AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

[opethdisciple]

What board are you using? I'll double check when I get in and report back. Unless it just stays white when the system automatically changes it, Pretty certain it still changes colour. I'm starting to doubt myself now!

Asus Prime Pro x370

 

[opethdisciple]

I was sent this guide from someone on Reddit to help me along the way.
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md

It states:
On Ryzen 3000, there's also CLDO_VDDG (not to be confused with CLDO_VDDP), which is the voltage to the Infinity Fabric. I've read that SOC voltage should be at least 40mV above CLDO_VDDG, but other than that there's not much information about it.

Most cLDO voltages are regulated from the two main power rails of the CPU. In case of cLDO_VDDG and cLDO_VDDP, they are regulated from the VDDCR_SoC plane. Because of this, there are couple rules. For example, if you set the VDDG to 1.100V, while your actual SoC voltage under load is 1.05V the VDDG will stay roughly at 1.01V max. Likewise if you have VDDG set to 1.100V and start increasing the SoC voltage, your VDDG will raise as well. I don't have the exact figure, but you can assume that the minimum drop-out voltage (Vin-Vout) is around 40mV. Meaning you ACTUAL SoC voltage has to be at least by this much higher, than the requested VDDG for it to take effect as it is requested.
Adjusting the SoC voltage alone, unlike on previous gen. parts doesn't do much if anything at all. The default value is fixed 1.100V and AMD recommends keeping it at that level. Increasing the VDDG helps with the fabric overclocking in certain scenarios, but not always. 1800MHz FCLK should be doable at the default 0.9500V value and for pushing the limits it might be beneficial to increase it to =< 1.05V (1.100 - 1.125V SoC, depending on the load-line).

Yep seen all that before. And in my case simply isnt the case as I am still unstable.

Also this is contrary to what it says in the Official AMD Ryzen Master guide on overclocking.

Where have those laymans got those values from? Who has told them that VDDG and SOC should have a 40mV difference?

It doesn't say that in the AMD documentation.

Now you see why I'm confused!!

But based on my experiences following the dram calc, it doesn't work. I've tried everything now multiple times and I am still unstable.

Only thing I havent tried is leaving both VDDG and SOC at 1.1v.

I've been running VDDG at 0.950v which is default.

 

[michty_me]

Ah ok, I just wasn't sure if you had or not. It was the first time that I had read it.

So we now have 40mV, I've read others sayong it should be 0.1V below SoC and my board is auto setting it at 1.1V which ties in with the AMD documentation.

I managed to get my 3200 CL16 Micron E die to run at 3600CL14 for days using Thaiphoon and DRAMcalc.

There appears to be a bit of misinformation on videos on how to run it too. I went down the importort from Thaiphoon way instead of manually entering stuff from a read file.

 

[humbug]

JayZ apology video (sort of)

LOL Jay's half apology 'but still not ma fault'

Yeah.. MY '2Cents...

"It was not really all my fault and AMD have wacky issues" You couldn't just stick to owning up to the mistakes you made you had to get defensive and stick the boot into AMD, again, if you want to talk about "AMD's wacky issues" then then do that in a dedicated video and we will decide for ourselves if AMD are wacky or if its just you, your whole premiss of the CPU's running 1.5v being too high comes from your lack of understanding of how these CPU 'are designed to work' and with that you are already pre programmed to think everything is AMD's fault which is why you jumped the gun on blaming your motherboard issues on AMD's "wacky" platform, this is not in any way shape or form partially or wholly AMD's fault "Because you know.. AMD has wacky issues so can you really blame me for jumping to conclusions?" yes actually, this was wholly your fault. You have a confirmation bias issue.

 

[humbug]

DRam Clac should be seen more as a guide rather than gospel, not all memory modules of identical type are equal.

 

[G550]

I do know tho that my ram doesn't like 1.45v. It was causing instability. I've tried it. It felt like I was forcing it to be stable and eventually it just couldn't carry on any more and started crashing out.

I don't understand, becuase this 8 pack ram according to Gibbo/8 pack is the same bin as the 3600MHz CL16 stuff so should 'just work'.

16 16 16 16 36 1.35v. But I can tell you, it doesn't! I've gone up to 1.45v and still not stable.

Though I don't have direct experience with that board, I suspect the root of your limited RAM frequency may be the physical design of the Prime X370 Pro mainboard.

At the point of release of original Ryzen and the BIOS updates in the following months, one thing I noticed amongst owners posting their experiences with the Prime X370 board was that virtually all of them reported a max realistic ram frequency of ~2933/3066 whereas most other comparable boards were able to reach 3200 or higher relatively easily with good ram (and of course a decent IMC on their particular Ryzen processor).
Subsequent internal timing tweaks from Asus may have raised its limits a little since then, but probably still several hundred MHz behind comparable boards.

I mentally ruled this board out as a candidate at the time for that very reason, and in the event waited until the 2xxx series and Prime X470 Pro before buying-in.

 

[humbug]

If its not stable with Profile Version V1 try V2, Profile V1 was too aggressive for my RAM but V2 is perfect.

 

[opethdisciple]

Though I don't have direct experience with that board, I suspect the root of your limited RAM frequency may be the physical design of the Prime X370 Pro mainboard.

At the point of release of original Ryzen and the BIOS updates in the following months, one thing I noticed amongst owners posting their experiences with the Prime X370 board was that virtually all of them reported a max realistic ram frequency of ~2933/3066 whereas most other comparable boards were able to reach 3200 or higher relatively easily with good ram (and of course a decent IMC on their particular Ryzen processor).
Subsequent internal timing tweaks from Asus may have raised its limits a little since then, but probably still several hundred MHz behind comparable boards.

I mentally ruled this board out as a candidate at the time for that very reason, and in the event waited until the 2xxx series and Prime X470 Pro before buying-in.

Yea. The limitation is probably my mobo. It's a gen1 board and budget as well.

Thing is the ram can run at 3600MHz CL16 for of 2-3 weeks with no problems. With a quite a few 2-3hrs gaming sessions, but will eventually crash/freeze.

 

[humbug]

400 series Motherboards did improve memory overclocking.

 

[Zeed]

And I'm back from my 3 week's Holidays in Azeroth. Best Holidays I'w had for last few years. Stacked up Food for week turned phone off locked house doors and TOTALLY ignored everything for a week. Not picked any phone calls or checked email or facebook or any forums in last 3 weeks. Besides 2 Text messages I'w read.... I Only get good news nowadays....
I'w put in 340 HOURS of gaming in them 3 weeks with 16-17 hours a day every day. With brake for food some sleep and shower. And small 30 minute run to Local Iceland for chicken grills and some green stuff to go with it.
It was just like good old times for first week till news that my friend got killed off by cancer and my best mate is in hospital.... From He texted me few hours after I got back from funeral. REALITY SUCKS BALLS.


Anyway. I not had a single BSOD reboot or crash since start of World of warcraft Classic. Defoo 100% stable after i reverted that ****** chipset driver in August.
Time to do some catching up (email 2000 unread spam christ ) and see what AMD fixed or messed up. I'w seen that they are still trying to fix the 1.0.0.3 microcodes

 

[michty_me]

See now this brings me back to something I mentioned ages ago. In the official AMD Ryzen master documentation they specify you should put VDDG to 1.1v when overclocking the ram.

But I was told that this was incorrect by people on these forums. Even the mobo highlights that value in red indicating it's a high value.

See page 34 for the screen shot from the Ryzen Maser documentation.

I'm going to try this again as a last resort. It did stabilize my system when I was using it, but I changed it because I was told this was not correct and was worried by the red VDDG value.

Looking at the note on the VDDG section it states:

VDDG represents voltage for the data portion of the infinity fabric. It is derived from the CPU SoC/uncore voltage (VDD_SoC). VDDG (input in V) can approach but not exceed VDD_SoC. This may help FCLK overclock.

I've just left it in auto for now.

 

[Woodsta888]

Yep seen all that before. And in my case simply isnt the case as I am still unstable.

Also this is contrary to what it says in the Official AMD Ryzen Master guide on overclocking.

Where have those laymans got those values from? Who has told them that VDDG and SOC should have a 40mV difference?

It doesn't say that in the AMD documentation.

Now you see why I'm confused!!

But based on my experiences following the dram calc, it doesn't work. I've tried everything now multiple times and I am still unstable.

Only thing I havent tried is leaving both VDDG and SOC at 1.1v.

I've been running VDDG at 0.950v which is default.

Have you, each time entered ALL the values from the DRAM calc in or just some? As the tech power up guide written by 1USMUS, states that you only need to test the first few parameters first and most people will get something stable at that point. The rest of the parameters are for tightening and tweaking. You need to follow his sequence for tuning RAM.

Iusmus states the first parameters you set are ProcODT + RTT (NOM, WR and PARK) trying the recommended and alternative values until the sytem has a mimimum number of errors using TM5 test package on the basic preset.

Then find optimal RAM voltage and SOC.

Once these are found half the cases will get a fully stable system at this stage. If TM5 does not find errors use the rest of programs he suggests to find errors.

Next step is debugging - changing the timings. Do not change all the timings at once. One at a time.

The rest of the parameters are for overclocking or tightening of the timings and reducing latency further.


See page 5 of his guide. There are flow charts for each step.
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/5.html

 

[opethdisciple]

Have you, each time entered ALL the values from the DRAM calc in or just some? As the tech power up guide written by 1USMUS, states that you only need to test the first few parameters first and most people will get something stable at that point. The rest of the parameters are for tightening and tweaking. You need to follow his sequence for tuning RAM.

Iusmus states the first parameters you set are ProcODT + RTT (NOM, WR and PARK) trying the recommended and alternative values until the sytem has a mimimum number of errors using TM5 test package on the basic preset.

Then find optimal RAM voltage and SOC.

Once these are found half the cases will get a fully stable system at this stage. If TM5 does not find errors use the rest of programs he suggests to find errors.

Next step is debugging - changing the timings. Do not change all the timings at once. One at a time.

The rest of the parameters are for overclocking or tightening of the timings and reducing latency further.


See page 5 of his guide. There are flow charts for each step.
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/5.html

I just entered in all the settings. What's TM5?

 

[michty_me]

I just entered in all the settings. What's TM5?

Looks like another memory test application

https://www.overclock.net/forum/27577522-post2594.html

 

[Woodsta888]

Yeah, it's Testmem5. What the DRAM guy says to use as other programs will count many other errors. As a memOC'ing pro best to use what he says.

What I did in the end was basically what 8PACK says to do, SOC 1.1 and enter the timings manually and use the system and check for stability for a while. No point diving in and attempting a RAM OC if the RAm at stock isn't stable.

My RAM is 3600CL16 as standard and it wouldn't work unless I did that. So get a stable system at 3200Mhz first for your system and I did for 3600 for mine. Then start with OC'ing to 3600. But it will take some time and patience. I think too many people misunderstood the calc. The calc just gives you the parameters that work with each other - or a range of. Looking at the imported timings through Typhoon or R- XMP, similar dies (B-die) are exactly the same speed wise when imported so my 3600 RAM shows the same imported latencies in ns as lower Mhz RAM. So even if I use the DRAM calc at 3600Mhz (which my RAM is anyway) it will give a different set of timings from what XMP is as it is already trying to use the typhoon import as the basis. No where does it collect the baseclock speed of the RAM and import that.

Also, when you change all the parameters at once, when the system reboots the BIOS has to cycle through and change some settings before others. You can see this mistake Jays2cents makes pointing out that CPU error in his video. But on his apology video he let the BIOS cycle and guess what - it booted...twice.

My system will hang on the BIOS sometimes at first cycle and I have to manually restart it and then it loads in the rest. I did start off thinking it had failed. I increased my BIOS cycles to 10 before the BIOS fails the bot cycle sequence. Default is 3 cycles. My MOBO doesn't have any POST numbers or screen and I think many look at this and the first message or numbers set they switch the machine off as a fail. Leave it to keep cycling and adjusting parameters. Even 1USMUS mentions this in his guide. Depending on the BIOS memory will be in the BIOS sequence before the CPU hence CPU error, particularly if you are changing en masse.

Lets use the car analogy and formula 1. The team don't change lots of parameters all at the same time. They change one - test, then another, test. Once they build a picture of how all the parameters affect each other, then then can calculate the limits of the other parameters. This is what the DRAM calc does. The initial parameters it uses are the physical latencies from the RAM chips imported from R-XMP or typhoon, the calc then give you values that will work within these ranges. This is why when you use the DRAM calc and input your RAm speeds as stock, as it's an overclocking tool, they appear tighter than what your actual XMP timings are.

These are my observations and my opinion only as I'm still n00b at this RAM OC'ing thing myself, so happy to be corrected. I am definitely no expert.

 

[Psycho Sonny]

And I'm back from my 3 week's Holidays in Azeroth. Best Holidays I'w had for last few years. Stacked up Food for week turned phone off locked house doors and TOTALLY ignored everything for a week. Not picked any phone calls or checked email or facebook or any forums in last 3 weeks. Besides 2 Text messages I'w read.... I Only get good news nowadays....
I'w put in 340 HOURS of gaming in them 3 weeks with 16-17 hours a day every day. With brake for food some sleep and shower. And small 30 minute run to Local Iceland for chicken grills and some green stuff to go with it.
It was just like good old times for first week till news that my friend got killed off by cancer and my best mate is in hospital.... From He texted me few hours after I got back from funeral. REALITY SUCKS BALLS.


Anyway. I not had a single BSOD reboot or crash since start of World of warcraft Classic. Defoo 100% stable after i reverted that ****** chipset driver in August.
Time to do some catching up (email 2000 unread spam christ ) and see what AMD fixed or messed up. I'w seen that they are still trying to fix the 1.0.0.3 microcodes

sounds like you are living the life bro. i'm glad to see you being so successful. keep up the good work.

 

[opethdisciple]

Thanks for all the help people. The issue is I still remain at an utter loss how to get my ram stable at 3600MHz CL16.

I think the most stable I had it was when I manually entered in only the base timings (16 16 16 16 36 - 1.35v - SOC 1.1v) This remained stable for about a month but then eventually exhibited instability.

I then tried 1.4v all the way to 1.45v on the dram, with 1.45v actually making things more unstable.

But the truth is that 3200MHz CL14 is a smidgen faster than 3600MHz CL16 any way?

3200MHz CL14: 8.75ns

3600MHz CL16: 8.9ns

 

[Panos]

Thanks for all the help people. The issue is I still remain at an utter loss how to get my ram stable at 3600MHz CL16.

I think the most stable I had it was when I manually entered in only the base timings (16 16 16 16 36 - 1.35v - SOC 1.1v) This remained stable for about a month but then eventually exhibited instability.

I then tried 1.4v all the way to 1.45v on the dram, with 1.45v actually making things more unstable.

But the truth is that 3200MHz CL14 is a smidgen faster than 3600MHz CL16 any way?

3200MHz CL14: 8.75ns

3600MHz CL16: 8.9ns

3866C16 you will find gains, not 3600C16. Submittings matter also.

 

[opethdisciple]

3866C16 you will find gains, not 3600C16. Submittings matter also.

But thats just for memory access times, is there any benefit on the CPU it's self when the IF is running at a faster frequency. Or is it literally just for access to memory?

Because 3600MHz will have a IF of 1800MHz whilst 3200MHz will be at 1600MHz.

 

[Panos]

But thats just for memory access times, is there any benefit on the CPU it's self when the IF is running at a faster frequency. Or is it literally just for access to memory?

Because 3600MHz will have a IF of 1800MHz whilst 3200MHz will be at 1600MHz.

How you get 8.75 and 8.9ns? What program you use?