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

[freeagent]

You are talking about the flareX modules? What is your current set-up with those memory modules if I may ask? I'm going to try to push those a bit to see what I can get. I'm sure there is still some performance left.

People say the only difference between them and tridents is the heatspreader. My current setup is in my sig

I read a G.Skill rep say 3200C14 is a very good bin.. you should be ok..

 

[Mussels]

PBO back on, this ram will not go even one setting higher, nor lower on the CAS latencies. Apparently it just likes C18.
I can kill windows services and do a billion tweaks to get the scores higher, it's just a happy stable speed.

 

[freeagent]

Awesome man! She’s rippin

 

[dont whant to set it"']

I've slotted back in my dual rank hynix's and give them another try on the new firmware, though so far not quite a week since , the rig is stable for my ''tea cups''.
Well it turns out that most likely it was user error preventing me clocking these thing past 2800MT on a few bios versions ago.
It wont post trying to go for 1900MHz IF with raised VSOC tension to 1.15V , all else the same , CPB off, 47.5x cpu multi @1.3625V , 1.38V dimm , gear down mode: default/auto , manual primary timings and trfc's , vrm's @ T.probe/standard/optimized.

 

[Taraquin]

PBO back on, this ram will not go even one setting higher, nor lower on the CAS latencies. Apparently it just likes C18.
I can kill windows services and do a billion tweaks to get the scores higher, it's just a happy stable speed.

View attachment 221711

The 3 primaries are not that important, they mean less for speed than most people think, but going from cl20 to cl16 won`t net your more than 1-2% on fps, going from tRFC 700 to 300 would easily gain you 5%. Do some basic tweaks to trrds\l, faw, trc, twr, trtp and trfc and you can easily shave off 5-10ns of latency and boost read and copy by 3-4k

 

[Mussels]

The 3 primaries are not that important, they mean less for speed than most people think, but going from cl20 to cl16 won`t net your more than 1-2% on fps, going from tRFC 700 to 300 would easily gain you 5%. Do some basic tweaks to trrds\l, faw, trc, twr, trtp and trfc and you can easily shave off 5-10ns of latency and boost read and copy by 3-4k

Yeah, i have no idea where to begin with that stuff. The last time i tried lowering TRFC i got nowhere... it's a multiple of another timing, isn't it?

 

[Mickey153]

My 5600X can run 4000 but Whea Errors not effecting stability, but IF 1966 rock stable

 

[Mussels]

Okay so at 4000, the system is rock stable but getting WHEA errors - LOTS of em

3800, none at all

Any ideas on how to fix those? Staying at 3800 for now, if i cant fix it i'll try timings instead

 

[Mickey153]

Okay so at 4000, the system is rock stable but getting WHEA errors - LOTS of em

3800, none at all

Any ideas on how to fix those? Staying at 3800 for now, if i cant fix it i'll try timings instead

Up your VDDG IOD voltage and Vsoc keeping the IOD voltage 40mv under Vsoc voltage these voltages top out safely at vsoc 1.2 v vddg iod 1.15v, personally I had no luck at 4000 without loosening the timings, and lost speed and latency to the point that 3933 is faster, but every chip is different so you could give it a try, ps because of vdroop keep iod 50mv lower or it will run at default voltage 900

 

[Taraquin]

Okay so at 4000, the system is rock stable but getting WHEA errors - LOTS of em

3800, none at all

Any ideas on how to fix those? Staying at 3800 for now, if i cant fix it i'll try timings instead

A new agesa might help, certain combos of voltages might help, but if you get very many it's often hard to fix. Tried 3866 or 3933? I know 4000 sounds better

As for timings at 3800 try 2T, gdm off, cl17, tRCDRD 21, tRP 21, tRAS 39, tRC 60, tRFC 480 (tRFC should be divideable by 8 if you have 2x8gb ram og by 16 if you have 2x16 or 4x8), tFAW 24, tRRDS 6, tRRDL 8, tWR 16, tRTP 8, tWTRS 4, tWTRL 8 or 9, SCL's 4, ProcODT 37-43 if 32gb or 28-37 if 16gb, soc volt 1.1, vddg iod 1.06, ccd 0.92, vddp 0.88. Rest on auto. Change timing one by one, boot and see if it works. If you can run this it would boost gaming by up to 10-15% vs stock. You might be able to run tighter too.

 

[Mussels]

3866 still had errors, 3800 was WHEA free.

I guess 3800 tweaked will be where it ends up, it's odd how it can be rock solid stable but spam the errors.

 

[Taraquin]

3866 still had errors, 3800 was WHEA free.

I guess 3800 tweaked will be where it ends up, it's odd how it can be rock solid stable but spam the errors.

Certain CPU and most full ATX boards gets wheas above 3800. Many mATX and ITX can run 4000-4200 whea free. Why is a good question. Try the timings I suggested, change all voltages, procodt first and then go for timings one by one

 

[Mussels]

Certain CPU and most full ATX boards gets wheas above 3800. Many mATX and ITX can run 4000-4200 whea free. Why is a good question. Try the timings I suggested, change all voltages, procodt first and then go for timings one by one

Nah, getting too much effort.

 

[Taraquin]

Nah, getting too much effort.

I can give you basic timing-input if you want to use less time? Would still yield you 5-10% performance easily If you keep first 4 at 18-22-22-22 set tRAS to 40, tRC to 60, tRFC to 480, keep rest at auto. But if you are satisfied then stick with what you got

 

[Mussels]

It's more about if i spend the time and get it all working perfectly, the next AGESA update, BIOS reset or slight change in hardware means i'm probably doing it all over again and i just dont have time for that.

 

[Anoniem]

So I have the memory settings in my sig dialed in, stable and no weird behaviour. I've had IF at 1900 for a while and have tried IF at 2000, 1967 and 1933 and all result in WHEA's clogging up the logs. After stumbling upon this post (massive thanks for the info you all!) it gave me some hope that 2000 might be possible so I flashed F35e but no dice. I set my ram at an easy 2133 with stock timings and tried 1.093 VSOC, nothing. 1933 with the same SOC voltage did boot but was incredibly unstable and caused the board to recover after 10 cycles.

What am I missing?

 

[Taraquin]

It's more about if i spend the time and get it all working perfectly, the next AGESA update, BIOS reset or slight change in hardware means i'm probably doing it all over again and i just dont have time for that.

If you use 1T gdm and all resistances, ProcIDT etc on auto it usually works across agesas. All the settings I used on agesa 1.1.0.0 also worked fine on 1.2.0.0, 1.2.0.1, 1.2.0.2 and 1.2.0.3. The settings I suggested abive should be very safe and there is much room for improvement, but the closer you get to max the more likely it is to be unstable. Your choice if you wanna play totally safe, generally safe and easily gain 10-15% perf or risk instability and possibly gain 20%+ in certain games/apps If you play at 1440p or 4k or 1080p with a gpu below 3060ti/3070 the gain is lower. The moment you use dlss even at 4k you get very high gains. Helped a friend of mine with crap ram and basic tuning. Going from 3200cl16 to 4000cl18 and some timings tuned fps in Cyberpunk at 4k dlss perf went from 145-155 at a certain scene to 170-180fps. With in depth tuning 200fps+ would have been possible

 

[freeagent]

What am I missing?

You aren't missing anything.. some can some cannot. I too have one that cannot. I get whea's at 1933 with 2 and 4 sticks. Voltage makes no difference in my case. Though I might try a higher IOD later.. but pretty sure I would be wasting my time. My 5600X starts tossing whea's at 2033. 5900X still destroys it at lower FCLK.. except in memory reads, and latency, but I think its pretty close.

 

[Anoniem]

You aren't missing anything.. some can some cannot. I too have one that cannot. I get whea's at 1933 with 2 and 4 sticks. Voltage makes no difference in my case. Though I might try a higher IOD later.. but pretty sure I would be wasting my time. My 5600X starts tossing whea's at 2033. 5900X still destroys it at lower FCLK.. except in memory reads, and latency, but I think its pretty close.

Yeah that's sadly pretty much what I expected. Anyway, 3800/1900 and tight timings is fine too, it's just the itch. the never ending itch to get it one notch higher

 

[purecain]

Damn my aida 64 memory benchmark is asking me to upgrade... meh...
Bit off topic but are those cheap keys online legit...???? managed to find a key..

@Mussels how are you getting the extra .25 multi. Is it a simple setting ive missed or have you tuned each core like we used to have to with the 1800x.

second run with less programs running in the background. Virtually the same results.

 

[Mussels]

Damn my aida 64 memory benchmark is asking me to upgrade... meh...
Bit off topic but are those cheap keys online legit...???? managed to find a key..

View attachment 222481
@Mussels how are you getting the extra .25 multi. Is it a simple setting ive missed or have you tuned each core like we used to have to with the 1800x.

I have PBO on +200, and have a small undervolt curve... could even be a measuring error

 

[Taraquin]

Damn my aida 64 memory benchmark is asking me to upgrade... meh...
Bit off topic but are those cheap keys online legit...???? managed to find a key..

View attachment 222481
@Mussels how are you getting the extra .25 multi. Is it a simple setting ive missed or have you tuned each core like we used to have to with the 1800x.

View attachment 222482

second run with less programs running in the background. Virtually the same results.

Post your zentimings if you want input on tweaking, can probably shave of 5-7ns latency and boost rest by 2-3k MB easily

 

[Valantar]

Decided to start tinkering a bit with RAM speeds on my main rig yesterday, after running it at straight DOCP settings since building it. ASRock B550 Phantom Gaming ITX-AX, 2x16GB G.Skill FlareX (dual rank B-die, B1 PCB, 3200c14), 5800X.

I don't know crap about RAM OCing, so for now I've
a) tried 1usmus' DRAM calc, which failed to POST even at 3600 safe (not really that strange given that it doesn't actually support Zen3 or B550), and
b) set everything to auto except the RAM and IF speed to see how high I could go.

The latter experiment was reasonably successful, though of course the timings are utter garbage - that's the next step, trying to get my head around which timings to tweak and which numbers are bad/acceptable/good. Hit 3800/1900 @ 1.2V, 4000/2000 @ 1.35V. 4066/2033 seemed to POST successfully once (got into the BIOS), but mostly just reset to safe settings by itself. 4000/2000 booted into Windows with no issues, did not show any WHEA errors, and was stable through a quick run of the Fire Strike CPU test, which is reasonably memory intensive. Didn't bother trying higher DRAM voltages, and it seems like people are recommending auto SoC voltage for Zen3, instead adjusting LLC? That's what I did, auto voltages (except DRAM) and level 3 SoC LLC.
Obviously haven't run any benchmarks yet as there's no point at these garbage timings, but at least I know my RAM and IF can both run faster than 3200 - though I don't know if they can do so at reasonable latencies and voltages. Just to reiterate: everything below except MCLK and FCLK is auto.

Edit: damn, looking at how much I paid for this kit vs. how cheap I can get a DDR4-4000c17/c18 kit today makes me a bit sad. Nearly 33% less!

 

[tabascosauz]

@Valantar you can run the loose timings to gauge where your IF stops scaling but beware, I've tripped Bus/Interconnect WHEA on unstable memory timings before. I previously thought IF and DRAM stability were completely distinct from one another (thus theoretically Bus/Interconnect should have nothing to do with timings) - still mostly true, but not completely so. Granted, I was doing some stupid shit with tRFC so it was blatantly unstable in TM5 to have done so.

Give it a shot with the timings you want to be running. 3200CL14 G.skill dual rank I'd say just shoot for 4000 16-16-16-36 right away at say 1.45V. Working on tRC and tRFC can come after you nail that down, then the rest can wait till later (the rest don't have too much an impact on VDIMM, within reason).

as for VSOC not sure where that info came from but you definitely want to be on manual control to have full control over what you're setting. If you don't like the droop you see on SOC, bump up the LLC, but otherwise just leave LLC alone (I'm usually fine with 0.015-0.02V during TM5). 1.2V is reasonable for 4000, but see if you can shave it down a bit - take it easy, small steps, and leave it at each for a couple of days to see if Bus/Interconnect comes up.

If you want to work out a 4000 daily, see about dropping VDDP a bit. Not seen high VDDP (>0.95) help stability for daily configs before. IOD is whatever, do what you need to do, I did need to support IOD about that much for 4000 when testing 2000MHz on mine.

of course, I say pump the VDIMM but that's assuming you have airflow. I'm not sure what 1.35V will get you - average B-die dual rank probably falls around 3600-3800CL16 with that imo

 

[Valantar]

@Valantar you can run the loose timings to gauge where your IF stops scaling but beware, I've tripped Bus/Interconnect WHEA on unstable memory timings before. I previously thought IF and DRAM stability were completely distinct from one another (thus theoretically Bus/Interconnect should have nothing to do with timings) - still mostly true, but not completely so. Granted, I was doing some stupid shit with tRFC so it was blatantly unstable in TM5 to have done so.

Give it a shot with the timings you want to be running. 3200CL14 G.skill dual rank I'd say just shoot for 4000 16-16-16-36 right away at say 1.45V. Working on tRC and tRFC can come after you nail that down, then the rest can wait till later (the rest don't have too much an impact on VDIMM, within reason).

as for VSOC not sure where that info came from but you definitely want to be on manual control to have full control over what you're setting. If you don't like the droop you see on SOC, bump up the LLC, but otherwise just leave LLC alone (I'm usually fine with 0.015-0.02V during TM5). 1.2V is reasonable for 4000, but see if you can shave it down a bit - take it easy, small steps, and leave it at each for a couple of days to see if Bus/Interconnect comes up.

If you want to work out a 4000 daily, see about dropping VDDP a bit. Not seen high VDDP (>0.95) help stability for daily configs before. IOD is whatever, do what you need to do, I did need to support IOD about that much for 4000 when testing 2000MHz on mine.

of course, I say pump the VDIMM but that's assuming you have airflow. I'm not sure what 1.35V will get you - average B-die dual rank probably falls around 3600-3800CL16 with that imo

Thanks for the input The "leave vSOC at auto" advice was from several OC guides I've skimmed (this one is the only one I still had open) - of course I have no idea of their quality, so any feedback is welcome. Btw, isn't 1.2V vSOC a bit on the high side for longevity? Thx for the guidance on other voltages though - auto settings do tend to make for some weird baseline readings, so getting some input on what is reasonable/expected stable for manual settings is great.

I'll keep the IF/memory interplay in mind for future testing - for now this has been a pretty simple "either it boots error-free or it fails to POST" scenario, but I'm sure I'll run into some grey areas soon enough.

I'm honestly not very picky in terms of what I'll end up at - I'd be equally happy with 3800 as 4000, and even 3600 would be perfectly fine as long as there's some performance increase over DOCP. I'm just looking to get this tuned better than stock, among other things because I have an inkling that this motherboard doesn't train these DIMMs all that well (I've been having some odd issues with long wake from sleep times and high DPC latency after long sleep cycles that I can't find any explanation for beyond this RAM not being on the QVL and the motherboard maybe not liking it - and ASRock support isn't helping either). We'll see how it plays out.

I'll have a look at tuning this a bit tonight or tomorrow