Memory training, or something else? (No BIOS, no display, plus more!)

[DanteCoal]

SOLVED!

Removed the Thermalright CPU bracket and put on the old stock retention bracket, booted with 2 DIMMs in B1 B2, and the system fired right up with no issues.  Updated drivers, checked the BIOS, all good.  Installed the other 2 DIMMs into A1 A2, and system took a good extra 30 seconds to fire up, but fire up it did.  Currently loading my files back onto the PC before I shut it down and cable manage, put panels on, etc.  In the end I can't imagine it was the bracket that caused the issue since I'd been using it on my MSI board with no issues, torqued down to the same spec and everything, but it's all working now, and since I can't see the bracket with my cooler on, I don't care anyway, lol.

Original Issue - Everything existing is known good, been using it all for a bit, but with an MSI MoBo that I haven't loved, so new MoBo.  Got everything installed, went to turn system on, GPU led's flashed green, RAM lit up, MoBo's RAM light lit up, that's it.  Let is sit for about 10 mins, nothing.  Turned it off, cleared CMOS (held onboard button for 10s sec), used the Q-Flash Plus feature to update BIOS, system reset itself once done, All other LEDs lit up, display came up with GIgabyte logo for half a second, then to a windows error that the system failed to load.  It reset itself.  Back to just the RAM and RAM indicator lights being on.  After 5 mins the rest of the lights came on again, still no display after about 30 mins.  At no point have the mouse of KB lit up or shown any sign of activity.

Steps taken (before solution) - Updated BIOS, reset CMOS, checked all connections, power cycled, reseated CPU, reseated RAM

System:
NEW Parts-

MoBo - GIgabyte X670 Gaming X AX V2 (Currently on latest BIOS version)

Existing (Known Good) Parts- 

CPU - Ryzen 5 7600X (w/ Thermalright bracket toqued to 0.6nm / 5in/lb) (w/ EKWB Block)
RAM - 4x16GB G.Skill Trident (F5-6000J3038F16GX2-TZ5N Verified compatible with new MoBo)
GPU - PowerColor Red Devil AMD Radeon 6950XT (w/ EKWB Block + backplate)
Storage - HP FX900 Pro M.2 (Existing Win10 OS on it, ready to re-install if I can get to/through BIOS)
PSU - EVGA Supernova G6 1000W (w/ Custom cables by Alchemy)
Cooling - Custom loop.  D5 Pump/res > 360 Rad > CPU > 360 Rad > GPU > Res.  9x Thermalright ARGB fans.
Display - KTC 32" Curved over DP using provided cable.
KB + Mouse - GMMK Pro / Model O.

Edited May 18 by DanteCoal
Solution found

[RONOTHAN##]

17 minutes ago, DanteCoal said:

RAM - 4x16GB G.Skill Trident (F5-6000J3038F16GX2-TZ5N Verified compatible with new MoBo)

Remove two sticks and report back. AM5 is kinda picky when it comes to 4x16GB compatibility, and while it can work, you usually need to be on the right BIOS revision for it to boot and function correctly. If that does fix everything, go through and try a few different revisions until you find one that actually works with 4x16GB. 

 

17 minutes ago, DanteCoal said:

(w/ Thermalright bracket toqued to 0.6nm / 5in/lb)

Remove this, it does nothing for temps (AM5 doesn't bend CPUs like LGA 1700 does) and they have a tendency to cause memory issues. 

[PDifolco]

38 minutes ago, DanteCoal said:

Issue - Everything existing is known good, been using it all for a bit, but with an MSI MoBo that I haven't loved, so new MoBo.  Got everything installed, went to turn system on, GPU led's flashed green, RAM lit up, MoBo's RAM light lit up, that's it.  Let is sit for about 10 mins, nothing.  Turned it off, cleared CMOS (held onboard button for 10s sec), used the Q-Flash Plus feature to update BIOS, system reset itself once done, All other LEDs lit up, display came up with GIgabyte logo for half a second, then to a windows error that the system failed to load.  It reset itself.  Back to just the RAM and RAM indicator lights being on.  After 5 mins the rest of the lights came on again, still no display after about 30 mins.  At no point have the mouse of KB lit up or shown any sign of activity.

Steps taken - Updated BIOS, reset CMOS, checked all connections, power cycled, reseated CPU, reseated RAM

System:
NEW Parts-

MoBo - GIgabyte X670 Gaming X AX V2 (Currently on latest BIOS version)

Existing (Known Good) Parts- 

CPU - Ryzen 5 7600X (w/ Thermalright bracket toqued to 0.6nm / 5in/lb) (w/ EKWB Block)
RAM - 4x16GB G.Skill Trident (F5-6000J3038F16GX2-TZ5N Verified compatible with new MoBo)
GPU - PowerColor Red Devil AMD Radeon 6950XT (w/ EKWB Block + backplate)
Storage - HP FX900 Pro M.2 (Existing Win10 OS on it, ready to re-install if I can get to/through BIOS)
PSU - EVGA Supernova G6 1000W (w/ Custom cables by Alchemy)
Cooling - Custom loop.  D5 Pump/res > 360 Rad > CPU > 360 Rad > GPU > Res.  9x Thermalright ARGB fans.
Display - KTC 32" Curved over DP using provided cable.
KB + Mouse - GMMK Pro / Model O.

In addition to what @RONOTHAN## suggests, try to post with no drive in the board

And why did you swap mobo for no gain I can think of if it worked ??

[DanteCoal]

6 minutes ago, PDifolco said:

In addition to what @RONOTHAN## suggests, try to post with no drive in the board

And why did you swap mobo for no gain I can think of if it worked ??

Currently in process of disassembling my loop to remove the TT bracket, so I'll update when that's done and I've got a chance to power things up again with just A1 and A2 DIMM slots filled.

To answer the MoBo swap question; my USB 3.2 headers for my front I/O stopped working, and the ARGB headers on the MSI board never worked properly to control my devices. (Not a huge issue to me, so I accepted that one after a day or two of tinkering).  Troubleshot the board and front I/O, found that the I/O was good, but the MoBo wouldn't work with any USB 3.2 headers.  Not to mention that the USB-C header on the MoBo didn't work.  Had done several BIOS revisions, all drivers updated, and every trick MSI told me (all of which I'd already done, but it can't hurt), and nothing got them working.  So I'm in the process of an RMA for the board that I'll keep as a spare, and now trying to get this board working in general.  MSI board in question was a "Pro x670-P WiFi"

[RONOTHAN##]

4 minutes ago, DanteCoal said:

with just A1 and A2 DIMM slots filled.

Make sure it's A2 and B2 as they're the optimized slots for high speed memory. A1 and A2 will have the same issues as you're currently experiencing. 

[OddOod]

4x RAM is not advised with DDR5. It's unlikely to improve until new memory managers come with new chips

[DanteCoal]

48 minutes ago, OddOod said:

4x RAM is not advised with DDR5. It's unlikely to improve until new memory managers come with new chips

Huh, wasn't aware that 4x RAM isn't good with DDR5; any quick reason as to why that is, or a link to something reputable I could study up on?  If I'm not getting any real world benefit from my 4x16 sticks and running 2x16 would benefit my system, I might just ditch the two sticks.

Oh, and my problem Is fixed, I'll be editing main post with the details basically copy/pasted from here;

Removed the Thermalright CPU bracket and put on the old stock retention bracket, booted with 2 DIMMs in B1 B2, and the system fired right up with no issues.  Updated drivers, checked the BIOS, all good.  Installed the other 2 DIMMs into A1 A2, and system took a good extra 30 seconds to fire up, but fire up it did.  Currently loading my files back onto the PC before I shut it down and cable manage, put panels on, etc.  In the end I can't imagine it was the bracket that caused the issue since I'd been using it on my MSI board with no issues, torqued down to the same spec and everything, but it's all working now, and since I can't see the bracket with my cooler on, I don't care anyway, lol.

[RONOTHAN##]

4 hours ago, DanteCoal said:

In the end I can't imagine it was the bracket that caused the issue since I'd been using it on my MSI board with no issues, torqued down to the same spec and everything

Do remember it's a different motherboard, and different motherboards have different sensitivities to mounting pressure. On LGA 1700, Gigabyte boards don't really care, while ASUS boards do (in general, it can vary from board to board), so it's very possible that the MSI board wasn't picky but this Gigabyte board is. 

[OddOod]

4 hours ago, DanteCoal said:

any quick reason as to why that is, or a link to something reputable

Honestly? no. I can google it up tomorrow, but I feel like I snatched it out of the zeitgeist. It's kinda to be expected for bleeding edge first gen. I remember thee first *few* chip gens. Zen 1 and 1+ were super picky about RAM speeds. 
From what I understand 4 sticks at stock speeds (no XMP/DOCP) is in spec and supported. But that's at low speeds and loose timings

Again, I cannot currently backup my claims, but this is my understanding and I'd be utterly stoked to be proven wrong

 

[DanteCoal]

18 minutes ago, OddOod said:

Honestly? no. I can google it up tomorrow, but I feel like I snatched it out of the zeitgeist. It's kinda to be expected for bleeding edge first gen. I remember thee first *few* chip gens. Zen 1 and 1+ were super picky about RAM speeds. 
From what I understand 4 sticks at stock speeds (no XMP/DOCP) is in spec and supported. But that's at low speeds and loose timings

Again, I cannot currently backup my claims, but this is my understanding and I'd be utterly stoked to be proven wrong

 

So, I did some looking around.  It seems that it's not only still an issue, but it's still a bleeding edge issue, and one that probably won't be solved until the next gen of CPUs.  If what I've read is correct, it's a combination of factors.  Current CPUs run in a dual split channel, and the speeds of DDR5 ram overflow what they're capable of.  Also an issue of timing compatibility, EXPO functionality, RAM sub-channels, and the overwhelming concensus is that while having more sticks can give you more RAM, your RAM speeds will suffer in a 4 slot setup.  So 4x is bad for gamers and causal users, but good for workhorse setups that need the extra GBs.

Also seems like this is an issue with Intel as well, since it's overall a DDR5 issue of just being too ahead of the current CPU curve.

As it stands though for my PC, I'm gonna keep the 4x16 and see how it plays out.  Done a few reboots now, and it goes from off to windows in just a few seconds now.  I'll see how my games and daily use work out, and hopefully it won't be an issue.  Wasn't one on the old mobo, but as @RONOTHAN## said, it could just be a difference in mobos.

[RONOTHAN##]

19 minutes ago, OddOod said:

I can google it up tomorrow,

I gotchu:

 

5 hours ago, DanteCoal said:

any quick reason as to why that is, or a link to something reputable I could study up on?

As for the general reasoning, it just comes down to how memory traces are routed. There's two main types: daisy chain and T-topology, where the daisy chain topologies allows for higher 1DPC (DIMM per channel) memory speeds but due to the slightly different lengths between slots 1 and 2 on the daisy chain they clock memory worse in 2DPC configs, and T-topology is the opposite where the 2DPC configs clock higher due to the equal length traces but clock worse in 1DPC configs due to the extra wire giving signal reflection issues and other interference. 

 

On DDR3, the daisy chain clocking advantages didn't really matter as memory didn't clock that high to begin with, so most boards were T-topology. On DDR4, it started to matter more, where the T-topology boards clocked 2DPC configs 200-400MT/s higher but clocked 1DPC configs 200-400MT/s lower as well. On DDR5, however, because the speeds are so much higher, daisy chain boards are more or less required (they're also much simpler to design, so early boards are almost always daisy chains), and the difference in 2DPC and 1DPC configs, both with dual rank memory, is anywhere between 400 and 1000MT/s depending on the BIOS and the specific board. 

[PDifolco]

1 hour ago, RONOTHAN## said:

I gotchu:

 

As for the general reasoning, it just comes down to how memory traces are routed. There's two main types: daisy chain and T-topology, where the daisy chain topologies allows for higher 1DPC (DIMM per channel) memory speeds but due to the slightly different lengths between slots 1 and 2 on the daisy chain they clock memory worse in 2DPC configs, and T-topology is the opposite where the 2DPC configs clock higher due to the equal length traces but clock worse in 1DPC configs due to the extra wire giving signal reflection issues and other interference. 

 

On DDR3, the daisy chain clocking advantages didn't really matter as memory didn't clock that high to begin with, so most boards were T-topology. On DDR4, it started to matter more, where the T-topology boards clocked 2DPC configs 200-400MT/s higher but clocked 1DPC configs 200-400MT/s lower as well. On DDR5, however, because the speeds are so much higher, daisy chain boards are more or less required (they're also much simpler to design, so early boards are almost always daisy chains), and the difference in 2DPC and 1DPC configs, both with dual rank memory, is anywhere between 400 and 1000MT/s depending on the BIOS and the specific board. 

About DDR5 shennanigans, my 7950X3D/X670E Asus Strix always take like 30sec to "train" RAM before posting, at each boot - I'm sure it's that because I have the corresponding postcode

I've 2x32GB GSkill Trident Z Neo, 6000CL30, Hynix M die, what is usually seen (by Buildzoid himself) as the best AM5 choice

Found there's a Memory Context Restore BIOS option that allows to have "sticky" RAM settings and make post instant, it worked, but also some guys don't recommend it because it could be unoptimal or even lead to issues, any insights on this ?

 

[RONOTHAN##]

6 hours ago, PDifolco said:

Found there's a Memory Context Restore BIOS option that allows to have "sticky" RAM settings and make post instant, it worked, but also some guys don't recommend it because it could be unoptimal or even lead to issues, any insights on this ?

There are BIOS revisions where it works, and there's BIOS revisions where it doesn't. If it's working for you, keep running it, I'd only turn it off if you occasionally have issues where the system doesn't turn back on.

 

6 hours ago, PDifolco said:

Hynix M die, what is usually seen (by Buildzoid himself) as the best AM5 choice

Slight astrick to this, M die is 2nd best on AM5, since it does most of the same timings as M die, but it tends to have the tRP and tRFC1 go lower than M die does. Granted, it's barely a difference, but it's still technically a difference. 

 

6 hours ago, PDifolco said:

my 7950X3D/X670E Asus Strix always take like 30sec to "train" RAM before posting, at each boot

Honestly, that's rather good. Both of the AM5 boards I've used with dual rank memory took over a minute to train whenever they did.

[GuiltySpark_]

I would be beside myself if I had to endure memory training at every boot. 

 

Frankly, I think many people are attributing the normal longer boot times to memory training when its not. On my board code 15 on the display indicates training and I ONLY see that after making BIOS changes. Normal bootups and restarts its skipped. (X670E Taichi) 

 

Unfortunately with so many boards not having proper 7-segment displays, its not always obvious what is actually happening and some are simply thinking long bootup = memory training all the time, when that isn't the case. 

 

When my system does train, its about an extra 20 seconds or so.

[PDifolco]

1 hour ago, GuiltySpark_ said:

I would be beside myself if I had to endure memory training at every boot. 

 

Frankly, I think many people are attributing the normal longer boot times to memory training when its not. On my board code 15 on the display indicates training and I ONLY see that after making BIOS changes. Normal bootups and restarts its skipped. (X670E Taichi) 

 

Unfortunately with so many boards not having proper 7-segment displays, its not always obvious what is actually happening and some are simply thinking long bootup = memory training all the time, when that isn't the case. 

 

When my system does train, its about an extra 20 seconds or so.

My BIOS always shows code 15 for the 30sec it takes to post, minus the couple final seconds, so it's definitely memory training

1 hour ago, RONOTHAN## said:

There are BIOS revisions where it works, and there's BIOS revisions where it doesn't. If it's working for you, keep running it, I'd only turn it off if you occasionally have issues where the system doesn't turn back on.

 

Slight astrick to this, M die is 2nd best on AM5, since it does most of the same timings as M die, but it tends to have the tRP and tRFC1 go lower than M die does. Granted, it's barely a difference, but it's still technically a difference. 

 

Honestly, that's rather good. Both of the AM5 boards I've used with dual rank memory took over a minute to train whenever they did.

Crystal clear, I'd put MCR on the, thanks !

[GuiltySpark_]

3 minutes ago, PDifolco said:

My BIOS always shows code 15 for the 30sec it takes to post, minus the couple final seconds, so it's definitely memory training

Something isn't working as intended then. 

[PDifolco]

9 minutes ago, GuiltySpark_ said:

Something isn't working as intended then. 

Why ? 15 is the code for RAM training and AM55 loves to always train RAM

[GuiltySpark_]

3 minutes ago, PDifolco said:

Why ? 15 is the code for RAM training and AM55 loves to always train RAM

Yes, on first boot and after something is changed. If properly configured that shouldn't be happening at every boot. Just like any other system in the past. AM5 (why do we call this by its socket again?) got this reputation as it was essentially broken for months and still is for people. 

[RONOTHAN##]

Just now, GuiltySpark_ said:

Yes, on first boot and after something is changed. If properly configured that shouldn't be happening at every boot. 

ASUS boards leave MCR disabled by default, and IIRC there's a couple BIOS revisions where it just doesn't work at all. 

[PDifolco]

22 minutes ago, RONOTHAN## said:

ASUS boards leave MCR disabled by default, and IIRC there's a couple BIOS revisions where it just doesn't work at all. 

Indeed my MCR is off by default, I'd go enable it after dinner :)

[GuiltySpark_]

25 minutes ago, RONOTHAN## said:

ASUS boards leave MCR disabled by default, and IIRC there's a couple BIOS revisions where it just doesn't work at all. 

This would infuriate me to the point of returning the board. Thankfully its not been a problem on my Asrock board.