Log FormD T1 meets MinisForum BD790i! (Build Log & benchmarks)

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
FormD T1 : SSD Heatsink trial (22th July 2025)
-------------------------------------------------------------------------------------------------------------------------------
Here is another trial inside the FormD T1, while using my Core i5 12600K setup.

It is time to try some heatsink on my verso SSD. By default, its idle temperature is not that terrible, i.e. 60°C while gaming (heat coming from the RTX 5090 FE). It is still safe, but as there is an offset with the riser, thanks to the 2-slots GPU, why not using this empty space to test some SSD heatsink.

First candidate will be the a Thermalright model, with bottom & top heatsinks :

Despite its size it is fitting nicely behind the riser. There is no clearance issue and I was able to put back the RTX 5090 FE…however I was unable to start the computer.

After some troubleshooting, I found the root cause : the bottom plate of the heatsink is making contact with the motherboard, creating some shortcuts. Hopefully the motherboard detected them and refused to power up.

Conclusion? It is suited to the recto SSD slot, not verso!

Second trial then.
I used the same heatsink as on my U-ITX setup : 3mm tall, held by rubbers.
It is also an easy fit!

PS : Note I found the best way to assemble my RTX 5090 FE inside the FormD T1. Indeed it can be tricky to put the entire screw through the GPU bracket. Finally the best way is simply to unscrew partially (you can see on above photo), and then to rotate the GPU bracket. Very simple & effective!

This time, no booting issue!

Time to test!
Best trial is a simple 3DMark Speedway torture test with a stock RTX 5090 FE (575W)!


The result is without surprise : -5°C versus without the heatsink…basically the same result as the U-ITX..

One interesting point is that the SSD temperature rose up to 58°C (and decreased quickly) after the benchmark finished…the GPU airflow (in/out) is finally well used by the heatsink..

PS : I’ve moved back my RTX 5090 FE and the SF1000 inside the Cerberus-X in order to define a truly stable undervolt of the RTX 5090 FE…but not only as I suppose that the remaining instabilities are coming from either my undervolted 12600K, XMP profile or the Nvidia drivers. The issue is that I need RTX 5090 FE to be fully utilized…and the R7 9800X3D is way better at this task. Based on my first testing, I measured a gap of 30%, even at 5120x1440 in Stellar Blade, especially in open areas.



More to come…

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
FormD T1 : System Stability (at last) & 6mm SSD Heatsink & comparison vs my Cerberus-X (27th July 2025)
-------------------------------------------------------------------------------------------------------------------------------

Debugging the instability!

My Cerberus-X setup with its R7 9800X3D proved very helpful as it allowed me to find the proper undervolting parameters for my RTX 5090 FE! So far no more crashes due the GPU in Stellar Blade, either in the Cerberus-X or the FormD T1.

However, Stellar Blade was very stable on my R7 9800X3D, but it was not the case with my FormD T1. After thinking, I supposed I could use the error message shared by the Unreal Engine (I got them on either Expedition 33 : Clair Obscur or Stellar Blade). Finally games using the Unreal Engine could be very useful to debug an unstable PC.

Here is my understanding (so far) on the Unreal Engine Debugger :

• Error Message : “FATAL ERROR”
  • This is translating an unstable computer, mostly on the CPU.
  • I got this message on both my R7 9800X3D (Curve Shaper -30/-10) and my Core i5 12600K (undervolt -0.085v). By reducing or removing the undervolt, instability disappeared!
• Error Message : “Lost Display….long message”
  • This is translating an unstable computer due to the GPU.
  • Once again, it appeared on both of my systems. By increasing slightly the tension (from 900mV to 910mV) removed the message on both systems!
• Error Message : None
  • The Unreal Engine debugger is not catching the error. I suppose it could be linked to either a Game Bug (if repeatable) or RAM stability issues (random)
  • Keep in mind, it is only my guess as the system was unable to catch the error nature.

Here is a long gameplay capture on my FormD T1, perfect!



6mm-height SSD heatsink

Very quickly I also tested 6mm-height SSD heatsink instead of my initial test with 3mm-height on my verso SSD :

Does it help? Quick answer is Yes by a 2°C additional decrease on the SSD temperature (using 3DMark Speedway torture test).
Here is a summary of the temperatures on the back/verso SSD, while heavy GPU load (575W) :

• Without heatsink : 60°C
• With a 3mm-height SSD heatsink : 55°C
• With a 6mm-height SSD heatsink : 53°C

FormD T1 vs Cerberus-X, initial comparison

After one month of intense use, it is time to make a first comparison between my beloved Cerberus-X and the FormD T1.

• AIRFLOW / COOLING / NOISE :
  • FormD T1 : Dual Chambers (CPU & GPU) design, herited from the Dan A4 case, is very, very effective, especially with the RTX 5090 FE 2-Slots passthrough design. Leaving a gap of 20 mm between the GPU and the motherboard/PSU tray is great for cooling as it allows an excellent airflow inside the case, without overheating the motherboard/PSU!
    With a 3-slots GPU, the case airflow is too restricted (based on my trial of the RTX 4080 FE).
    The 2 top exhaust 120 mm fans (Corsair RS120 Max + Noctua Offset Grid) are very effective.
    Overall the FormD T1, with my setup, can handle a stock RTX 5090 FE (575W), even if an undervolt (510W, ie 0.895mv) is a better compromise performance/noise for me.
  • Cerberus-X : With my latest cooling setup (dual fans bracket just in front of the GPU), the Cerberus-X is keeping the crown of the GPU cooling, either on gaming load & on idle (keeping the GPU fans inactive).
    My only issue remains the relatively poor (below my expectations) performance of the Noctua C14S / D12L on a R7 9800X3D. I got very little improvements by increasing the fan speed from 1000 rpm to 1500 rpm (1-2°C improvement on the CPU core). There is clearly a bottleneck on the heat dissipation of the R7 9800X3D. Neither Noctua nor the Cerberus-X are to blame here.
• MOTHERBOARD
  • FormD T1 : Mini-ITX motherboards have been a long topic of frustration for me, i.e. poor longevity, i/o, aso. I must admit, since 2024, the average level of mini ITX motherboards is improving, but slowly. On the AM5 side, B850 models are pretty decent to good (like Asus B850i or MSI B850i). But so far X870 models are pretty dull (Asus or Gigabyte). I’m putting high hopes in the future MSI X870i with its 4 m.2 slots, but let’s see.
  • Cerberus-X : Simply the best as it is compatible with ATX motherboards.
• MAINTENANCE
  • FormD T1 : It is a very important point for me. Well the FormD T1 is very easy to maintain. Thanks to its design, it is possible to open it while still powered/plugged in. Its tool-less side panels are a must.
  • Cerberus-X : It is still my reference for maintenance. Every maintenance operation is easy & fast. It is simply the best!
• EVOLUTION
  • FormD T1 : Simply the best. The FormD T1 managed to evolve overtime, while keeping its initial overall dimensions (10L). I loved its modular (Lego) & smart design. The only drawback I find is the bad handling manual.
  • Cerberus-X : My biggest criticism of the Cerberus-X would be its almost complete lack of evolutions over the last 8 years. This is a common criticism with other great Sliger cases like the SM550/560/570/580 or the S610/620. They were ahead at their release date but now they are outdated versus competition.

As a conclusion, my FormD T1 has managed to divert me from my Cerberus-X. The U-ITX managed the same but fell short due to some points, one of them was the mini-ITX motherboards versus ATX ones, especially for m.2 slots. Depending on the MSI X870i mini-ITX MB, I might consider switching my main setup to the FormD T1 as it is working very well with a RTX 5090 FE, undervolted or not.
Meanwhile I’m also frustrated with my R7 9800X3D. It is very true that it is faster than my i5 12600K, but it is also way harder to cool, despite using less power (W).

So let’s see…

 

[MarcParis]

For information I've just stopped the trial of the FormD T1 with a 12600K & a RTX 5090 FE. Overall, it is a viable combo with some standard drawbacks (mini ITX Form Factor, hotter temperatures & louder overall) but nothing critical/prohibitive.

For now I'm back to my Cerberus-X as I have an interesting experimentation to do...

Stay Tuned !

 

[MarcParis]

Well, well, as I just finished Stellar Blade for the 3rd time (4th run ongoing) and I don't have just yet another game to platinum, I'm considering to solve those instabilities issues met on my 12600K mini ITX.
After some thoughts I guess the issue is linked to the DDR5 speed, ie 6000MTS. Alder lake is only guaranted up to 4800MTS with DDR5 while Raptor Lake is 5600MTS. Thus I suppose those random issues I faced on Stellar Blade are linked to my DDR5, especially as Stellar Blade is pretty gourmand at RAM (~22-23Go). Issue is not the RAM itself but the RAM controller I would say.

I will let you know!

 

[MarcParis]

What is happening to FormD T1 (Ncase) : Why are they changing the orientation of their gen 5 riser??! This making the T1 no longer compatible with RTX 5080/5090 FE models!

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
Minisforum BD790i : R9 7945HX : Time to re-test it (20th December 2025)
-------------------------------------------------------------------------------------------------------------------------------

Background

As you might remember, when I got my RTX 5090 FE (June 2025), I tried to pair it with my Minisforum BD790i but I experienced a nasty bug : shutting down the computer completely while under load. Since I did not have time to test my BD790i properly.

This type of bug could be linked to hardware or software (driver) issues…especially as I experienced it with my RTX 4080/5080 with the “not so great” Nvidia drivers from the “blackwell” release

Why am I doing it right now? The main reason is I still think the minisforum motherboard is the best implementation for mini-ITX form factor :

• CPU cooling : 120mm fan, 50mm tall (15mm fan) - 60mm tall (25mm fan) with direct exhaust to the rear (Screws M2.5 thread)
• SSD Cooling : Dual SSD x4 gen 5 on the recto of the motherboard, with good heatsink, vented by the CPU fan
• No additional Chipset : “Straight to the point” architecture, allowing significantly faster booting time and better cooling of components. The downside is that the i/o is limited to the CPU capabilities.
• HDMI 2.1 with 120hz output : pretty unique feature on mini ITX AM5 boards.. The others are always limited to 60hz. This feature could be very useful for a gaming streaming system. I still have to check it.

Bench table

As a starting point I went to the test my setup outside any case, with this hardware :

• Motherboard : Minisforum BD790i (Bios 1.12, manual CMOS by unplugging the battery)
• CPU : AMD R9 7945HX
• RAM : Crucial (RIP) 2x32Go 5200MTS (CAS 42-42-42)
• SSD : Samsung 950 Pro 512 Go (10 years old)
• PSU : Corsair SF750 Platinum
• CPU Fan : Fractal Design HP-12 Venturi (great fan)

The installation went smoothly, in general, but with one issue :
After the Windows update I experienced some blue screen of death (BSOD) because of the hiberfil.sys crashing at a “cold” start. It was happening 100% of the time. Hopefully the fix was simply to force a hard shutdown...restart and “Voilà”!
The root cause was that Windows 11 was not shutting down the computer but was putting it in a deep sleep mode instead. The easy fix was to disable the deep sleep with this command : “powercfg -h off”
Job’s done and the booting time is still very fast.

I tested the CPU with Cinebench 2024 and it went pretty well! Single Core : 117 points and Multi Core 1710 points.


Sliger S620

Time to move the setup inside a proper case. As I wanted to exclude any riser for now, I chose to use my Sliger S620 as a bench case with the capability to plug a discrete GPU.

Here are the hardware changes :

• CPU Fan : Noctua NFA12x15 Chromax (already anticipating the next case switch..)
• Exhaust Fan : 2xFractal Design HP-12 Venturi on the side bracket
• Discrete GPU : My “almighty” MSI GTX 1080 Ti Gaming X (8 years old)!

The build is looking awesome, despite using only some spare parts I have…

Everything went smoothly, again!

I tried also to play the “furnace” Stellar Blade Demo and it performs great (relatively speaking, 60fps, FSR3) :



Once again this combo is really quiet, even under load (270W on the GPU, 100W on the CPU)

I’ll continue to push & test the system, the obvious next will be obvious with the pairing of my RTX 5090 FE…but it will have to wait for a second SF1000.

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
Minisforum BD790i : R9 7945HX + RTX 5090 FE (23rd December 2025)
-------------------------------------------------------------------------------------------------------------------------------

Build update

I finally received a second Corsair SF1000. As it was a second hand (terrible packaging, but the parts themselves were in good shape) I checked carefully all the cables, the connectors and especially the terminals, to see if there were any visible anomalies. I found nothing suspicious.

I tried it quickly with my test setup, inside the S620, ie BD790i + GTX 1080 Ti…everything happens normally!
Then it was time for the real trial, testing the RTX 5090 FE stock, using 575W!

Compared to the previous build :

• CPU Fan : I swapped it to the 25mm Noctua version (NFA12x25), for more air pressure
• GPU Stand : I tried the GPU stand, which is effective while the S620 is in vertical stand, its magnetic base sitting on the steel frame. However in horizontal stand, it is too hazardous as the heat generated by the RTX 5090 FE is smoothly a bit all the materials and the final grip is no longer enough and the GPU stand can fall (impossible while in vertical stand)
• PSU : Corsair SF1000 with its stock cables

I love looking at the RTX 5090 FE from the bottom of the case…it is almost mesmerizing.
Note it is possible to fit 2x25mm thick fans at the bottom of the S620. That would help keeping the fanless mode of the RTX 5090 FE at idle, but not its cooling. However that would have a major downside on the maintenance as it would require first to remove the fans, before removing the RTX 5090 FE itself.


Thermals

Let’s start with now my standard “torture’ speedway test, 20 min at full GPU load, no CPU bottleneck



Thermal test outcome :

• R9 7945HX Max Stabilized Temperature : ~93°C with fan at 1700 rpm, running at 100W. No thermal throttling reached as the maximum temperature of the R9 7945HX is at 100°C. However I had to switch to the 25mm thick fan instead of 15mm as with this later the CPU temperature was reaching 97°C at the middle of the test (I stopped it there).
• RTX 5090 FE Max Stabilized Temperature : ~76°C with fan at 1623 rpm, running at 550-575W. In comparison, with the Cerberus-X the RTX 5090 FE is at 74.6°C at 1600 rpm and the FormD T1 at 80°C at 1800 rpm in similar conditions.

(You can find the link to the different videos on the youtube description)

Overall thermals are pretty nice. Inside the S620, obviously the RTX 5090 FE is heating up the CPU, despite a pretty nice exhaust system, buit the global system remains very effective.

The RTX 5090 FE cooling is very effective inside the S620, achieving a pretty similar performance/noise ratio as the Cerberus-X. The FormD T1 already showed it is noticeably less effective in cooling and noisier compared to the S620/Cerberus-X.
Also it is clear that bottom intake fans (up to 25mm thick) on the S620 are not required at all in order to cool the RTX 5090 FE under load. However they can help to keep its fanless mode forever (like on the Cerberus-X), but at the cost of longer maintenance as it will be mandatory to remove the bottom fans before removing the RTX 5090 FE.

I’m even thinking that a C-shape cooler like the Thermalright AXP120-67 or the ID Cooling IS67-XT should work very well while allowing very good maintenance level in the S620, compared to tower cooler that would require to be disassembled before removing the RTX 5090 FE.


R9 7945 HX, is it a Gaming Bottleneck??

Short answer is “Yes BUT” not really.
The R9 7945HX could be a true bottleneck for some games, in specific area…but on the other hand, the downside is not killing the gaming experience and for some tasks, like shaders compilation, the R9 7945HX is even better than the R7 9800X3D.

Let’s start with Stellar Blade. Depending on the area, the bottleneck from the R9 7945HX could damper the framerate by ~0%-40%. The journalists will focus on the -40% to highlight the power of the R7 9800X3D, but those areas are not impacting combat (like the city of Xion -40%, reaching 83 fps). In open areas, like the great desert, the impact in fps may seem important with similar loss of ~40% / 40f ps but the achieved framerate is still within 110 fps, really smooth & stable. In closed areas, negative impact is limited (-0%/-15%).
Overall my comment is that the R9 7945HX is not limiting, at all, the gaming experience, even in Stellar Blade, where the UE4 engine is demonstrating its ugly/terrible CPU optimization.
It seems Unreal is working on it, in collaboration with CD Projekt for their new open worlds.

Example of the “worst” case scenario for the R9 7945HX in Stellar Blade (Great Desert, Xion)


Example of the “best” case scenario for the R9 7945HX in Stellar Blade (Eidos 7)


Overall, the R7 9800 X3D is clearly acting as a mitigation tool to hide the pretty bad optimization of the game engine.


On the total opposite, we have Assassin’s Creed Shadows, with its homemade “Anvil” game engine, dedicated to the open worlds and the impact of the R9 7945HX is basically 0. The RTX 5090 FE is used at its maximum, even in crowded areas.



Overall the gaming experience with the combo R9 7945HX + RTX 5090 FE is really good, stable, smooth. My opinion remains that the X3D cache is a patch for not optimized game engines (unfortunately the very popular Unreal Engine is still part of them). My gaming experience is based while using a 4k monitor up to 240hz, solo gaming. I’m completely unable to comment on other use cases, like 1080p 500hz, ultra fast multiplayer fps.

 

[MarcParis]

------------------------------------------------------------------------------------------------------------------------------
Back to FormD T1 12600K + RTX 5090 FE (23rd December 2025)
-------------------------------------------------------------------------------------------------------------------------------

Before moving to my next project I decided to try again my 12600K with the RTX 5090 FE, especially to compare it vs the R9 7945HX / R7 9800X3D and to solve my instabilities issues.

Regarding the instabilities, I made up my mind that it was finally related to the DDR5 speed. My hypothesis was that my 12600K was unable to keep up to 6000 MTS in some specific cases. Thus I reduced its speed to 5600 MTS, while keeping the same subtimings as the XMPs. Meanwhile I restore my 12600K undervolting to -0.085v.

In terms of the CPU performance I can simply conclude that the 12600K is at the same level as the R9 7945HX in terms of FPS, but the R9 7945 HX is WAY better in terms of loading / shader compilation times.
You can compare the 3 CPU with those 3 videos (timestamp) :

AMD R7 9800 X3D : 130 fps


AMD R9 7945 HX : 83 fps


Intel Core i5 12600K : 81 fps



At last, here is the thermal test with the RTX 5090 FE undervolted on Stellar Blade (Demo).



As reminder, here is the full setup specs :

• Case : FormD T1
• CPU : Intel Core i5 12600K (UV -0.085v)
• RAM : Corsair Vengeance 2x16Go DDR5 6000 MTS CL30 (XMP I profile) BUT limited to 5
• 600 MTS for stability concerns
• CPU Cooler : Thermalright AXP90-CU + Noctua NFA9-14 (2200rpm Model)
• RTX 5090 FE : Undervolted
  Core Curve : 825-910 : 2700 MHz
  GDDR7 +3000
  Power limit : 100%
• Top Fans : 2xCorsair RS120 Max (exhaust ~1400rpm)
• Fan Grids : Noctua NA-FG1-12 Sx2 with offseted mount
• Custom fan curve : Yes ("Fan control" Software)
• Nvidia drivers : 591.59

 

[MarcParis]

------------------------------------------------------------------------------------------------------------------------------
FormD T1 : Road to Linux : 12600K + RX 9070 XT (14th February 2026)
-------------------------------------------------------------------------------------------------------------------------------

Following the late massacre of Windows 11 bugs, like Recall, abuse telemetry (almost spyware at this stage), broken Outlook importation in POP3 mailbox, FPS divided by 2 like in Assassin’s Creed Shadows, I’ve decided to seriously explore Linux as an alternative to Windows 11, and overall Microsoft products. I also have some resentments against Nvidia, but it is not as strong and one step at a time… I also added to this project to find a French-EU secure cloud service that works both on Linux & Windows & Mobile.

For this project, I decided to use my setup within the FormD T1, based on the Core i5 12600K.

I initially used my MSI GTX 1080 Ti Gaming X, thus I decided to move back the riser of the FormD T1 to its original position…and the first issue appeared…


The standoffs of the Thermalright AXP90-47 Cu were punctionning through the riser itself!! This was literally blocking the plugging of the graphic card inside the PCIe slot! This simply means that the APX90-47 is not compatible with a FormD T1 with its original layout, it is incredible to see people recommending it…it is simply a non-sense! I always criticized Thermalright coolers for their mediocre mounting mechanism…for the AXP90-47, it is terrible!




Thus I decided to switch to the ID-Cooling IS-53-XT, I bought it last summer. I found it interesting for the FormD T1 as it is using a 17mm thick fan. Thus if I use it with a 14mm tall fan, like the Noctua NF-A9x14 the CPU cooler will reach only 50mm height, avoiding most of the air turbulence while measuring 53mm.

Like all the other ID-Cooling cpu coolers I used, the IS-53-XT was very simple to use. They even improved the assembly process, by using the same standoffs for both AMD & Intel.




As a simple spoiler, I would say this combo of the IS-53-XT heatsink + a Noctua NFA9-14 is one the best combo as cpu cooler for the FormD T1 in the 3-slots GPU configuration.

I also noted that using a GPU in the original position of the FormD T1 is preventing the use of any m.2 SSD at the back of the motherboard. It is not because it won’t fit, but it won’t benefit from any airflow or any cooling. That’s why I still recommend having a mini-ITX motherboard with both SSDs on the recto side, like the Asus X870i or the minisforum BD79xi.

Let’s speed up a little bit..

After the build, I decided to test different distros of Linux, with a simple installation on my loyal 950 Pro 512 Go. No dual boot for my early tests.
Having a GTX 1080ti was pretty cumbersome as some distro crashed during the installation, like CachyOS. I tried Bazzite Gnome, Bazzite KDE, Ubuntu 25.10, CachyOS, ZorinOS. From those trials, I appreciate Bazzite & Ubuntu with their easier approach, especially to install apps (Flatpack for example). At the end I selected Bazzite KDE as it was handling my 32:9 way better than Gnome.

Bazzite was also the distro that is able to launch a steam game out of the box. Then I tried the demo of Stellar Blade, as I already tried it with my GTX 1080ti in Windows 11…and it was pretty terrible…45 fps on average…ok…but 15fps on 1% low…
I also tried Doom 2016, it was also pretty ugly and finally crashed…
Unfortunately the Nvidia drivers for my GTX 1080ti on Linux are simply terrible. Also to mention that my screen was flickering like hell during the installation process…but it was back to normal once installation completed.

My Linux experience could have stopped here, but I decided it was the perfect opportunity to upgrade my GTX 1080ti with a more potent & capable GPU, that is also working well on Linux…thus I decided to go for the Asus Prime RX 9070 XT, as its price remains reasonable, unlike the RTX 5070Ti/RTX 5080.









Thanks to its 3x 8 pins power connectors, the RX 9070 XT is a tight fit inside the FormD T1, but it is completely fine at the end.

After the addition of the RX 9070 XT, my experience with Linux was largely improved…but even if Bazzite remains an easier start, it is hard to master. I spend all my evenings / free times to replicate what I was able to do on Windows. Hopefully, I discovered new paths and I was delighted with my progress. The list of “can’t” remains very small versus the list of “can” and there are some items as “surpass”.

Let’s only list the “can’t” :

• HDMI : no adaptive sync and other advanced features thanks to the hdmi consortium that lock them on open source drivers…some kind workarounds are their way, let’s see
• Hardware monitoring : You know my love for fan & temperature monitoring & management...due to the lack of drivers I cannot monitor my motherboard fans. It is a pity, especially as there is great software to manage their fans. I need to rely only on my motherboard bios. But let’s note that even on Windows, this fan management arrived only late for the last generation of AM5 motherboard (2025). On top of that I already have a workaround I’ll share below.
• Firmware updates for Samsung SSD & Dual sense controller (closed programs) : Even with wine, it is a dead end. I still require Windows, either as dual boot or as a virtual machine.

Except that, everything is working pretty well. I noticed some huge effort on ease of program installation like Flathub or appimages format.

My main achievement was to free myself from the shackles of my Outlook pst archive (12go file). This point alone took me one full day to find a way to convert, with no loss, to Mbox / Thunderbird. Cherry on the top is that I found a proper FR-EU cloud service (Shadow Drive) that worked well to store the new email archives.

Before moving to the gaming section, I needed to get rid of the lack of the MB fan management on Linux, due to the missing drivers. For that I simply applied the same technique as for Windows, when Speedfan / Fan control were / are not working : using a thermal probe connected to the MB to control the fans… Both my mini-ITX motherboards, ie Asus B760i and X780i have one thermal sensor pins…thus let’s use them!




For the FormD T1, I placed it on the PSU, in front of the exhaust of the RX 9070 XT. It is more delicate to set up the fan curve…as I cannot monitor the thermal probe in Linux. However, thanks to my former experiments, I know that the temperature while the GPU is under load should be between 50-60°C inside the FormD T1. Thus I managed properly the top exhaust fans.

Time for some gameplay videos…even on that part I had to evolve, as I was really, way too much relying on Nvidia Shadowplay to record them.
Speaking of the game setup, even applying the FSR4 was a cumbersome path, as it required a lot of understanding how the proton is working..



Overall, I’m really pleased with my current project, as it is successful !

• Bazzite is a great distro, easy to start but still hard to master.
• AMD RX 9070 XT : it is working pretty well, trading blows with the 5070ti, silent at idle, cool & quiet under load.
• Outlook shackles gone !

More to come on my journey to my journey to Linux!

 

[rfarmer]

Nice testing. I still can't convince myself to try Linux but maybe soon. You are right about the AXP90-X47, the weird thing is you go to the FormD sub on Reddit and those guys swear by that cooler, recommending it above pretty much anything else. I know it's good for it's size but I have also seen way too many posts about the poor mounting.

 

[MarcParis]

Nice testing. I still can't convince myself to try Linux but maybe soon. You are right about the AXP90-X47, the weird thing is you go to the FormD sub on Reddit and those guys swear by that cooler, recommending it above pretty much anything else. I know it's good for it's size but I have also seen way too many posts about the poor mounting.

For sure I would have not tried Linux with my main desktop directly. After 2 weeks it could be possible but not yet too. My main issue is that I cannot get rid at 100% of windows 11.
I've now moved to a dual boot setup and I can start a full comparison. Proton is far from being free on the performance. On the image quality side I also have some doubts as I'm lacking some monitoring tools..
As a simple example, I launch FSR3.1 on Stellar Blade on both Win 11 & Bazzite. In terms of image quality, it was ok for Win11 but terrible on Bazzite...where FSR4 has more significant improvements.

I don't regret my Linux experiment as I'm learning tons of stuffs, either on Linux, Windows 11 and myself (my limits).

 

[Skripka]

Nice testing. I still can't convince myself to try Linux but maybe soon. You are right about the AXP90-X47, the weird thing is you go to the FormD sub on Reddit and those guys swear by that cooler, recommending it above pretty much anything else. I know it's good for it's size but I have also seen way too many posts about the poor mounting.

AXP90 is a great little cooler...depending on what you're wanting to cool. That being said the stock fan is only okay, and finding a similar low-profile fan is a problem. I had one way back for a bit in my NCase days.

Just give Bazzite a go. It is easy to psyche yourself out. With a 5080, Nvidia might be a turd bucket. But then again that is what Windows and its 10,000 Copilot nag buttons are. So pick your poison LOL

 

[MarcParis]

AXP90 is a great little cooler...depending on what you're wanting to cool. That being said the stock fan is only okay, and finding a similar low-profile fan is a problem. I had one way back for a bit in my NCase days.

Just give Bazzite a go. It is easy to psyche yourself out. With a 5080, Nvidia might be a turd bucket. But then again that is what Windows and its 10,000 Copilot nag buttons are. So pick your poison LOL

The issues of the APX90-47 are concentrated on the mounting mechanism (socket & fan), pretty terrible. The worst is that the AXP90-47 is not compatible with the FormD T1 or pure sandwich style case as it can block the PCIe slot..

 

[MarcParis]

------------------------------------------------------------------------------------------------------------------------------
Linux Experience : AMD, Intel, Nvidia (01st March 2026)
-------------------------------------------------------------------------------------------------------------------------------
Following the successful first trial with Bazzite, I decided to deepen my trials. Thus the first step was explore the dual boot Bazzite - X11 (Windows 11)

After installing X11 on a 500Go partition it was “relatively easy” to install Bazzite on the remaining 1.5To of the SSD…however with the following prerequisites :

• Follow strictly partitions guide for each Linux distro (Bazzite & CachyOS have different partition architecture)
• On Bazzite, potentially re-enter the key for the secure boot (enroll MOK)
• Be able to see the Bios… It may sound very stupid but with my RTX 5090 FE, my monitor (Samsung Neo G9 mini LED) is not displaying the Bios by default (It is working just fine with my RX 9070 XT). The work around is simply to display the monitor menu during the boot sequence with my RTX 5090 FE…

The experience was very positive as I was able to install/replace my Linux installation while maintaining the X11 partition fully functional.

Once again, after this positive trial, I finally decided to test Linux on my main setup (9800X3D+RTX 5090 FE) and build my own experience on the Nvidia GPU on Linux.

As for the FormD T1 build, I relied on an external thermal sensor to monitor the GPU temperature and setup the fan speed following its reading. X11 was useful to monitor everything.




Overall, installing Bazzite on a disk fully setup for X11 was very easy. The Partition tools on Linux are definitely at another level compared to the X11 base tools.

However I faced a new issue versus my Intel setup, related to the absence of Linux drivers for the Mediatek MT7927 Wifi 7 - Bluetooth card. There is ongoing reverse engineering to work around this issue, let’s see. The only Wifi 7 Linux compatible cards are from Intel, but locked to Intel CPU, so far. One solution would be to replace the Mediatek M.2 card with an Intel AX210, but it is not that obvious on the X870i. Let’s see.

So compared to X11, what am I losing while gaming on Bazzite? Those tests were done the Nvidia Drivers 590.48.01

• Fanless Mode on idle? Well no. Make sure you select “Adaptive Synchronization” to Automatic. The RTX 5090 FE is able to lower it power consumption to 25W approx, like on X11 and I’m able to get the same fanless mode as on X11
• Undervolt the RTX 5090 FE. This can be also achieved through LACT, with the same method as on MSI Afterburner on X11. Just apply GPU offset +700Mhz while fixing the max GPU speed to 2700Mhz, and voilà! Cherry on the top is that you can also overclock the GDDR7 to 17000MTS like on X11. The only drawback is that you cannot setup the curve like on MSI Afterburner to avoid the offset frequencies on idle, leading to a small power consumption increase (~2-3W at idle).
  
• What about DLSS 4.5 or DLSS Indicator? Well this one was almost easier than on X11 as you can force a DLSS preset & indicators as environment parameters. It is showing the same information as on X11.
• What about video recording? Well, as for AMD RX 9070 XT, I switched to OBS successfully, using Nvidia codecs.
• What about the gaming performances? That was quite a surprise as most of my recent games worked pretty well but others don’t.. It seems that Unreal Engine games are working very well.
  
  • Here is the list of games tested with my comments :
  • Stellar Blade (UE4) : Running like a champ!
    
  • Wuchang Fallen Feathers (UE5) : Running like a champ!
    
  • Echoes of the end (UE5) : Running like a champ!
    
  • A Plague Tale Requiem (Asobo) : Running like a champ!
    
  • Dragon Quest VII Reimagined : Running like a champ!
  • Assassin’s Creed Shadows (Anvil) : Running like a champ, including the ray tracing!
    
  • Star Wars Outlaws (Snowdrop) : The game is launching but not playable due to strong stutters (the same as AMD RX 9070 XT)
  • Resident Evil 9 Requiem (RE engine) : The game is launching but not playable due to strong stutters and graphical glitches. Ray Tracing option is not available on Linux. On the RX 9070 XT, RE9 is working great, but no ray tracing option either. I expect a Nvidia driver updates will make the game playable but still without Ray Tracing option, as this one is linked to the RE engine itself
    
  
  

I tried some benchmarks to compare X11 vs Bazzite performance on the RTX 5090 FE and the RX 9070 XT. When the game is working well on Bazzite, there is always a performance hit versus X11.
I summarize it on this video, of Stellar Blade the Demo :


The outcome is :

• Nvidia RTX 5090 FE
  • FPS : 128 (X11) vs 110 (Bazzite), -14% (Bazzite / X11)
  • Frametime : 7.8ms (X11) vs 9.1ms (Bazzite), +17% (Bazzite / X11)
  • RTX 5090 Power : 570W (X11) vs 528W (Bazzite), -7.4% (Bazzite / X11)
  • FPS / Frametime at the same GPU power : -7% / 10% (Bazzite / X11)
  • The 1% lows are behaving differently and it is difficult to conclude, but the value was -25% (Bazzite / X11)
• AMD RX 9070 XT
  • FPS : from -14% to -3% (Bazzite / X11)
  • RX 9070 XT Power : no significant gap
  • 1% Lows : from -10% to -25% (Bazzite / X11)

Overall, there is a performance hit between X11 & Bazzite, either on AMD & Nvidia GPU. I was expecting way worse performance from the Nvidia GPU on Bazzite, but I was wrong.
But my main outcome is that we can play very well on Linux, even though there are still some bugs. The performance impact is negligible as long as your fps (incl. 1% lows) entered into the VRR range of your monitor.
Thus I’ll keep a X11 partition on my setups to bypass any issues.

Let’s conclude for today by this video, RX 9070 XT on RE 9 Requiem, on Bazzite at its release date, pretty convincing. Hopefully the Ray Tracing is not really impactful on this game..

 

[MarcParis]

While experimenting on my next build inside the FormD T1, a strange stuff occured to me...a first in 30 years of building PC...

I always promoted aircooler to be more reliable than AIO...I'm still right as it occured only once in 30 years...but it can happened...

Before discovering it, I experienced unstable PC with hard reset, very hot R7 9800X3D (even limited to 65W), very loud or compute errors. After installing Bazzite, CachyOS...twice on two different SSDs I decided to check again the hardware...

Look at my IS-53XT ! The cold plate was desoldered from the heatpipes!!

At least thermal paste was properlly in place!

I just replaced the IS-53XT by its little brother the IS-47XT and my setup worked well again!
Versus an AIO failure, the consequences were nothing!

 

[rfarmer]

Oh wow I saw that on Reddit but didn't realize it was you, that's wild

 

[MarcParis]

Oh wow I saw that on Reddit but didn't realize it was you, that's wild

This event is big enough to be broadly shared..
I just bought a second IS-53-XT as it worked well...when it was in one piece..

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------

FormD T1 + R7 9800 X3D + RX 9070 XT : Try hard ! (03th May 2026)

-------------------------------------------------------------------------------------------------------------------------------

With little surprise, as I already shared my experience with my IS-53-XT, I’ve decided to upgrade my FormD T1 v2.5 (Ncase) setup.

Why? Well, since I’m using my RX 9070 XT inside the T1, I’m impressed with its ratio performance / noise, especially under Linux as I’m able to tweak its fan curve, while keeping the fanless mode, without any additional fans, like on my RTX FE GPUs.

Also, I noticed that with my previous 12600K, there was a noticeable performance hit on Linux versus Windows 11, on some games. For example, in Solasta 2, I was reaching 83-84% of RX 9070 XT use versus 99-100% in Windows 11.I supposed that, maybe, the 3DV cache of the R7 9800X3D could help with the proton passthrough.

Thus I used my Asus X870i as a motherboard that will allow proper 2xm2 and 2x USB4 support.

I ran some early trials. Despite my IS-53-XT issues (just before it collapsed),I was able to reach 97-98% GPU utilization on Solasta 2, thus a pretty improvement versus the 12600K.

Overall my early trials were positive. Thus I decided to upgrade my FormD T1, not only with the R7 9800X3D, but also with a new riser Gen 5 and main SSD. While parts would be delivered, I continued my experiments with the IS-47-XT with my R7 9800 X3D limited at 65W PPT.

Time to upgrade!
I switched the original gen 4 (19cm) riser to Link up (AVA5 PCIe 5.0) Doubles Inversées (v3) 18cm, suited to FormD T1, with no GPU offset.

Clearly with the riser, I have no space to put back an 2cm offset to fit the RTX 5090 FE, however, I would not recommend it as the RTX 5090 FE will not be able to remain fanless on idle…and its fan, even at 1100rpm -30%), are distracting at idle, where everything else is silent.
I’ve tested the Gen 5 Riser on Windows 11 (GPU-Z and 3DMark PCIe) and they were all positive!

Next upgrade would be to put back another IS-53-XT. Compared to my first attempt, I guess what happened. In my first trial, I switched the original fan (17mm thick) to a Noctua NFA9x14 (14mm thick). By securing the screws, it pushed on the heatsink, creating some mechanical pressure on the heat pipe / cold plate. Thus this time I used IS-47-XT screws (using a 15mm thick fan) with my Noctua fan. Future will tell if I was correct, or not.
Here a comparison between the IS-47-XT and the IS-53-XT heatsinks.

Last update was the addition of a nvme SSD Gen 5, 9100Pro 2To. Hopefully I bought it at approximately the same price as its release (early 2025). I fear it will be even more inflated later.

Here is the final look :

As a side note, securing the GPU remains the most difficult part of the global assembly. Hopefully this GPU bracket remains in steel...

Here is the setup
Case : FormD T1 2.5
CPU : AMD R7 9800 X3D (CO -14), PPT 105W
RAM : Corsair Vengeance 2x16Go DDR5 6000 MTS CL30 (XMP I profile)
CPU Cooler : ID-Cooling IS-53XT + Noctua NFA9-14 (2200rpm Model)
SSD #1 : Samsung 9100 Pro 2To (2 Partitions, 1 Bazzite + 1 CachyOS KDE)
SSD #2 : Samsung 950 Pro 512 Go (Windows 11 25H2)
ASUS RX 9070 XT Prime : Quiet Bios + Specific fan curve (LACT, Target 76°C, max RPM 1650)
Top Fans : 2xCorsair RS120 Max (exhaust ~300-1400rpm) managed by either by Cooling Control (Linux)
Fan Grids : Noctua NA-FG1-12 Sx2 with offset mount
Custom fan curve : CPU / GPU - Exhaust Fan
Chipset Fan : forced to 0rpm
VRM Fan : not manageable
As an interesting point, I managed to push my R7 9800 X3D (CO -14) up to 105W, under Cinebench 2024 with the IS-53-XT. As I consider the Noctua NF-A9x14 (2200rpm version) not distracting at full speed, it is fine for me.

It is important to reach this CPU power, as more and more games are pushing the CPU, like Arc Raiders or Dragon Age : The Veilguard.

Some words about my Linux experience. Well I’ve switched now completely to Linux (mostly CachyOS, Bazzite as back-up). I’m just keeping Windows 11 on a separate SSD, mostly for firmware updates and also benchmarking (like Cinebench or comparing Win 11 with CachyOS in gaming).
In terms of gaming performance I’m impressed with my RX 9070 XT, especially with FSR4 compatible games. On CachyOS, the RX 9070 XT is generally on par with my RTX 5090 FE, on gaming. Even with an Nvidia GPU, gaming experience on Linux is really good, as I got only one non-compatible game so far, ie Solasta 2, with Nvidia Driver 595. With my RX 9070 XT, playing on Linux is really as good as on Windows 11. I would say even better as I’m able to manage my fan curve on Linux without losing the fanless mode on idle.

At last, mediatek 7927 (Wifi 7 + BT) Linux is coming and forecasted to Kernel 7.1.
But the biggest news is the official support of HDMI 2.1 by AMD (3 years to come!), with a first step by the Kernel 7.2.

To finish, here is the gameplay on The Expanse : Osiris Reborn!



Enjoy!

 

[rfarmer]

Very nice, glad it is working for you. How is that riser, any issues? I've been considering upgrading mine.

 

[MarcParis]

Very nice, glad it is working for you. How is that riser, any issues? I've been considering upgrading mine.

so far so good. 3D Mark PCIe tests were clear. With my RX 9070 XT from 23.22GB/s (Gen 4) to 35.37GB/s (Gen 5). As reference, my RTX 5090 FE connected in Gen 5 is reaching 53.56GB/s.

Just be careful on the exact reference. There is a similar model with bigger PCIe slot that does not fit.