AM5 AGESA 1.0.0.7b

[AusWolf]

Thats also what I noticed and questioned, the article doesnt say anything about that. I thought maybe a bios version but that doesnt seem to be the case.

I don't remember such versions. There was 1.60 and a beta for 1.65 at some point.

 

[chrcoluk]

@Mussels This is Zen1 all over again with DDR5 instead. Give it another generation and I'm sure all these oddites will dry up.

AMD isn't as nice to me when I swap ram without disabling XMP/EXPO first. Usually it will refuse to boot ever again unless I clear the CMOS. Kinda a hassle for me since I'm swapping things daily. Intel doesn't seem to care what I swap to. Either it will try the XMP1 profile or default to JEDEC, either way it boots regardless.

Yeah thats odd, I would expect a bios to auto recover to jedec after X post failures. You think its something ageasa side that prevents this? As I would assume its something that could be just copied over from intel boards.

On my Z370 I can configure the post attempts before auto recovery, I set it to 2, as in my opinion even 1 retry isnt good enough and suggests instability, on my Z690 board its just a on or off, and it annoyingly tries way too many times, but will eventually post to jedec.

 

[Konyalee]

Has anyone tried running 4x16GB 6000Mhz C30 RAM after new update, they claim to fix compability issues now.

I'm on 1.0.0.7a on Crosshair X670E Hero and can't get my board to boot into Windows when any EXPO profile is set - it simply boots to BIOS in safe mode. Running at stock 4800MHz is the only way I can run things right now.

Going to check and see if this update is on ASUS as well.

ETA: Just checked. Still 1415.

I have the same board, when I flashed bios it reset all of my setting so if you have any settings boot device related might be worth to check it out.

 

[Mussels]

Have a question to some of you here having a Asrock board. I saw this test review, where it seems Asrock somehow uses some way slower RAM timings, even if you use the same RAM kits, resulting in a 10% performance loss in gaming compared to other boards from MSI or Gigabyte. How can that be? Anyone can confirm that? Maybe thats also the reason Asrock boards seem to be more "stable" when it comes to "boot or not to boot" with 6000+ kits.

The article says:

"The Asrock boards were hit and miss, oddly the Riptide was good at the 6000 speed, but the Lightning and Pro RS both failed, forcing us to run at 5600. The Lightning and Pro RS were both tested with the BIOS version "1.11.AS03" which was the latest version, the Riptide though has seen two more updates and the latest 1.14.AS15 worked just fine with DDR5-6000. So presumably Asrock will eventually provide that BIOS revision to the rest of their B650 lineup, solving the memory issues."

So not sure if that is still true, for latest AGESA or not.

AMD B650 Motherboard Roundup: VRM Thermal Testing

For those wanting to jump on the AM5 platform while spending as little as possible, which affordable AMD B650 motherboards are the best? We'll be torture-testing eleven...

www.techspot.com

I was thinking of getting a ASRock B650M PRO RS WIFI board with a G.Skill Flare EXPO X5 DDR5-6000CL32-38-38-96 kit for a new Ryzen 7800X3D build.

Look at ASUS who happily threw 1.3v+ at the SoC on AM5 to get ram overclocks stable, or MSI who ran the MCLK at 850 instead of the 1800 everyone else did with DDR5 6000 (from one of those youtube videos, GN or HUB)

SoC voltage and secondary/tertiary timings do not come from the RAM - that's upto the board makers to choose and implement in their BIOS. Looser timings means more support such as running 4x sticks of the RAM, or future models with looser timings but lower performance, and this is the situation where they have to find a compromise that works best for their hardware.

I'm sure AMD says to use the JEDEC standard timings and then tune as they see fit, and they all did their testing with the early DDR5 modules and did some basic math, scaling the timings around... and then manufacturers went nuts with DDR5 7200 CL32 before anything could damn well use it.


Honestly sometimes it's like they throw a change at these by commitee and hope for the best
"SoC voltage helps ram OC, so lets push it 0.1v higher for every 100MT/s they go up and not actually test that for safety on any boards"
(oops, this went poorly on boards without any hardware to cap the voltages)
"Loosening these timings helped get better stability at DDR5 6000, so let's just loosen them for everything"



Be nice if they just had a compatibility and performance mode, defaulting to compatibility with looser secondary/tertiary timings

 

[ir_cow]

This comes from ASUS BIOS modder not me. ASUS is really the only one who kinda tricks people into choosing XMP 1 because it reads like it would be the main XMP profile. Not the case. Better tweaked settings, but with AMD, that can lead to stability issues quickly.

XMP I = asus optimized XMP profile, it only load tCL, tRCD, tRP and tRAS from original XMP profile, won’t load another timings from XMP profile. in addition it will optimized some timings by auto for best performance.
XMP II = manufacturer XMP profile, will load tCL, tRCD, tRP, tRAS, tREF1/2 and tRFCsb timings from XMP profile, and addition auto tuning some timings for better performance.
XMP tweaked = almost same as XMP I

 

[Mussels]

I recall reading on that, part of the scandal that got overshadowed by the exploding CPU's is that ASUS EXPO/XMP was not actual EXPO/XMP, but a custom overclocking profile on their behalf

 

[mkdr]

Yeah, what's 162 and 163 next to the MSi boards?

I found out in a comment on the review article, that 162 and 163 indeed seem to refer to different bios versions. Yet I looked up at the MSI site, and none of the bios updates had any version with that number at all. wonder if the latest Asrock bios versions also fixed that performance regression seen in the tests, where a newer MSI bios fixed it. Not sure right now if I should buy the MSI Pro B650-P WIFI or the Asrock B650M PRO RS WIFI.

 

[Mussels]

I know asus release BETA versions and delete them when the final comes out - they also delete a lot of finals, if they find issues with them.
Asrock/MSI etc could be doing the same.

 

[BES]

So, I just updated to the ComboAM5 1.0.0.7b why is it now memory training forever on every boot? ..in the bios it's already set to not memory train once it's done it before, but it keeps doing it anyway on every restart, it's the correct ram from the motherboard support list.. in XMP profile..

I didn't really see in this thread a resolution to this, it took over 10 minutes just to boot the PC into safe mode, then booting it back into regular mode because it insists on training EVERY TIME..the previous bios it just booted in like 10 seconds after it train that initial first time...

 

[TheLostSwede]

So, I just updated to the ComboAM5 1.0.0.7b why is it now memory training forever on every boot? ..in the bios it's already set to not memory train once it's done it before, but it keeps doing it anyway on every restart, it's the correct ram from the motherboard support list.. in XMP profile..

I didn't really see in this thread a resolution to this, it took over 10 minutes just to boot the PC into safe mode, then booting it back into regular mode because it insists on training EVERY TIME..the previous bios it just booted in like 10 seconds after it train that initial first time...

1.0.0.7c seems to fix that.

 

[BES]

1.0.0.7c seems to fix that.

Well, I think im just going to revert to the old one for now because of this... a simple restart shouldn't train every time..

 

[TheLostSwede]

Well, I think im just going to revert to the old one for now because of this... a simple restart shouldn't train every time..

I'm not having any issues with that on 1.0.0.7c.

As you can see, I'm running some rather high speed modules (for AMD) and tight timings, XMP profily only, non QVL RAM.

 

[BES]

I'm not having any issues with that on 1.0.0.7c.

As you can see, I'm running some rather high speed modules (for AMD) and tight timings, XMP profily only, non QVL RAM.

View attachment 310766

I have an ASRock X670E Pro RS, not seeing that BIOS version, so I think ill revert for now until they add it..

 

[TheLostSwede]

I have an ASRock X670E Pro RS, not seeing that BIOS version, so I think ill revert for now until they add it..

Could be ASRock hasn't relased the 1.0.0.7c AGESA yet.

 

[Deleted member 232534]

asrock never released the 1.0.0.7c

 

[BES]

Yeah it's annoying, the motherboard has been working perfectly, even with EXPO ram not on the support list, until this last Bios where the EXPO ram wouldn't work at all with any profile, unless I used the default ram speed and then I switched RAM to a supported one, then it's doing the training every single reboot. Before with the unsupported EXPO ram it was booting in 10 seconds or less after it trained..

 

[Deleted member 232534]

Yeah it's annoying, the motherboard has been working perfectly, even with EXPO ram not on the support list, until this last Bios where the EXPO ram wouldn't work at all with any profile, unless I used the default ram speed and then I switched RAM to a supported one, then it's doing the training every single reboot. Before with the unsupported EXPO ram it was booting in 10 seconds or less after it trained..

you have the "memory context restore" set to 'enable' or 'disable'?

 

[Steevo]

What's your experiences so far with 1007 so far? A friend put together a 7950X3D setup a few days ago with a 64GB 6000CL32 Hynix kit. Could only boot 5600, even at 1.25V SOC. No 1007b BIOS available yet for his X670E Strix-A.

Also, man...they weren't kidding about long POST times.

I installed and have booted 6400 at 32 36 36 38 but in gear 2, at 1.2v SOC only, memory at 1.45 forcing gear 1 at 6000 and 28 32 32 32 will still cause occasional BSOD and a hour of running men tests it fails. I booted to 6800 with default 6400 SK Hynix timings provided in 1.0.07c Agesa in the newest BIOS but what I gained in read and write cost me 6ns latency and I haven’t finished going higher on memory speed to see where that wall is and tuning timings. I am hoping to be able to dedicate a few more hours today then doing some FPS and latency tests. But it overall seems that 6000-6400 with tight timings is still the best option even with X3D chips.

If my system takes more than 45 seconds for memory training I can already tell it will fail testing or the IMC and DDR5 are having issues and latency and performance is going to suffer.

 

[BES]

Actually I figured out the issue with the ComboAM5 1.0.0.7b, it wasn't the ram or the ram profile, it was the processor profile I set, it seems, now it's working fine, I even restarted it a few times to test it and it's now booting up instantly with a PBO 85C -20mV overclock profile with EXPO enabled: https://valid.x86.fr/yctdxm .. only weird thing im seeing now is two ram sticks showing up as quad channel in CPU-Z ..lol HWinfo64 says it's dual channel, so must be a CPU-Z issue

 

[R-T-B]

Has anyone tried running 4x16GB 6000Mhz C30

Been running my setup with that ram type across all updates. I have to tune some voltages but it works.

only weird thing im seeing now is two ram sticks showing up as quad channel in CPU-Z ..lol

That's normal. It's because of how DDR5 channels are setup.

 

[mkdr]

Really annoying that AsRock doesnt seem to release 1.0.0.7c, it seems they are waiting for 1.0.0.8 which sadly will contain the Inception mitigation microcode. I dont want 1.0.0.8 I want 1.0.0.7c.

 

[Panther_Seraphin]

Actually I figured out the issue with the ComboAM5 1.0.0.7b, it wasn't the ram or the ram profile, it was the processor profile I set, it seems, now it's working fine, I even restarted it a few times to test it and it's now booting up instantly with a PBO 85C -20mV overclock profile with EXPO enabled: https://valid.x86.fr/yctdxm .. only weird thing im seeing now is two ram sticks showing up as quad channel in CPU-Z ..lol HWinfo64 says it's dual channel, so must be a CPU-Z issue

DDR5 is a dual channel per DIMM

In DDR1-4 you could have 2 DIMMs per Memory Channel and that memory channel is a single 64bit bus
In DDR5 you can have 2 DIMMs per Memory channel but that memory channel is two 32bit bus hence "quad channel" memory

 

[Mussels]

So, I just updated to the ComboAM5 1.0.0.7b why is it now memory training forever on every boot? ..in the bios it's already set to not memory train once it's done it before, but it keeps doing it anyway on every restart, it's the correct ram from the motherboard support list.. in XMP profile..

I didn't really see in this thread a resolution to this, it took over 10 minutes just to boot the PC into safe mode, then booting it back into regular mode because it insists on training EVERY TIME..the previous bios it just booted in like 10 seconds after it train that initial first time...

If the new update doesnt help (c) make sure you manually set things - a lot of people with slow boots are simply because the boards dont try to run high voltages as fast like they did in the past.

(It was something very vaguely along the lines of every failed boot add 0.5v to the SoC until it worked - but now they go up in smaller increments so it takes longer)
Manually setting voltages helps some people, memory training isn't what people are seeing, you're seeing failed boots.


"fast boot' tries to use the saved settings from the previous boot, (voltages, timings etc it's using) but can fail if something was borderline like the PSU standby voltage dropped over time as it got cold - or it can lose power entirely from being unplugged and need to do a cold boot

I'm a fan of getting every system working at cold boots every time, or it doesn't feel stable to me. My x370 board is like this where it can take 3-5 POST cycles to run at DDR4 3200 (If it passes 5, it resets to 2133 silently), but 2933 works every single time and boots fast every time.

 

[mkdr]

Maybe a tip which could be useful to some people (with Asrock boards?):

I got a new build last week, with an Asrock B650M Pro RS WiFi (bios 1.28, AGESA 1.0.0.7b) + 7800x3d + G.Skill 6000 cl32 32gb (2x16) sk hynix A-die kit. I noticed, that the PC became way more stable, if I lowered the following voltage settings, which were set by the EXPO profile:

SOC 1.30V (EXPO) => 1.15V
VDDP 1.15V (EXPO) => auto (0.95V)

Especially lowering VDDP from 1.15V to auto, which sets 0.95V, helped a lot. Not sure why, but this build seems to be way more happy with lower voltage settings, especially for VDDP.

I even lowered afterwards following settings down and thy system seems still be stable:

VDD 1.35V (EXPO) => 1.25V
VDDQ 1.35V (EXPO) => 1.25V
VDDIO (memory controller) 1.35V (EXPO) => 1.25V

I left the RAM primary as they came, 32 38 38, but used Buildzoids timings for SKH 6000 with infinity fabric at 2033MHz. Very important: had to leave DDR power down mode on auto, disabling froze the PC also in bios.

and I set a curve optimizer of all negative -30, and reduced PPT to 75W. I get 18500 in Cinebench R23 with these settings, and PC seems to be stable, also in OCCT cpu + memory SSE tests, where I got errors before when VDDP was still on 1.15.

I have super fast boot times, around 10 seconds or so to Windows 11 desktop (with fast startup disabled).

 

[Mussels]

I did some googling to try and find a simple description of what memory training actually is - there is complex steps involved but the summary is that it's really just measuring resistance to gauge the length of the wiring involved, so it can compensate for timing and voltages.
Things like one slot being further than the other it'll know to delay things so they match up, or send slightly more voltage to counter resistance.


Because the few good descriptions are like reading a technical manual written by Tolkien, people think every slow or failed POST is memory training - usually it's just a black screen while it reboots a few times before loading defaults, or on the newer boards kicking up specific voltages like SoC. (boards with diagnostic LED's make this really clear as you can see each cycle)