MSI Motherboards Unleash Extreme Power with Memory Capacity Boosted To 256GB

[ir_cow]

I would love to see people trying to run this at 6000 on Ryzen 7000 or Intel 12/13/14. Good luck getting it to post.

 

[Minus Infinity]

Finally the perfect antidote to Adobe software's woeful memory leaks.

 

[mechtech]

What's actually required for this? Hardware modifications or just a bios update?

 

[Dr. Dro]

It looks like a bug to me. On my dual-channel DDR4 system it shows "2x 64-bit". On my dual-channel DDR5 AM5 system it shows "2x 32-bit" like in the screenshot, which is incorrect. Both AIDA64 and HWiNFO confirm that dual-channel DDR5 is enabled. It should probably show something like "2x (2x 32-bit)" for dual-channel DDR5.

Did you ensure to populate slots A2 and B2 instead of placing them next to each other? If memory is installed correctly, it should report:

Dual 64-bit channel DDR4 with two sticks (2x 64-bit)
Quad 32-bit channel DDR5 with two sticks (4x 32-bit)

 

[Chaitanya]

I would love to see people trying to run this at 6000 on Ryzen 7000 or Intel 12/13/14. Good luck getting it to post.

Did you request couple of kits for review?

 

[TumbleGeorge]

AMD Phoenix

This specs exist from more of 6 months. Propably someone with PC will explore webpage for date were is published, please?

 

[gurusmi]

I would love to see people trying to run this at 6000 on Ryzen 7000 or Intel 12/13/14. Good luck getting it to post.

By using that at a Asus board one should have bars that need a max of 1,25V. This seems to be a limit on using 4 DDR5 bars in general.

 

[ir_cow]

Did you request couple of kits for review?

I don't have a Kingston contact.

 

[Wirko]

Finally the perfect antidote to Adobe software's woeful memory leaks.

Why yes, the poison and the antidote both cost a lot.

 

[ncrs]

Did you ensure to populate slots A2 and B2 instead of placing them next to each other? If memory is installed correctly, it should report:

Dual 64-bit channel DDR4 with two sticks (2x 64-bit)
Quad 32-bit channel DDR5 with two sticks (4x 32-bit)

Yes, dual-channel is working, and as I wrote in the post you replied to both AIDA64 and HWiNFO report that it is on. It's just CPU-Z that is misreporting it.

 

[FreedomEclipse]

Now what if you can do something stupid, like allowing the AMD APU to use 128GB+ of ram??

 

[Dr. Dro]

Now what if you can do something stupid, like allowing the AMD APU to use 128GB+ of ram??

I'm not sure that would serve any meaningful purpose, as more often than not even granting an unusually large aperture size to the iGPU results in no improvement than simply letting Windows manage the graphics memory itself... some anecdotal cases of reserving memory at the BIOS level being faster, but those are rare afaik

 

[FreedomEclipse]

I'm not sure that would serve any meaningful purpose, as more often than not even granting an unusually large aperture size to the iGPU results in no improvement than simply letting Windows manage the graphics memory itself... some anecdotal cases of reserving memory at the BIOS level being faster, but those are rare afaik

Fallout/Skyrim and tonnes of graphical mods, addons and hi-res texture packs

 

[A Computer Guy]

So quad channels? That's what HEDT is for.
I've been saying for years that I think it's been a mistake to have the top "mainstream" CPUs (only) on the mainstream platforms. I wish that CPUs like i7-13700K and 7800X and up would be on the Xeon-W and Threadripper platforms, respectively, offering the same high clocks, but quad channel memory and plenty of PCIe lanes. I think this mix is exactly what many developers and content creators needs these days. Today we have to choose between fast cores and memory + IO (+ lots of slower cores).

I do wonder if we get to see Arrow Lake and Zen 5 supporting 256 GB of RAM. I would seriously consider it.

It seems kind of weird to me threadripper even offers lower core count parts when that's pretty much covered by AM4/AM5.
The 5955WX is approx. double the cost of 5950x for a 16 core Zen3 CPU just because?

 

[TumbleGeorge]

It seems kind of weird to me threadripper even offers lower core count parts when that's pretty much covered by AM4/AM5.
The 5955WX is approx. double the cost of 5950x for a 16 core Zen3 CPU just because?

More instructions activated for professional purpose. Support 8 channel RAM and up to 2TB RAM. Much more PCIe lines. Make me AM4/5 PC with more of 192GB RAM right now, please.

 

[A Computer Guy]

More instructions activated for professional purpose. Support 8 channel RAM and up to 2TB RAM. Much more PCIe lines. Make me AM4/5 PC with more of 192GB RAM right now, please.

Also threadripper is ECC RAM certified too I thought.

 

[AnotherReader]

More instructions activated for professional purpose. Support 8 channel RAM and up to 2TB RAM. Much more PCIe lines. Make me AM4/5 PC with more of 192GB RAM right now, please.

I don't think that Threadripper supports more instructions than Ryzen.

 

[TumbleGeorge]

I don't think that Threadripper supports more instructions than Ryzen.

Where I wrote that is question of support? Read me again.
Here has detailed instructions that works for Ryzen Threadripper Pro series 5000. (On the bottom of the page)

 

[AnotherReader]

Where I wrote that is question of support? Read me again.
Here has detailed instructions that works for Ryzen Threadripper Pro series 5000. (On the bottom of the page)

I've gone through the page and I don't see anything in the feature list that isn't supported by Ryzen. The one exception would be support for RDIMMs.

 

[Dr. Dro]

Fallout/Skyrim and tonnes of graphical mods, addons and hi-res texture packs

Yeah, that's the reason I have always opted for GPUs with 16+ GB VRAM for the past 5 years or so, starting with the Vega FE. But I'm not sure if it matters when Windows manages the memory all the same, beyond this manually reserved aperture. One could argue that leaving the aperture at a smaller size, say, 128 to 256 MB reserved, and Windows manage the rest would even be beneficial since the memory that is free could be used by the OS if VRAM requirements are not so high.

 

[FreedomEclipse]

Yeah, that's the reason I have always opted for GPUs with 16+ GB VRAM for the past 5 years or so, starting with the Vega FE. But I'm not sure if it matters when Windows manages the memory all the same, beyond this manually reserved aperture. One could argue that leaving the aperture at a smaller size, say, 128 to 256 MB reserved, and Windows manage the rest would even be beneficial since the memory that is free could be used by the OS if VRAM requirements are not so high.

Thats why I said it was stupid in my original post. I just wanted to see something extremely dumb. I mean what are you gonna do with 256GB of ram?? Play Minecraft??

 

[A Computer Guy]

I've gone through the page and I don't see anything in the feature list that isn't supported by Ryzen. The one exception would be support for RDIMMs.

I can say for Pro cpus such as 4750g (zen2) vs consumer 3800x (zen2) ECC support is different. For example Memtest86 ECC error injection actually seems to work.
Not sure if this is a UEFI/BIOS limitation or chip limitation.

 

[efikkan]

It seems kind of weird to me threadripper even offers lower core count parts when that's pretty much covered by AM4/AM5.
The 5955WX is approx. double the cost of 5950x for a 16 core Zen3 CPU just because?

As me and others have mentioned, there are many other features than just pure core count that matters, including but not limited to;
- Memory support (bandwidth, ECC, ...)
- PCIe lanes (for GPUs, SSDs, NICs, ...)
Such features are though fairly irrelevant for those not into any professional/semi-professional use cases.
Unfortunately, even the 12 and 16 core Threadrippers have a bit lower clock speeds (of the same generation), and are far behind the current generation Ryzen CPUs.

The "HEDT" segment have largely faded away, but should be more relevant now than ever, considering the rise of content creation. Having the fastest cores combined with more memory and PCIe shold be the ideal mix for (semi-)professionals. And even with the addition for a few more PCIe lanes, mainstream platforms are still falling short.

I can say for Pro cpus such as 4750g (zen2) vs consumer 3800x (zen2) ECC support is different. For example Memtest86 ECC error injection actually seems to work.
Not sure if this is a UEFI/BIOS limitation or chip limitation.

There are certainly different levels of ECC support and features, it's not as simple as either/or.
And any such feature must be supported on every level in order for it to work, so in most cases the ECC memory itself, CPU memory controller, BIOS and OS(!). If one is lacking the desired support, the entire feature is just a placebo. And even though most CPU memory controllers have support technically, if they don't support it officially, they are probably not verified, so it might not offer the protection that you want, so a BIOS "enabling" ECC on an unverified CPU might not be a good idea.

I wish I had time to do a deep-dive into researching what ECC levels are supported on various mainstream and workstation platforms, and do real validation on it, not just trust marketing gimmicks.

But even moreso, get some real answer to find out when ECC actually makes a real world difference. Even though I understand well how it works, I still couldn't give a clear answer to where to draw the line on having it or not. Clearly, mission critical servers should have it, but should your home "workstation" have it?
And I'm not primarily worried about crashes. I mostly use Linux, and system crashes and reboots are hardly in my vocabulary. I'm more worried about undetected memory corruption leading to corrupted files/data. When applications are running for months, what are the likelihood of corrupted data?

 

[A Computer Guy]

As me and others have mentioned, there are many other features than just pure core count that matters, including but not limited to;
- Memory support (bandwidth, ECC, ...)
- PCIe lanes (for GPUs, SSDs, NICs, ...)

Aren't those feature provided by the I/O die though? In essence you are paying for the CPU cores and the more advanced I/O die with Threadripper.

Such features are though fairly irrelevant for those not into any professional/semi-professional use cases.
Unfortunately, even the 12 and 16 core Threadrippers have a bit lower clock speeds (of the same generation), and are far behind the current generation Ryzen CPUs.

The "HEDT" segment have largely faded away, but should be more relevant now than ever, considering the rise of content creation. Having the fastest cores combined with more memory and PCIe shold be the ideal mix for (semi-)professionals. And even with the addition for a few more PCIe lanes, mainstream platforms are still falling short.

There are certainly different levels of ECC support and features, it's not as simple as either/or.

I think the basics are reporting, correction, and fail action (halt or reboot). Not sure what else there is but I imagine servers likely have more advanced features like chipkill.

And any such feature must be supported on every level in order for it to work, so in most cases the ECC memory itself, CPU memory controller, BIOS and OS(!). If one is lacking the desired support, the entire feature is just a placebo. And even though most CPU memory controllers have support technically, if they don't support it officially, they are probably not verified, so it might not offer the protection that you want, so a BIOS "enabling" ECC on an unverified CPU might not be a good idea.

I think you have a good point here. Without proper validation your just taking a chance it's not going to work properly when you need it. I almost shelled out the bucks for Xeon but ended up going with Zen+ R5 2600 when I read multiple people were reporting ECC was working just fine on those chips. I think I got a better deal for the price/performance over Xeon and managed to confirm it was working via ram overclocking and memtest86 error injection. UEFI/BIOS updates later borked memtest86's error injection feature so it was a bit of a reminder AMD could close the door on unverified CPU's at any time.

I wish I had time to do a deep-dive into researching what ECC levels are supported on various mainstream and workstation platforms,

I would go with one of those Asrock AM4/AM5 workstation boards but the I/O ports are lacking and is really meant for server boxes. Also they lack RGB.

and do real validation on it, not just trust marketing gimmicks.

It's that validation cost of course vendors are avoiding (or so people say) on consumer platforms and provides continued fuel for segmentation that people often complain about who are looking for ECC support. Asrock listed support for ECC on a wide variety of their AM4 boards which was a nice nod that things were changing. MSI didn't seem to do that and not sure if any of their AM4 boards ever supported it. I think Gigabyte and Asus have some AM4 boards that support ECC.

I'm not sure wtf is going on with AM5 though. From Asrock forums there were messages about ECC support having trouble and support eventually coming in UEFI/BIOS updates. Recently I see they have been updating their AM5 motherboard pages to indicate if they are supporting ECC. AM5 Pro cpus are now being listed in compatibility lists which is a good sign that the motherboards likely actually validated and support it but their QVL lists are NOT showing any ECC modules yet. Supposedly AM5 CPU's officially support ECC and AMD is actually listing it as such provided the motherboard supports it.

I was hesitant to bite on AM5 when I heard Asrock pulled their advertising of ECC support for motherboards sometime after the launch of AM5. I thought perhaps AMD finally pulled the plug and decided to segment ECC support with Threadripper. Perhaps it took them some time to adjust for DDR5 like with higher capacity ram modules.

But even moreso, get some real answer to find out when ECC actually makes a real world difference. Even though I understand well how it works, I still couldn't give a clear answer to where to draw the line on having it or not. Clearly, mission critical servers should have it, but should your home "workstation" have it?

Since I starting using ECC ram I've had very few PC problems (that I didn't cause myself) but that's anecdotal evidence and sample size of one that's not a very convincing argument. It's not just cosmic blips that can be a problem but with the increase in high speed components, busses, and overcall capacity of RAM with more to go wrong, and mix in overclocking you've got a recipe for disaster. I'm surprised it hasn't become more of a problem but I guess the quality of technology was keeping up in pace with good enough reliability. Eventually wall started closing in with the DDR5 and on-chip ECC was born just to get it to work. Obviously though people have had working and stable non-ECC systems.

And I'm not primarily worried about crashes. I mostly use Linux, and system crashes and reboots are hardly in my vocabulary. I'm more worried about undetected memory corruption leading to corrupted files/data. When applications are running for months, what are the likelihood of corrupted data?

Since I run VM's for daily work I worry about it more than the average person since memory corruption there can cause a lot of damage and headaches if things go wrong. Other issues of importance surrounding memory corruption includes things like SSD/NVMe caching which is something people probably don't think much about. Then there are also software out there like Primocache and of course ZFS.

 

[efikkan]

I would go with one of those Asrock AM4/AM5 workstation boards but the I/O ports are lacking and is really meant for server boxes. Also they lack RGB.

Depends on whether they are validated or not.
I see Ryzen 7x00 series have listed ECC support, I assume this means they have passed validation prior to binning, while the earlier generations did not have official support.

It's that validation cost of course vendors are avoiding (or so people say) on consumer platforms and provides continued fuel for segmentation that people often complain about who are looking for ECC support. Asrock listed support for ECC on a wide variety of their AM4 boards which was a nice nod that things were changing. MSI didn't seem to do that and not sure if any of their AM4 boards ever supported it. I think Gigabyte and Asus have some AM4 boards that support ECC.

While I have no issue with sub-$150 boards lacking ECC, I do think they should include it on "workstation"-ish boards (along with real heatsinks instead of these giant metal blobs).

I see that Asus' ProArt B650-/X670E-CREATOR boards lists ECC, while the Intel counterparts ProArt B760-/Z690-CREATOR don't, probably because they need a different chipset. Anyone who wants that have to stick to the Pro WS W680 boards on Alder Lake/Raptor Lake, when just talking about Asus. (There are others like Asrock and Gigabyte, but they aren't the best in terms of reliability. SuperMicro is a good option, but are hard to find and expensive.)
So in this case, AM5 have an advantage over LGA1700 in affordability.

(I see that my ProArt B550-CREATOR now lists support, but those CPUs weren't validated so I don't plan on replacing my RAM to gain ECC)

As mainstream platforms, these do have fairly limited IO though. Like if someone needs GPU(s), SSDs(possibly some in RAID1), 10G NICs etc. it quickly runs out, and chipset PCIe lanes are not suitable for everything.

I'm not sure wtf is going on with AM5 though. From Asrock forums there were messages about ECC support having trouble and support eventually coming in UEFI/BIOS updates.

BIOS issues is a common problem with Asrock.
I'm not very "forgiving" when it comes to this, fundamental things like this should have been ironed out during development. People buy this feature for reliability, so then we expect them to do it properly. And this isn't something that should be hard; server and proper workstation boards from others manage to do this.

I was hesitant to bite on AM5 when I heard Asrock pulled their advertising of ECC support for motherboards sometime after the launch of AM5. I thought perhaps AMD finally pulled the plug and decided to segment ECC support with Threadripper. Perhaps it took them some time to adjust for DDR5 like with higher capacity ram modules.

None of which should be an issue, they either aren't competent or aren't prioritizing proper validation, and this is not a good sign for anyone who wants a computer that "just works".
There is a big difference between let's say a teenager building a computer for fun (gaming and OC), and a "grownup" building a computer for their hobby/side income/etc. (or even full time work), as they usually have limited time or deadlines.

Since I starting using ECC ram I've had very few PC problems (that I didn't cause myself) but that's anecdotal evidence and sample size of one that's not a very convincing argument.

Yeah, it's really hard to tell.
As I mentioned, I rarely see system crashes at all. While the Windows computer at my day job crashes 1-2 times per month, at previous jobs which have been mostly Linux users people "never" experience such crashes, so software is a big factor here.

Yet still, I have been thinking about testing it in a way where I write a program to allocate let's say 16 GB of a distinct pattern, run the program for 6 months non-stop, and every 24 hours or so do a complete validation against a file, to see if any bits are flipped. And if this produces any interesting result, then test the same thing on machines with ECC.

I'm not concerned about corruption during a day or two, I keep my machine always on and my programs running, so when I have a little time to do some work, everything is ready.

Since I run VM's for daily work I worry about it more than the average person since memory corruption there can cause a lot of damage and headaches if things go wrong. Other issues of importance surrounding memory corruption includes things like SSD/NVMe caching which is something people probably don't think much about. Then there are also software out there like Primocache and of course ZFS.

Absolutely important points.
Especially if you run your computer for months at the time, data can stay in RAM risk getting corrupted. This is why I would like to find out whether such corruption is likely or astronomically unlikely.