so much.

Well I’m on 1440p Ultra with RT max, so that’s why things differ here. I could use your 3600 it would be enough with my setup as was my 3700X, no big difference between these CPUs. However, maybe I’ll fiddle around a bit more with ram speed, i just hate it when it bootloops again and I need to reset the bios because of that.

 

[Taraquin]

Well I’m on 1440p Ultra with RT max, so that’s why things differ here. I could use your 3600 it would be enough with my setup as was my 3700X, no big difference between these CPUs. However, maybe I’ll fiddle around a bit more with ram speed, i just hate it when it bootloops again and I need to reset the bios because of that.

Understand that, you could aim for a safe oc/tuning, no need to aim for perfection with your setup at 1440p/ultra

Usually small adjustments to RAS, RC, RRD, FAW, RFC gives you the biggest boost, RDWR/WRRD, WR/RTP, SCL's and WTR can boost a bit more.

Typically auto vs safe tuning if you have Samsung B-die 16-16-16 1.45v:
RAS 36 vs 32
RC 64 vs 48
RRD 8/10 vs 6/8
FAW 38 vs 24
RFC 630 vs 300
WTR 5/14 vs 4/12
WR/RTP 24/12 vs 16/8
SCL's 5 vs 4
RDWR/WRRD 18/10 vs 10/3

Should be easy and stable on 99% of all B-die

 

[eazen]

Understand that, you could aim for a safe oc/tuning, no need to aim for perfection with your setup at 1440p/ultra

Usually small adjustments to RAS, RC, RRD, FAW, RFC gives you the biggest boost, RDWR/WRRD, WR/RTP, SCL's and WTR can boost a bit more.

Typically auto vs safe tuning if you have Samsung B-die 16-16-16 1.45v:
RAS 36 vs 32
RC 64 vs 48
RRD 8/10 vs 6/8
FAW 38 vs 24
RFC 630 vs 300
WTR 5/14 vs 4/12
WR/RTP 24/12 vs 16/8
SCL's 5 vs 4
RDWR/WRRD 18/10 vs 10/3

Should be easy and stable on 99% of all B-die

Yea that’s the problem I have Hynix whatever RAM, it’s a TridentZ RGB 3600 CL16 kit. I tried ram tuning at one time with the help of Ryzen tuner by 1usmus, but the settings recommended were barely better than the standard ones and then I opted to use 3800 instead of these slightly sharper timings.

 

[Taraquin]

Yea that’s the problem I have Hynix whatever RAM, it’s a TridentZ RGB 3600 CL16 kit. I tried ram tuning at one time with the help of Ryzen tuner by 1usmus, but the settings recommended were barely better than the standard ones and then I opted to use 3800 instead of these slightly sharper timings.

CJR or DJR? CJR can be a bit tricky, DJR is easier.

 

[Bockoama]

Hello techpowerup community,

i randomly found this post while looking for additonal information on overclocking RAM.

I have experience with overclocking CPU and GPU but not much with overclocking RAM.

I have used a guide which i found on Github to get started.
Maybe someone here can help me to tighten the timings of my RAM Kit.
I use Ryzen 9 5900x and 3200mhz RAM kit 4 x 8GB modules.
According to my research this RAM Kit should be B-Die.

With the guide on Github i so far went up to 3800mhz 1900 IF 1.45v.

Timings are pretty loose right now. I dont know if should up the frequency more until i crash or start to tighten timings?