Ryzen Owners Zen Garden

[tabascosauz]

So, whatever LLC setting my board is doing on auto, sets my Vsoc to 1.075v when i set to 1.1v in BIOS, but doing so seems to have murdered my systems performance. Thoughts?

That's a sizeable droop but not unreasonable unless it's dropping even lower when you stress IF (ie. memtesting). Could you elaborate on "murdered" system performance? Do you mean reduced memory benchmark performance? If you're not pushing enough VSOC, then memory performance and stability can suffer. If IF is on the edge, I mainly pay attention to VSOC and VDDG_IOD.

You really shouldn't need 1.1V for 3600 though? For an average 5900X maybe even 1.05V is a bit much (1.05V get, not set). I set 1.125V for 3800 and get 1.106V or so.

If you find the droop excessive just up the LLC a notch or two. Auto LLC on Gigabyte is usually middle of the road. There's no iGPU so SOC power draw really doesn't vary at all, don't see much risk in running even High or Turbo, though I don't see why it would ever be necessary.

 

[MxPhenom 216]

That's a sizeable droop but not unreasonable unless it's dropping even lower when you stress IF (ie. memtesting). Could you elaborate on "murdered" system performance? Do you mean reduced memory benchmark performance? If you're not pushing enough VSOC, then memory performance and stability can suffer. If IF is on the edge, I mainly pay attention to VSOC and VDDG_IOD.

You really shouldn't need 1.1V for 3600 though? For an average 5900X maybe even 1.05V is a bit much (1.05V get, not set). I set 1.125V for 3800 and get 1.106V or so.

If you find the droop excessive just up the LLC a notch or two. Auto LLC on Gigabyte is usually middle of the road. There's no iGPU so SOC power draw really doesn't vary at all, don't see much risk in running even High or Turbo, though I don't see why it would ever be necessary.

Performance in the AC Valhalla benchmark went down to about 70 FPS from about 80-85fps that is was. Once I bumped Vsoc to 1.15v again and set LLC to Turbo, which has it showing at 1.138v in Windows performance went back up.

Shit box 5900x lol?

I dropped it back to 1.125, which reads at 1.112 in Windows, and performance has stayed constant with what im expecting. Let me run a few more tests.

Okay shit seems okay now. Im at 1.1v in BIOS, 1.087v in Windows. Performance normalized. LLC set to Turbo for the Vsoc.

Is there any way to read DRAM voltage in hwinfo? Its set to AUTO in BIOS, but XMP is enabled so it should be getting 1.45v, just want to confirm though.

 

[tabascosauz]

Performance in the AC Valhalla benchmark went down to about 70 FPS from about 80-85fps that is was. Once I bumped Vsoc to 1.15v again and set LLC to Turbo, which has it showing at 1.138v in Windows performance went back up.

Shit box 5900x lol?

I dropped it back to 1.125, which reads at 1.112 in Windows, and performance has stayed constant with what im expecting. Let me run a few more tests.

Okay shit seems okay now. Im at 1.1v in BIOS, 1.087v in Windows. Performance normalized. LLC set to Turbo for the Vsoc.

Is there any way to read DRAM voltage in hwinfo? Its set to AUTO in BIOS, but XMP is enabled so it should be getting 1.45v, just want to confirm though.

It should be 1.45, but if you're worried just type in 1.45 in the box. Don't count on the software VDIMM sensor, I've yet to come across a single AM4 Gigabyte board that doesn't purport to overvolt VDIMM by 0.03-0.06V according to the software sensor. just gigabyte things

I'm not entirely convinced that the software reading is even accurate, because when I move the same kits to non-Gigabyte boards I still go off the BIOS value and not the overvolted software value.

 

[freeagent]

I use 1.156-1.162 SOC for 2K and up on my 5600X

My Z77 OC Formula and Strix B550-F don't show vdimm which makes me sad

 

[HD64G]

Obviously not for him.

Since it was an AGESA bug and the previous to current published AGESA solved it for most, then it isn't an AMD bug but a motherboard vendor. My 5c.

 

[MxPhenom 216]

Since it was an AGESA bug and the previous to current published AGESA solved it for most, then it isn't an AMD bug but a motherboard vendor. My 5c.

AGESA updates are pushed to mobo vendors from AMD. And USB issues arent unique to a specific board vendor. Its widespread. It has literally effected all of them. Not just x570 either. B550 has its share of usb issues too

 

[HD64G]

AGESA updates are pushed to mobo vendors from AMD. And USB issues arent unique to a specific board vendor. Its widespread. It has literally effected all of them. Not just x570 either. B550 has its share of usb issues too

But since May this bug was solved for MSI X470 & B450 motherboards and for X570 and B550 it was even earlier. So, for other vendors to still have problems it might have something to do with their UEFIs and not the AGESA code.

 

[MxPhenom 216]

But since May this bug was solved for MSI X470 & B450 motherboards and for X570 and B550 it was even earlier. So, for other vendors to still have problems it might have something to do with their UEFIs and not the AGESA code.

It doesn't sound like for everyone who owns an MSI board that it has been fixed though is what I am saying. There are still those outliers for every vendor where its still an issue. Theres mega threads all over reddit since the first release of AGESA update that addresses the issue where people still report problems, and its not trending to one vendor or another.

It should be 1.45, but if you're worried just type in 1.45 in the box. Don't count on the software VDIMM sensor, I've yet to come across a single AM4 Gigabyte board that doesn't purport to overvolt VDIMM by 0.03-0.06V according to the software sensor. just gigabyte things

I'm not entirely convinced that the software reading is even accurate, because when I move the same kits to non-Gigabyte boards I still go off the BIOS value and not the overvolted software value.

In the BIOS to the side where it has RAM and CPU information, ram was at 1.47v for me. I manually set to 1.45 to see if thatll give it a more accurate value.

 

[tabascosauz]

It doesn't sound like for everyone who owns an MSI board that it has been fixed though is what I am saying. There are still those outliers for every vendor where its still an issue. Theres mega threads all over reddit since the first release of AGESA update that addresses the issue where people still report problems, and its not trending to one vendor or another.


In the BIOS to the side where it has RAM and CPU information, ram was at 1.47v for me. I manually set to 1.45 to see if thatll give it a more accurate value.

You mean the CHA A/B DRAM VOLTAGE reading in the sidebar? Yeah that one's overvolted by about +0.02V for me. So in summary, the number I enter in BIOS is what I go off of but it's not displayed anywhere else in BIOS or Windows; the BIOS sidebar number is +0.02V and doesn't show up anywhere else; and the software sensor is +0.01-0.06V depending on load

#justgigabytethings


In all seriousness, probably best to just treat it like any other board and only go off the value you personally punch into BIOS. It sounds like your board is probably +0.03V like most Gigabyte boards. My Unify-X is better than my Aorus in that the BIOS sidebar is correct, but the software sensor is still +0.01V.

 

[MxPhenom 216]

You mean the CHA A/B DRAM VOLTAGE reading in the sidebar? Yeah that one's overvolted by about +0.02V for me. So in summary, the number I enter is what I go off of but it's not displayed anywhere else in BIOS or Windows; the BIOS sidebar number is +0.02V and doesn't show up anywhere else; and the software sensor is +0.01-0.06V depending on load

#justgigabytethings


In all seriousness, probably best to just treat it like any other board and only go off the value you personally punch into BIOS. It sounds like your board is probably +0.03V like most Gigabyte boards. My Unify-X is better than my Aorus in that the BIOS sidebar is correct, but the software sensor is still +0.01V.

Indeed.

I just need to get in and dial down some of those subtimings a bit for this RAM and I think ill be set. I just noticed my tRFC is 630, which is fricking insanely loose.

@tabascosauz

Why in the timing section does it list my tRFC as 312. But when I go into windows and I open up Zentimings its actually showing as 630?

Is this because of Gear down mode being enabled?

 

[tabascosauz]

Indeed.

I just need to get in and dial down some of those subtimings a bit for this RAM and I think ill be set. I just noticed my tRFC is 630, which is fricking insanely loose.

@tabascosauz

Why in the timing section does it list my tRFC as 312. But when I go into windows and I open up Zentimings its actually showing as 630?

Is this because of Gear down mode being enabled?

No it's not got anything to do with GDM. Are you sure you didn't look at the wrong number? In Gigabyte BIOS the pesky column on the right I think is always there and shows the current and default values I think

Make sure you didn't accidentally put it in tRFC2 or tRFC4 leave those alone on Auto

 

[MxPhenom 216]

No it's not got anything to do with GDM. Are you sure you didn't look at the wrong number? In Gigabyte BIOS the pesky column on the right I think is always there and shows the current and default values I think

Make sure you didn't accidentally put it in tRFC2 or tRFC4 leave those alone on Auto

yeah idk. the columns show CH A CHB and said it should be 312, but hwinfo and Zentimings was reading 630. I just changed mine a bit. Between the ones you posted, and what the calculator suggested to me:

Is it possible to run Cmd rate at 1T without doing GDM out of curiosity? Im still clueless on what the hell gear down mode even does.

 

[tabascosauz]

yeah idk. the columns show CH A CHB and said it should be 312, but hwinfo and Zentimings was reading 630. I just changed mine a bit. Between the ones you posted, and what the calculator suggested to me:

Is it possible to run Cmd rate at 1T without doing GDM out of curiosity? Im still clueless on what the hell gear down mode even does.

View attachment 213864

Latency is really high? My 3600 14-14-14 profile is 56.5ns or so, about the same timings, nothing too tight.

Command rate is just a question of stability. 1T is faster than GDM is faster than 2T, but they are less stable/harder to stabilize in that order.

For pure 1T you can either brute force more VDIMM (not always possible) or do the cadbus trick (60-20-24-24 iirc dont quote me) (also not always possible).

Unless you're a benchmark addict, no point. Sometimes depending on CPU/board, 2T may be slightly faster or slower than GDM, the two of them are pretty close usually.

 

[MxPhenom 216]

Latency is really high? My 3600 14-14-14 profile is 56.5ns or so, about the same timings, nothing too tight.

Command rate is just a question of stability. 1T is faster than GDM is faster than 2T, but they are less stable/harder to stabilize in that order.

For pure 1T you can either brute force mote VDIMM (not always possible) or do the cadbus trick (60-20-24-24 iirc dont quote me) (also not always possible).

Unless you're a benchmark addict, no point. Sometimes depending on CPU/board, 2T may be slightly faster or slower than GDM, the two of them are pretty close usually.

I think the only differences between my profile and yours is the tRFC and GDM. Those 2 things for me might be able to get me below 60ns

 

[tabascosauz]

I think the only differences between my profile and yours is the tRFC and GDM. Those 2 things for me might be able to get me below 60ns

I run 140ns tRFC and I don't run GDM anymore just 2T. I don't expect 2T to have any latency penalty nowadays on Zen 3, and 20ns tRFC can probably account for like 0.5ns. tRFC makes a big difference but only really for big jumps (50+ns).

Is F2 the AGESA 1203b BIOS? I'm still convinced that Gigabyte fucked this release in various different ways, wouldn't be surprised if they fix it in the next release.

But your clock in AIDA is quite low. Run it a few more times, see if it changes. Benching AIDA on Zen 3 is a nightmare, v unpredictable latency

 

[MxPhenom 216]

I run 140ns tRFC and I don't run GDM anymore just 2T. I don't expect 2T to have any latency penalty nowadays on Zen 3, and 20ns tRFC can probably account for like 0.5ns. tRFC makes a big difference but only really for big jumps (50+ns).

Is F2 the AGESA 1203b BIOS? I'm still convinced that Gigabyte fucked this release in various different ways, wouldn't be surprised if they fix it in the next release.

Indeed it is. I think I am just going to leave it at this for now.

Maybe its DPC vs Daisy Chain since the unify x you have is DPC isnt it? But that may not be that big of a difference

 

[tabascosauz]

Indeed it is. I think I am just going to leave it at this for now.

Maybe its DPC vs Daisy Chain since the unify x you have is DPC isnt it? But that may not be that big of a difference

Nah performance difference is negligible between direct and daisy chain. Maybe 0.5ns, but even then it's impossible to tell if it's due to other board differences. Unify-X actually seems to be on the slower side when it comes to memory, think both my Aorus ITX boards were tighter (GB is generally pretty good on memory imo)

Maybe run a few times membench (in DRAM calc) to see if it's just AIDA acting up

 

[MxPhenom 216]

Nah performance difference is negligible between direct and daisy chain. Maybe 0.5ns, but even then it's impossible to tell if it's due to other board differences. Unify-X actually seems to be on the slower side when it comes to memory, think both my Aorus ITX boards were tighter (GB is generally pretty good on memory imo)

Maybe run a few times membench (in DRAM calc) to see if it's just AIDA acting up

Where is membench download? I cant find it anywhere.

 

[tabascosauz]

Where is membench download? I cant find it anywhere.

It's inside the DRAM calc app. Maybe like 3rd or 4th tab at the top. Just click run, let it do its thing and come back in a few minutes

 

[MxPhenom 216]

It's inside the DRAM calc app. Maybe like 3rd or 4th tab at the top. Just click run, let it do its thing and come back in a few minutes

Run 1:

Run 2:

 

[tabascosauz]

View attachment 213868

Seems reasonable enough. Maybe 140ns will do the trick.

 

[freeagent]

I run 140ns tRFC and I don't run GDM anymore just 2T.

I was actually just dabbling in that neighborhood today, but on the other side of the spectrum

 

[MxPhenom 216]

I was actually just dabbling in that neighborhood today, but on the other side of the spectrum

View attachment 213871

Is TestMem5 considered the standard for mem stability?

I haven't ran it yet at my settings, but i also havent had problem in anything.

 

[freeagent]

Is TestMem5 considered the standard for mem stability?

It seems to be fairly aggressive, I don't know if it is the standard though.. I just saw a bunch of guys running it when I got my first Zen chip. It seems to work fairly well.. Usually if that goes I don't have any problems unless they are IF/interconnect problems. I saw Tabascosauz liked the Anta Extreme profile, so I use it. I came here knowing nothing, and these guys did an awesome job coaching me. But I still know nothing

 

[tabascosauz]

Is TestMem5 considered the standard for mem stability?

I haven't ran it yet at my settings, but i also havent had problem in anything.

TM5 with the anta777 extreme1 .cfg is basically the standard yes. Some like using 1usmus v3 .cfg instead, I stick to anta. It's one of the quickest tests.

But you have to keep running it manually every 1h20m or so which makes it a bit of a pain. So I only use it for testing during the day. For overnight I use HCI Memtest Pro, paid $5 for it I think. A lot more convenient than the free HCI, though that still has all the functionality that matters.

Overall anta777 seems more aggressive than HCI. Therefore usually more effective.

But relying on TM5 alone is stupid, like any other test alone, regardless of how good it is. If you want daily stability, always run a bunch of different tests.

e.g. a few runs of TM5 anta + 1000-2000% HCI coverage + LinpackXtreme or Prime95 Large FFT

Even TM5 certainly isn't foolproof. It's better than HCI at detecting most timings, but is fucking terrible at evaluating tRFC on APUs. Vega iGPU are very sensitive to tRFC, so while it might pass any # of TM5 runs at 133ns it'll instantly bluescreen a game.