AMD Readying AGESA 1.0.0.7c for AM5 Motherboards

[TheLostSwede]

According to a post by @g01d3nm4ng0 on Twitter/X, we now know that AMD is readying yet another AGESA update for AM5 motherboards. The new version is, based on information from our own sources, a minor update to the current version. As such, AMD will be moving from 1.0.0.7b to 1.0.0.7c. @g01d3nm4ng0 didn't reveal any details of the new AGESA apart from the screenshot below, but we asked around and managed to find out what the new AESA addresses.

The update is specifically for those with Samsung DDR5 memory in their AM5 motherboards and it addresses multiple memory related stability issues. We weren't given the full details as to what those are, but there have been some reports about there being issues specifically with Samsung DDR5 memory in some AM5 boards and hopefully this will solve all those problems. We don't have a release time frame for the updated AGESA, but with 1.0.0.7b barely out the door, it might take a few weeks before this one makes it through all the internal testing at the motherboard makers.



View at TechPowerUp Main Site | Source

 

[bug]

As usual, if your system is already stable, don't mess with it.

 

[trparky]

As usual, if your system is already stable, don't mess with it.

Unless, of course, you need to update your BIOS because of some kind of vulnerability. Then, you're kind of forced to update.

The update is specifically for those with Samsung DDR5 memory in their AM5 motherboards and it addresses multiple memory related stability issues.

But why? What is Samsung doing with their DDR5 memory as versus the other manufacturers? What is Samsung doing wrong? Are they not holding themselves to the DDR5 specifications?

 

[ir_cow]

But why? What is Samsung doing with their DDR5 memory as versus the other manufacturers? What is Samsung doing wrong? Are they not holding themselves to the DDR5 specifications?

Multiple here reported Samsung memory not working correctly on their ASUS MBs. The only problem I have with the only Samsung kit I have, is that going above DRAM 1.4V causes the system not to boot. Kinda odd, but honestly not a big deal for me. Its above the EXPO profile, so really you are overclocking beyond what the manufacture sold it as.

 

[trparky]

Multiple here reported Samsung memory not working correctly on their ASUS MBs.

But what's the root cause? Is it AMD or Samsung at fault here? I'd have to put the blame on Samsung if you ask me since they're being singled out here.

 

[ir_cow]

But what's the root cause? Is it AMD or Samsung at fault here?

I don't know. it seems like AMD hasn't implemented Samsung ICs correctly. Remember this is AMD first DDR5 platform. Samsung memory has no issues on Intel that I am aware of.

 

[trparky]

I don't know. it seems like AMD hasn't implemented Samsung ICs correctly

But I'm going back to the statement... Where did Samsung go wrong? Did they not adhere to the DDR5 specification like they should have whereas the other manufacturers did and thus they had no issues? I thought that was the whole point about having standards. The industry creates a standard and a specification, the manufacturers build products based upon those standards and specifications, and if they run astray, then yeah... you're going to have issues.

 

[Vya Domus]

it seems like AMD hasn't implemented Samsung ICs correctly.

What does that even mean ? AMD isn't making the RAM kits nor the ICs used in them, so implement what ?

 

[trparky]

What does that even mean ? AMD isn't making the RAM kits, implement what ?

That's what I mean.

If you ask me, AMD shouldn't be fixing the issue. If the issue is with Samsung memory while other manufacturers aren't having issues, then the problem is Samsung's problem, and they should be forced to recall any products that have that issue.

Again, I cite standards and specifications. If AMD is adhering to the standards and specifications while Samsung isn't, then Samsung should be the one that's forced to fix things.

 

[ir_cow]

What does that even mean ? AMD isn't making the RAM kits nor the ICs used in them, so implement what ?

All timings that are not in the SPD data is generated by the motherboard. If MB manufacturers are just copying/pasting whatever AMD has specified, there is a good chance it may not work in all configurations. It could be something like the tWTRS doesn't scale linearly, or maybe RttWr, RtPark is the wrong values. Maybe even ProcODt. There is a ton of settings that is set by the motherboard. I think the guidelines set by AMD for Samsung memory was incorrect for certain configurations. That is my thinking. Otherwise it wouldn't be the AESA update. Instead, individual motherboards / brands.

 

[trparky]

All timings that are not in the SPD data is generated by the motherboard. If MB manufacturers are just copying/pasting whatever AMD has specified, there is a good chance it may not work in all configurations. It could be something like the tWTRS doesn't scale linearly, or maybe RttWr, RtPark is the wrong values. Maybe even ProcODt. There is a ton of settings that is set by the motherboard. I think the guidelines set by AMD for Samsung memory was incorrect for certain configurations. That is my thinking. Otherwise it wouldn't be the AESA update. Instead, individual motherboards / brands.

OK, but I argue that no matter what memory kit you buy, it should simply work out of the box at base DDR5 specifications. No EXPO, no DOCP, no XMP, no overclocking of any sort.

Right?

 

[ir_cow]

OK, but I argue that no matter what memory kit you buy, it should simply work out of the box at base DDR5 specifications. No EXPO, no DOCP, no XMP, no overclocking of any sort.

Right?

Yeah Samsung ICs using JEDEC specs does work

 

[xSneak]

I still have problems getting my msi x670e carbon to boot with 2 x 32GB sticks of hynix memory at 6000 mhz on the beta 1.0.0.7b bios. This platform is a mess and I wish I would have gone with intel instead.

 

[ir_cow]

I still have problems getting my msi x670e carbon to boot with 2 x 32GB sticks of hynix memory at 6000 mhz on the beta 1.0.0.7b bios. This platform is a mess and I wish I would have gone with intel instead.

You can do the following choices to resolve this:

#1 Lower it to 5600
#2 Raise the SoC and VDDIO-MEM voltages manually.
#3 Set FLCk/MEM ratio to /2

All are easy to do.

 

[trparky]

Yeah Samsung ICs using JEDEC specs does work

Then Samsung needs to fix their shit. There's a reason why the JEDEC standard exists.

I still have problems getting my msi x670e carbon to boot with 2 x 32GB sticks of hynix memory at 6000 mhz on the beta 1.0.0.7b bios.

I have 2x16 GB sticks of G.Skill memory, no issues even with EXPO enabled. What brand of memory do you have?

 

[ir_cow]

Then Samsung needs to fix their shit. There's a reason why the JEDEC standard exists.

What do you mean? Samsung is following the JEDEC standard. Hence why it works.

Reports from people with issues is only with enabling XMP/EXPO. Like I said, that is where its up to AMD to have some guidance from MBs vendors. It's all speculation of course to the real issue, but I don't think Samsung is to blame. Nothing wrong with the ICs themselves.

 

[trparky]

What do you mean? Samsung is following the JEDEC standard. Hence why it works.

Reports from people with issues is only with enabling XMP/EXPO. Like I said, that is where its up to AMD to have some guidance from MBs vendors. It's all speculation of course to the real issue, but I don't think Samsung is to blame. Nothing wrong with the ICs themselves.

Oh, I was under the impression that Samsung memory ICs were having issues running on AMD systems at bone stock configs.

 

[Why_Me]

AMD and RAM issues go together like coffee and creamer.

 

[ir_cow]

Oh, I was under the impression that Samsung memory ICs were having issues running on AMD systems at bone stock configs.

Maybe. Not everyone reports problems. Either way, it's good AMD is addressing this.

 

[Gmr_Chick]

What do you mean? Samsung is following the JEDEC standard. Hence why it works.

Reports from people with issues is only with enabling XMP/EXPO. Like I said, that is where its up to AMD to have some guidance from MBs vendors. It's all speculation of course to the real issue, but I don't think Samsung is to blame. Nothing wrong with the ICs themselves.

Yep, can confirm. When I had my Asus board, my Samsung-based, X670E Hero QVL-listed, G.Skill kit wouldn't boot when EXPO was enabled - it would only boot into safe mode BIOS due to "system instability" and if I wanted it to boot, even after trying suggestions provided here, the only thing that would guarantee a boot into Windows was running the kit at stock JEDEC speed.

 

[Vya Domus]

There is a ton of settings that is set by the motherboard.

Exactly, this isn't up to AMD. AMD can't test every RAM kit out there for every motherboard and provide those settings it's obviously up to the motherboard manufacturer to test the RAM compatibility, it always has been.

 

[unwind-protect]

OK, but I argue that no matter what memory kit you buy, it should simply work out of the box at base DDR5 specifications. No EXPO, no DOCP, no XMP, no overclocking of any sort.

Right?

AMD doesn't say. On an issue like this that could mean runtime memory corruption they should really be a little more specific.

If you really rely on data integrity a change message like this could make you not use Samsung memory at all until the new BIOS is available (and doesn't break anything else you rely on).

 

[bug]

But why? What is Samsung doing with their DDR5 memory as versus the other manufacturers? What is Samsung doing wrong? Are they not holding themselves to the DDR5 specifications?

I have no idea why AGESA has such a troubled history. It's supposed to handle just the initialization of a few things.

At one point, Microsoft had to patch Win95 because during init there was a hardcoded waiting time somewhere. As CPUs got faster, other things started up faster and that delay started to get in the way of other things.

 

[Deleted member 232534]

 

[bug]

View attachment 307519

Wth is this "memory training" and why is it taking so long? I haven't seen that with Intel.