MSI Raider A18 HX A9WJG-091TR RAM and SSD Incompatibility Issues - Is a BIOS Update Available?

joao.sousa.domingo158102d7 said:
First run:
View attachment 201807
That's with win OS SSD on a gen4 slot and laptop without battery - but I believe that doesn't affect these results.
I'll do more runs laters.

I'm latest Titan 18 HX AI A2XWJG (285HX) - oh and Kingston 128Gb 6400 (52-51-51-102)
Click to expand...
Wow, great score!

Unfortunately you got only 197 runtimes installed, they have a total of 199. Two runtimes - and rather juicy ones - are missing. So the test is slightly incomplete, and these two runtimes might have put you over the 11 minute mark (if my earlier findings are any indication).

So you probably downloaded the "Single File" installer? Since the juicy runtimes would bump the file size above 4 GB, and Windows has a hard limit on executable file sizes at 4 GB - these are now only available in the "Web" installer. It's easy enough to switch your installer:

- uninstall the "Single File" package using Add Remove Programs
- download the "Web" installer and run it, telling it to do a full setup instead of the default minimum setup
- you can also instruct it on one of the screens before setup starts to save all downloaded files (so you won't have to re-download them again later)

That'll get us to an apples-to-apples comparison with 199 runtimes.

BUT, you'll still have one element superior to my testing - 128 GB RAM! I'm sure that'll help plenty, too - so I wouldn't be surprised if you outdo my reported score again (although, it would surprise me if the 285HX matches or exceeds the 9955HX3D score).

Any thoughts on where I might be able to order 128 GB laptop ram at 5600 MHz and 6400 MHz rated speeds? That's pretty much the only missing component in my rigs (being stuck at 96 GB for now).
 
electrostingz said:
I don't have enough RAM to be competitive, 32GB installed 5600M/T but it's a 14900HX laptop and I'm using modified CPU settings with HT disabled.


View attachment 201815

Just a note on that screenshot, it doesn't sustain 97 Deg C, only peaks momentarily at times so it's looks worse than it is.
Click to expand...
This is so cool!

It would probably have been the best performing processor of all of ours combined, but the RAM is indeed ruining the fun.

On that note, I don't even know if the 14900HX can sustain 5600 MHz RAM speed at 96 GB. The 13980HX would not run at that speed at anything above 32 GB, so 64 GB or more - and you were down to 5200 MHz - giving the 7945HX(3D) a competitive advantage.

At least having HT disabled reduced part of your memory pressure down.

That's hands down the best score I've seen ever with just 32 GB RAM.

Edit: During the run, especially towards the very end of it; what is your best rated CPU speed that the Windows Task Manager itself showed you?

Running on the 285HX, I never have this go above 4.8GHz; on the 7745HX3D it was pretty constant around 5.3GHz, and 5.1GHz for the 9955HX3D.
 
BTW I haven't forgotten about 3DMark - I run a full suite starting with 3DMark 6, collecting all scores through Vantage, 11, and the newer iterations starting with Fire Strike, etc.

It's especially interesting to see how DX9 scores with 3DM6 don't really improve at all, from generation to generation of the latest and very best nVIDIA has to offer :D

And let's not forget the venerable SuperPI for single thread benchmark scoring (which always seems to favor Intel) and wPrime for multi core testing!

Call me old fashioned, but I still use HDTune for SSD testing too - yes, it is totally obsolete nowadays (especially their last free version I have).

I hope to get to those once my operating environment has stabilized somewhat, seeing as I'm still fighting to get my Windows images updated and running successfully on not one, but two brand new devices (the 9955HX3D keeps acting up in more ways than I would imagine - or want to burden you with - if it only weren't the performance crown holder).
 
mimarsina158b02db said:
Wow, great score!

Unfortunately you got only 197 runtimes installed, they have a total of 199. Two runtimes - and rather juicy ones - are missing. So the test is slightly incomplete, and these two runtimes might have put you over the 11 minute mark (if my earlier findings are any indication).

So you probably downloaded the "Single File" installer? Since the juicy runtimes would bump the file size above 4 GB, and Windows has a hard limit on executable file sizes at 4 GB - these are now only available in the "Web" installer. It's easy enough to switch your installer:

- uninstall the "Single File" package using Add Remove Programs
- download the "Web" installer and run it, telling it to do a full setup instead of the default minimum setup
- you can also instruct it on one of the screens before setup starts to save all downloaded files (so you won't have to re-download them again later)

That'll get us to an apples-to-apples comparison with 199 runtimes.

BUT, you'll still have one element superior to my testing - 128 GB RAM! I'm sure that'll help plenty, too - so I wouldn't be surprised if you outdo my reported score again (although, it would surprise me if the 285HX matches or exceeds the 9955HX3D score).

Any thoughts on where I might be able to order 128 GB laptop ram at 5600 MHz and 6400 MHz rated speeds? That's pretty much the only missing component in my rigs (being stuck at 96 GB for now).
Click to expand...
I got two 10:35 in the beginning but was running chrome and some background stuff. For the remaining tests, I got a 10:32 and then a bunch of straight 10:31s which I could not improve upon.
I can't remember which one I installed (I downloaded both initially). I will do that tomorrow and post all the tests I run.

I stress tested the 128Gb Kingston 6400 on my laptop here: https://forum-en.msi.com/index.php?posts/2332855/
It's all stable as you can see in the tests.
It's not CL38, but at least the 128Gb (2x64GB) have the same exact 6400 at 52-51-51-102 as the OEM 64Gb (2x32Gb) Micron sticks. They were still expensive tho at €260 each. I don't expect any CSODIMM of 2x64Gb to be made of lower than CL52 anytime soon tho - it's a lot of RAM and will likely be unstable even for clocked RAM.
I bought my 2 sticks on amazon.es, and you also have other stores like Amazon.de.

But, this is only for the 285HX. AMD CPUs are limited to 5600MHz and CSODIMMs will give no benefit. And as far as I've seen to get even the 5600 you'd likely have to stick with 64Gb only.. Someone else needs to confirm.
 
mimarsina158b02db said:
That's hands down the best score I've seen ever with just 32 GB RAM.

Edit: During the run, especially towards the very end of it; what is your best rated CPU speed that the Windows Task Manager itself showed you?

Running on the 285HX, I never have this go above 4.8GHz; on the 7745HX3D it was pretty constant around 5.3GHz, and 5.1GHz for the 9955HX3D.
Click to expand...

I was wondering if this was actually ok for 32GB :-)

So I did a few test runs and always use hardware monitor to see the sensor readouts and it appears towards the end it goes down to 2 cores, as the program implies 2 thread per runtime, my CPU is max boost on 2 cores at that point, 5.8GHz.
 
joao.sousa.domingo158102d7 said:
I got two 10:35 in the beginning but was running chrome and some background stuff. For the remaining tests, I got a 10:32 and then a bunch of straight 10:31s which I could not improve upon.
I can't remember which one I installed (I downloaded both initially). I will do that tomorrow and post all the tests I run.

I stress tested the 128Gb Kingston 6400 on my laptop here: https://forum-en.msi.com/index.php?posts/2332855/
It's all stable as you can see in the tests.
It's not CL38, but at least the 128Gb (2x64GB) have the same exact 6400 at 52-51-51-102 as the OEM 64Gb (2x32Gb) Micron sticks. They were still expensive tho at €260 each. I don't expect any CSODIMM of 2x64Gb to be made of lower than CL52 anytime soon tho - it's a lot of RAM and will likely be unstable even for clocked RAM.
I bought my 2 sticks on amazon.es, and you also have other stores like Amazon.de.

But, this is only for the 285HX. AMD CPUs are limited to 5600MHz and CSODIMMs will give no benefit. And as far as I've seen to get even the 5600 you'd likely have to stick with 64Gb only.. Someone else needs to confirm.
Click to expand...
Thanks for the links!

The web installer will download stuff online, so please be sure to give it some time.

I am happy to confirm that with my Crucial/Micron 48 GB sticks, the AMD 9955HX3D is indeed running at 5600 MHz.

For the record, I'd originally bumped the HX3D to 96 GB via the Norse Titan's RAM sticks, which were reportedly running at only 3200 MHz.

But switching between those sticks made no difference to the benchmark runtime speeds, perhaps due to the latency that improves when the MHz rating goes down.
 
electrostingz said:
I was wondering if this was actually ok for 32GB :-)

So I did a few test runs and always use hardware monitor to see the sensor readouts and it appears towards the end it goes down to 2 cores, as the program implies 2 thread per runtime, my CPU is max boost on 2 cores at that point, 5.8GHz.
Click to expand...

Wow, that's the highest GHz speed I've seen so far of any (mobile) CPU for sure!

I'm 99% positive based on what you've reported that the 14900HX then, when paired with 96 GB RAM (hopefully at 5600 MHz), would outdo both the 285HX and the HX3D's handsomely. Hard to beat 5.8GHz with 5.3GHz (7945HX3D), 5.1GHz (9955HX3D), or 4.8GHZ (285HX).

This finding would basically justify claims Intel made when they first shipped the 14900HX that it was capable of outdoing the 7945HX3D.

Edit: Does anyone have a way to disable SMT (AMD's name for HT) on the HX3D's? It's not there in the BIOS, on none of the AMD devices I've come across as of late actually.

It'd be really great to have additional findings based on AMD performance when SMT is turned off.
 
electrostingz said:
I was wondering if this was actually ok for 32GB :-)

So I did a few test runs and always use hardware monitor to see the sensor readouts and it appears towards the end it goes down to 2 cores, as the program implies 2 thread per runtime, my CPU is max boost on 2 cores at that point, 5.8GHz.
Click to expand...
what CPU score do you get in Time Spy? that one doesn't need RAM.
 
joao.sousa.domingo158102d7 said:
what CPU score do you get in Time Spy? that one doesn't need RAM.
Click to expand...
Pretty sure the Norse Titan would perform better there, only because of the RTX 5090 though.

Edit: Just saw you mentioned the CPU score, that'd be a good comparison, and I'd expect the 14900HX to outperform the 285HX.

3DMark actually has a dedicated CPU benchmark too, so if your version allows that, it'd be great to run that through too - it has both single and multi core tests available.
 
It's an interesting one when you compare the 285HX with the14900HX, the 285HX E-cores will out shine the 14900HX so I think there might be some advantage under the right conditions, and Intel did mention multi-core optimisations are better on the 285HX.

joao.sousa.domingo158102d7 said:
what CPU score do you get in Time Spy? that one doesn't need RAM.
Click to expand...

My laptop is a ASUS ROG Strix G16, 14900HX + RTX 4060, with default settings it's terrible and will run too hot for what it's capable of, which kills performance. So everything I show is non standard / tweaked settings, this was when I limited the Performance cores to 5GHz only with voltage adjustments.

Screenshot 2025-05-01 134204.png
 
electrostingz said:
It's an interesting one when you compare the 285HX with the14900HX, the 285HX E-cores will out shine the 14900HX so I think there might be some advantage under the right conditions, and Intel did mention multi-core optimisations are better on the 285HX.



My laptop is a ASUS ROG Strix G16, 14900HX + RTX 4060, with default settings it's terrible and will run too hot for what it's capable of, which kills performance. So everything I show is non standard / tweaked settings, this was when I limited the Performance cores to 5GHz only with voltage adjustments.

View attachment 201834
Click to expand...
wow.. that is a massive CPU score.. which tells me I must have some real CPU problem on my end..
 
electrostingz said:
It's an interesting one when you compare the 285HX with the14900HX, the 285HX E-cores will out shine the 14900HX so I think there might be some advantage under the right conditions, and Intel did mention multi-core optimisations are better on the 285HX.



My laptop is a ASUS ROG Strix G16, 14900HX + RTX 4060, with default settings it's terrible and will run too hot for what it's capable of, which kills performance. So everything I show is non standard / tweaked settings, this was when I limited the Performance cores to 5GHz only with voltage adjustments.

View attachment 201834
Click to expand...
Good point. Although I think I'd take single core peak velocity over multi core optimizations pretty much any day, given that most software/tasks we use/do on a daily basis (unless you're specializing in AI or rendering etc.) is still primarily single core in nature.
 
Wow, the number of technical curveballs I've been getting lately have been spell binding!

Remember the MSI Raider 9955HX3D BIOS thinking the brand-new Sabrent 8 TB Rocket Plus drive I'd plugged into it was just a 2 MB device?

Now when I connect it elsewhere, I'm getting other PCs thinking the same.

May there have been like some low level format or something the BIOS did to this SSD, does anyone know how I might be able to put the drive back the way it was?
 
mimarsina158b02db said:
Wow, the number of technical curveballs I've been getting lately have been spell binding!

Remember the MSI Raider 9955HX3D BIOS thinking the brand-new Sabrent 8 TB Rocket Plus drive I'd plugged into it was just a 2 MB device?

Now when I connect it elsewhere, I'm getting other PCs thinking the same.

May there have been like some low level format or something the BIOS did to this SSD, does anyone know how I might be able to put the drive back the way it was?
Click to expand...
Use diskpart to clean and format. Follow this:
www.windowscentral.com

How to use DiskPart to clean and format drive not working on Windows 10

If a drive is causing problems on Windows 10, it could be a logical issue that can be fixed with a few commands.
www.windowscentral.com www.windowscentral.com
 
So one of the other oddities I've been working on is that 90% of my download bandwidth (and only that, not the upload) was magically disappearing somehow.

After days of sleuthing, more bad news for Intel.

They had this thing called Intel Technology Access Filter Driver enabled on my WiFi card. Unchecking it restored all my bandwidth.

Wait, it gets worse. Much worse.

Whatever this is, it was also preventing DISM from completing any and all operations on my same Windows image affected by this issue.

After disabling it in this manner, DISM started working as well.

Seriously - this is malware-grade behavior. From best I can tell, this filter driver was installed when I opted in to the Intel Driver Support Assistant software.

You know, back in the days of XP, I seem to remember NOT installing ANY drivers from Intel made my PC run substantially faster.

I guess it's good to see some things never change?

Somebody seriously needs to file a CVE about this. What a nightmare of a company.

MSI need to seriously think about norse-dragonifying their AMD line, not their Intel line.
 
So my Norse Titan is finally settled in (I have two OS images I use in production).

My best score is 11:40 with the Pre-Compiled Runtimes Tool "benchmark", running with Windows Defender temporarily disabled, and all background apps (I have tons of SQL and similar servers hosted on this thing) on idle (so some performance loss over the peak 11:13 time is nominal).

I'll be following up with more tests on the Norse Titan and the AMD device on the same and secondary OS images I have in production.

All tests using the full complement of 199 runtimes on Max Power.
 
electrostingz said:
It's an interesting one when you compare the 285HX with the14900HX, the 285HX E-cores will out shine the 14900HX so I think there might be some advantage under the right conditions, and Intel did mention multi-core optimisations are better on the 285HX.



My laptop is a ASUS ROG Strix G16, 14900HX + RTX 4060, with default settings it's terrible and will run too hot for what it's capable of, which kills performance. So everything I show is non standard / tweaked settings, this was when I limited the Performance cores to 5GHz only with voltage adjustments.

View attachment 201834
Click to expand...
Regarding the difference between the 285HX and 14900HX, I looked at the HWiNFO .csv data on Total CPU Usage (= Core Usage) and CPU Package Power and noticed that during the entirety of Time Spy (2 GPU + CPU tests) the Total CPU Usage reaches only a maximum of 63.1% and only has an average of 20.8%.. (skewed towards the CPU test as expected). As such, this only draws a maximum peak of 126 W and an average of 46.8 W for a CPU (285HX) that can get a maximum of 220 W - this occurs in the same way for several other Time Spy runs.
image (1).png
image.png


I ran Prime95 for 45 minutes and got 100% average Total CPU Usage - it's 100% the entire 45m - and during that time it draws an average of 175.6 W with a peak maximum of 209.5 W (as shown in the attached HWiNFO .csv file). Among other Prime95 runs, the maximum CPU Power I saw was 214 W (out of a maximum of 220 W):
CPU_214Wmax.png


In other words, the Time Spy benchmark is not making full use of the Intel 285HX, especially during the CPU test - not even close.. So the score suffers and stays around 17800. I guess during those runs where I got 18400 CPU score, the CPU just happened to be used a little bit more (and certainly draw more power). But in the end, we can't use Time Spy CPU test to compare processors - needs to be a benchmark like Prime95 that makes 100% use of the CPU all the time.
 

Attachments

  • prime95_run.zip
    504.7 KB · Views: 15
Back
Top