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

[mimarsina158b02db]

Very interesting observations!

I think for a realistic CPU performance assessment though, you need something that exercises a single core only (like the venerable SuperPI), as most software today is still single core primarily.

IOW even if 285HX outperforms 9955HX3D when all cores are maxed out for an hour at least (which isn't a given and should be tested for explicitly), this wouldn't automatically guarantee the highest real world performance; especially if the latter bests the former when used with fewer cores that more accurately match typical software CPU utilization patterns.

That's why I like the Pre-Compiled Runtimes Tool, as it blends both the most extreme and the most typical usage patterns into a single number in a neatly packaged, summarily executed tool - everything except the GPU, of course - seeing as both HDD/SSD and RAM performance are factored in.

Although I suppose this tool requires the most inordinate amount of RAM that's not typically required in daily usage amongst all the other benchmarks out there; and this would be its main weakness IMHO, with all due respect to its manufacturers/publishers.

 

[electrostingz]

In general I would ignore certain aspect of benchmarking if it doesn't apply to your usage, for instance if you are only gaming on the laptop this is the area that you should try to maximise. If the E-cores are never used in games why have the E-cores active and boosting for no reason? It's power wasted that could assist the P-Cores (the available power will distribute to the P-core only).

You should try a timespy run with the E-cores disabled to see if it improves the GPU benchmark, might be more consistent and hopefully higher to some extent, forget the low CPU test score as you won't have 16 E-cores active. And see in game performance, how does it feel and performance, maybe more responsive? Also other gaming benchmarks that are not using the E-cores, including actual games with benchmarks.

And for a 200 series CPU your 36MB L3 cache is shared between the P and E cores, so 36MB all to the P-cores.

You also have design considerations in play, this is a given no matter what, with more cores there is an increase in latency. It's due to a few aspects, internal CPU design and how the cores are accessed and data passed on. Internally all the cores are connected on a ring interconnect bus, all the cores are linked and going from one to the other which means time. Then you have the CPU scheduler which allocates work to and from the cores, so it's another point of interest in latency and Intel have added to the mix with their Thread director, some stuff to try and improve P / E core allocation. And factor in more issues such as tasks switching across cores you got more latency issues to consider. Anyway the point I make, all of this is reduced if you don't need the E-cores active including power and thermals.

Give it a try if gaming is your only concern / goal and see if it's any different.

 

[mimarsina158b02db]

In general I would ignore certain aspect of benchmarking if it doesn't apply to your usage, for instance if you are only gaming on the laptop this is the area that you should try to maximise. If the E-cores are never used in games why have the E-cores active and boosting for no reason? It's power wasted that could assist the P-Cores (the available power will distribute to the P-core only).

You should try a timespy run with the E-cores disabled to see if it improves the GPU benchmark, might be more consistent and hopefully higher to some extent, forget the low CPU test score as you won't have 16 E-cores active. And see in game performance, how does it feel and performance, maybe more responsive? Also other gaming benchmarks that are not using the E-cores, including actual games with benchmarks.

And for a 200 series CPU your 36MB L3 cache is shared between the P and E cores, so 36MB all to the P-cores.

You also have design considerations in play, this is a given no matter what, with more cores there is an increase in latency. It's due to a few aspects, internal CPU design and how the cores are accessed and data passed on. Internally all the cores are connected on a ring interconnect bus, all the cores are linked and going from one to the other which means time. Then you have the CPU scheduler which allocates work to and from the cores, so it's another point of interest in latency and Intel have added to the mix with their Thread director, some stuff to try and improve P / E core allocation. And factor in more issues such as tasks switching across cores you got more latency issues to consider. Anyway the point I make, all of this is reduced if you don't need the E-cores active including power and thermals.

Give it a try if gaming is your only concern / goal and see if it's any different.

How do you mean? Does the MSI BIOS have a hidden setting to disable efficiency cores?

BTW the "Thread Director" is garbage. It uses retarded heuristics which don't scale very well in the real world (ex: if your process doesn't have at least one visible pixel, it gets relegated to efficiencies).

 

[joao.sousa.domingo158102d7]

I will be trying CoreDirector (for now) as it's simpler than Process Lasso

 

[electrostingz]

How do you mean? Does the MSI BIOS have a hidden setting to disable efficiency cores?

Not sure as I don't have an MSI laptop to experiment with, what do you have in the hidden section? (I assume you accessed?)
Based on what I can see in the forums,

Fn + Right Copilot + Right Shift + Left Alt + F2
or
Right Ctrl + Right Shift + Left ALT + F2

 

[mimarsina158b02db]

Not sure as I don't have an MSI laptop to experiment with, what do you have in the hidden section? (I assume you accessed?)
Based on what I can see in the forums,

Fn + Right Copilot + Right Shift + Left Alt + F2
or
Right Ctrl + Right Shift + Left ALT + F2

So cool! Will have to check this out! Do you enter this sequence once you're in the BIOS, or at another time?

Is there a similar setting for the AMD counterpart of the MSI device?

 

[joao.sousa.domingo158102d7]

For MSI (intel or amd) Advanced BIOS it's: left ALT + F2 + right CTRL + right SHIFT (once you are in BIOS)

oh what @electrostingz said

 

[mimarsina158b02db]

I will be trying CoreDirector (for now) as it's simpler than Process Lasso

Really cool software, but I don't know if it actually works on the latest Intel devices.

My Surface (on which I keep typing these reports as I alternate between torturing and being tortured by the latest and greatest MSI has on offer) has the Ultra 7 165U processor, which has two ultra efficiency cores.

According to the Task Manager, these are never active. They're always parked.

And none of these de-parkers have been able to get them to be actually used, at least in Task Manager's view.

Yet I do enjoy performance benefits from utilizing all 14 logical cores (including the ultra efficiencies), so this is part of the reason why I am unable to trust anything coming from Intel at this point. Apparently they're now lying to the OS too, on which cores are being used.

It's just crazy.

BTW, the AMD just completed three runs of the Pre-Compiled Runtimes Tool.

Ironically, the first run was 10:53, with the second and third coming in at 11:14 and 11:12.

Still 30+ seconds ahead of the best Intel of this generation, but slower than I was expecting for sure.

This is on the fully loaded OS image, so I'm not too surprised - except to say that the CPU GHz, again according to the Task Manager, never scaled 5GHz or above at any time throughout!

 

[mimarsina158b02db]

For MSI (intel or amd) Advanced BIOS it's: left ALT + F2 + right CTRL + right SHIFT (once you are in BIOS)

oh what @electrostingz said

That didn't work, but this did:

Fn + Right Copilot + Right Shift + Left Alt + F2

Saw primarily overclocking related controls on the AMD, nothing with respect to disabling SMT (HT).

There were a few SSD settings though, they're obscure - I'll compare them with the ones on the Intel (when it's ready for a reboot) and see if that takes care of the detection issues I've been having.

 

[joao.sousa.domingo158102d7]

That didn't work, but this did:

Fn + Right Copilot + Right Shift + Left Alt + F2

Saw primarily overclocking related controls on the AMD, nothing with respect to disabling SMT (HT).

There were a few SSD settings though, they're obscure - I'll compare them with the ones on the Intel (when it's ready for a reboot) and see if that takes care of the detection issues I've been having.

yeah that's because we don't have right CRTL in our keyboards - needs to be copilot key for us. But both sequences are the same.
I didn't look yet - just went straight to core-parker software first

 

[joao.sousa.domingo158102d7]

Really cool software, but I don't know if it actually works on the latest Intel devices.

My Surface (on which I keep typing these reports as I alternate between torturing and being tortured by the latest and greatest MSI has on offer) has the Ultra 7 165U processor, which has two ultra efficiency cores.

According to the Task Manager, these are never active. They're always parked.

And none of these de-parkers have been able to get them to be actually used, at least in Task Manager's view.

Yet I do enjoy performance benefits from utilizing all 14 logical cores (including the ultra efficiencies), so this is part of the reason why I am unable to trust anything coming from Intel at this point. Apparently they're now lying to the OS too, on which cores are being used.

It's just crazy.

BTW, the AMD just completed three runs of the Pre-Compiled Runtimes Tool.

Ironically, the first run was 10:53, with the second and third coming in at 11:14 and 11:12.

Still 30+ seconds ahead of the best Intel of this generation, but slower than I was expecting for sure.

This is on the fully loaded OS image, so I'm not too surprised - except to say that the CPU GHz, again according to the Task Manager, never scaled 5GHz or above at any time throughout!

So it doesn't seem that CoreDirector (and Process Lasso) are very much compatible with Arrow Lake processors yet:
- none of the 3 enforcement methods have any effect on the Pre-Compiled Runtimes Tool - always shows enforcing on 0 (and I run it just in case and results were the same as without using it)
- for 3DMark it does show as enforcing, but results are the same or worst as without using it

- With enforcement method "Efficiency Mode OFF": (same results as before)

- With enforcement method "CPU Affinities": (worst results than before)

- With both enforcement methods "Efficiency Mode OFF" + "CPU Affinities": (worst results than before - same results as with "CPU Affinities")

- With the least restrictive enforcement method "CPU Sets": (same results as before)

Time Spy CPU test is still not using the full capabilities of this CPU.
So unless I can desable e-core in BIOS and test, for now it seems that using core-parking software like CoreDirector is useless - Arrow Lake is still new tho.

On the other hand, my current Win OS is installed on a Samsung 9100 Pro in the gen5 slot - with speeds below (not yet Raid0) - and I have run the Pre-Compiled Runtimes Tool.

I installed InstallAware from Web installer with full features instead of default minimum setup, and the Pre-Compiled Runtimes Tool now always shows 199 runtimes available:

(runs with battery inserted, everything in, and with Real-time Protection turned off)

Doesn't get any better than 10:23.
I will reinstall OS again soon and will try two Raid0 systems of 2 and 3 SSDs, and will then run it again.

 

[ronvessGL76]

Wow ! AMD vs Intel & more! Raid in current day systems has a minimal benefit, if any. Next, shutting down SMT in AMD has a minimal benefit, if any & only in certain games.
Current SSD drives negate the need for Raid drives. Bench mark programs, sure. For me I'm more concerned how my games play out & survive a Windows install.

 

[mimarsina158b02db]

So it doesn't seem that CoreDirector (and Process Lasso) are very much compatible with Arrow Lake processors yet:
- none of the 3 enforcement methods have any effect on the Pre-Compiled Runtimes Tool - always shows enforcing on 0 (and I run it just in case and results were the same as without using it)
- for 3DMark it does show as enforcing, but results are the same or worst as without using it

- With enforcement method "Efficiency Mode OFF": (same results as before)
View attachment 201973

- With enforcement method "CPU Affinities": (worst results than before)
View attachment 201974

- With both enforcement methods "Efficiency Mode OFF" + "CPU Affinities": (worst results than before - same results as with "CPU Affinities")
View attachment 201975

- With the least restrictive enforcement method "CPU Sets": (same results as before)
View attachment 201976

Time Spy CPU test is still not using the full capabilities of this CPU.
So unless I can desable e-core in BIOS and test, for now it seems that using core-parking software like CoreDirector is useless - Arrow Lake is still new tho.

On the other hand, my current Win OS is installed on a Samsung 9100 Pro in the gen5 slot - with speeds below (not yet Raid0) - and I have run the Pre-Compiled Runtimes Tool.
View attachment 201978

I installed InstallAware from Web installer with full features instead of default minimum setup, and the Pre-Compiled Runtimes Tool now always shows 199 runtimes available:
View attachment 201979 View attachment 201980 View attachment 201981
(runs with battery inserted, everything in, and with Real-time Protection turned off)

Doesn't get any better than 10:23.
I will reinstall OS again soon and will try two Raid0 systems of 2 and 3 SSDs, and will then run it again.

That is a fantastic score!!!

Pretty sure the 128 GB RAM is to thank for that.

What is your peak RAM usage during the test run - does it go above 96 GB?

 

[mimarsina158b02db]

Wow ! AMD vs Intel & more! Raid in current day systems has a minimal benefit, if any. Next, shutting down SMT in AMD has a minimal benefit, if any & only in certain games.
Current SSD drives negate the need for Raid drives. Bench mark programs, sure. For me I'm more concerned how my games play out & survive a Windows install.

Glad to hear I'm not missing out by being unable to disable SMT.

Yes, I avoid RAID nowadays - more of a boot hassle than anything else - but there's no way around it sometimes.

Without merging in the RAID drivers into my OS image, I couldn't get it to boot on the Norse Titan - even when I wasn't actually trying to boot into a RAID array,

 

[joao.sousa.domingo158102d7]

That is a fantastic score!!!

Pretty sure the 128 GB RAM is to thank for that.

What is your peak RAM usage during the test run - does it go above 96 GB?

So, RAM is actually not used too much throughout the whole test. You can monitor it yourself.

Just before the 3min test mark there is a maximum peak of 99.1% of RAM usage - along with virtual memory usage as well (here the SSD will be relevant). But after 3 mins, RAM usage drops to 50%, and after that it drops to 23.6% for the remainder of the test - this is why, on average, for the entire 10 minutes of the test, RAM usage was just 52.1%.
In summary, it initially gradually climbs to max RAM up till 3 min, then after that drops to 50% for a while and eventually just to 23% for the remainder of the test.

 

[mimarsina158b02db]

So on my second production OS image, which has finally settled in - and is the actual main OS I use as my daily driver; on the first run of the benchmark, BOTH devices scored exactly 10:55. Amazing, right? EXACTLY the same score.

On the second run, the Intel bested the AMD by literally one second - 10:42 vs. 10:43. Mind blowing. This makes the first time I've seen the Intel exceed the AMD on this generation of hardware, with the above being its first example of meeting it.

Does anyone have an 14900HX with ample RAM to run these tests?I really can't wait to see if the higher GHz on the 14900HX bests both the 285HX and the 9955HX3D combined.

So, RAM is actually not used too much throughout the whole test. You can monitor it yourself.

View attachment 201985

Just before the 3min test mark there is a maximum peak of 99.1% of RAM usage - along with virtual memory usage as well (here the SSD will be relevant). But after 3 mins, RAM usage drops to 50%, and after that it drops to 23.6% for the remainder of the test - this is why, on average, for the entire 10 minutes of the test, RAM usage was just 52.1%.
In summary, it initially gradually climbs to max RAM up till 3 min, then after that drops to 50% for a while and eventually just to 23% for the remainder of the test.

Peak RAM still seems really high. This sort of suggests that even 128 GB is being saturated at the maximum demand point of the benchmark.

This is a really interesting finding.

Until we have (mobile) PCs with 192 GB RAM, I guess we just won't know at what point memory saturation won't be an issue for this benchmark. The figures they self-advertise under the memory pressure heading are purely theoretical, but not entirely reassuring in this regard.

 

[mimarsina158b02db]

Third round, and AMD strikes back! Intel 10:37, AMD 10:34.

Still an 11 second far cry from Joao's results, but with only 96 GB RAM, and with fully loaded systems running multiple background servers; I find these results more than admirable (not in the least when AMD manages to eke out a win).

Will run one or two more iterations, before I pivot to dedicated CPU testing in 3DMark, and then GPU testing outright.

 

[mimarsina158b02db]

Uh-oh...must be really stressing the system (although I have no overclocking enabled in the BIOS whatsoever).

I just relaunched the benchmark on the AMD, and it popped up without the OK/Cancel buttons! That's a first for sure.

 

[mimarsina158b02db]

Fourth run: Intel 10:47, AMD: 11:02.

 

[joao.sousa.domingo158102d7]

Uh-oh...must be really stressing the system (although I have no overclocking enabled in the BIOS whatsoever).

I just relaunched the benchmark on the AMD, and it popped up without the OK/Cancel buttons! That's a first for sure.

hmm might not be 100% stable.. but if its not OC'ed anywhere don;t know what it can be.. running hot?

I did fully test my system before for stability, and I never OC'ed these laptops before: https://forum-en.msi.com/index.php?posts/2332855/

My runs can start like 1 or 2 seconds higher, but they all stabilise after to the best time. If any further run is worst is usually only by a second and nothing more.