Log SMALLEST ATX CASE : CERBERUS-X (Complete build log)

[hrh_ginsterbusch]

Thanks for the clarification, I was still trying to understand .

I know it was only about the PSU, but Machine and more component combo left me speechless... 500€+ ITX mobo, 500€+ PSU... M2 on air should be paired with mATX for better VRM cooling, more storage, more fans header ... and reduced cost.

I mean, its what I thought first, too. If you can stuff mATX inside, then why dont ..? Could be that, indeed.

On the other hand: Its a passive PSU. At the very start of the system, the fan of the X870 board jumps on, very loud .. so that would explain why the MB is supposedly "ruining user experience" if its trying "to support double CCD chips". A lot of (m)ATX boards dont have that, acc. to the OP. They stress in the post before, that the mainboard used is the "MSI MAG X870 Tomahawk (fully passive cooling, 5Gb ethernet, allowing 2xm.2 x4 gen 5 at the same time, without hurting GPU bandwidth)".

On that topic: Lets not ignore the fact, that ATX boards got much more headroom, because .. uhm .. there is more PCB space? So the comparison is kinda wonky.

Aside of the fact that you can still disable those fans - yes, in some cases, you need to unscrew the VRM etc. pp., and certainly unplug the fan, but still, nobody hinders you to do so. Those heatsinks are massivelly overspec'd anyway. Yes, in theory, you could OC your CPU, but I guess the chances for this are much lower in an SFF scenario than eg. with a fullsize ATX or even just a simple MFF build.

cu, w0lf.

 

[MarcParis]

took me a while to understand. for anybody else reading: OP is referring to the MB VRM fan. so that depends on your board.

Thanks for the clarification, I was still trying to understand .

I know it was only about the PSU, but Machine and more component combo left me speechless... 500€+ ITX mobo, 500€+ PSU... M2 on air should be paired with mATX for better VRM cooling, more storage, more fans header ... and reduced cost.

Sorry for the lack of precision.
I was very upset by Asus, ultra expensive, board that is able to make gigantic fan noise at its startup.

Thus having a fan on mini ITX board to cool VRM for the most power hungry CPU (like R9) is a strong downside I found, especially versus ATX or Micro ATX that do not have any fan (at least on those generation..)

 

[hrh_ginsterbusch]

Sorry for the lack of precision.
I was very upset by Asus, ultra expensive, board that is able to make gigantic fan noise at its startup.

Thus having a fan on mini ITX board to cool VRM for the most power hungry CPU (like R9) is a strong downside I found, especially versus ATX or Micro ATX that do not have any fan (at least on those generation..)

Well, its Asus. They seem to be premium, but have failed repeatedly, at least in my POV, when it comes to mainboards. Been with MSI and Asrock for ages. Had one Asus board 15 years ago, because "they make solid stuff", and the failure rate with this board was horrible. Next on, "cheap" Asrock board, its still around, being used as a placeholder for anything mATX, and technically still working, too. Then there was a lot of notebooks, and fast forward to 2021, its MSI and Asrock all over again

This very system I'm writing this reply on, is running on an MSI Z690 mATX board. Solid, fun, no issues so far. Gonna be moved from the Asus AP201 to the Meshroom S / Mesh-mATX, when I finally finish with my optimization work on the K77 Lite workstation build (hopefully in the next few weeks).

That particular build, the K77 Lite workstation, with a 7950X, is set up with the Asrock B650E PG/ITX, which is also very solid, despite the lack of USB4. Esp. no weird VRM fan noises - because its got no VRM fan to begin with

TL;DR: Asus might have been good in terms of mainboards at some point, but most of the time, you pay a lot of premium you dont really need in the end. The peripherals, like monitors, or eg. the before-mentioned AP 201 case, are really solid, and most of it doesnt feel like overpaying the premium, but the rest feels .. disconnected, to say the least. Plus their current support / RMA issues

cu, w0lf.

 

[MarcParis]

Well, its Asus. They seem to be premium, but have failed repeatedly, at least in my POV, when it comes to mainboards. Been with MSI and Asrock for ages. Had one Asus board 15 years ago, because "they make solid stuff", and the failure rate with this board was horrible. Next on, "cheap" Asrock board, its still around, being used as a placeholder for anything mATX, and technically still working, too. Then there was a lot of notebooks, and fast forward to 2021, its MSI and Asrock all over again

This very system I'm writing this reply on, is running on an MSI Z690 mATX board. Solid, fun, no issues so far. Gonna be moved from the Asus AP201 to the Meshroom S / Mesh-mATX, when I finally finish with my optimization work on the K77 Lite workstation build (hopefully in the next few weeks).

That particular build, the K77 Lite workstation, with a 7950X, is set up with the Asrock B650E PG/ITX, which is also very solid, despite the lack of USB4. Esp. no weird VRM fan noises - because its got no VRM fan to begin with

TL;DR: Asus might have been good in terms of mainboards at some point, but most of the time, you pay a lot of premium you dont really need in the end. The peripherals, like monitors, or eg. the before-mentioned AP 201 case, are really solid, and most of it doesnt feel like overpaying the premium, but the rest feels .. disconnected, to say the least. Plus their current support / RMA issues

cu, w0lf.

Well I've mostly got Asus & Asrock MB since the last 20 years. I got one Gigabyte MB (OMG its bios...)
Here is my point of view.
In terms of reliability, Asrock was mediocre till 2021 (approx) where they reached the bottom...but nowadays they made very strong effort and they are back.
In terms of functionality, Asus tends to be ahead (main example are Bios and combine PWM/DC fans header), let's say during last decade.

Over the last MB generations, Asus is not so appealling, because they lost their main advantage (I would say Bios, hands down) as competitors improved a lot on this field. On top ratio functionality/price ratio is bad for Asus MB.
To be specific, the MSI X870 Tomahawk is the best AM5 MB, so far, in terms of functionality/price ratio. to switch to more expensive MB, you need to have a ultra specific needs (I see mainly extreme overclocking or mandatory needs of 10gb ethernet).

 

[GuilleAcoustic]

mandatory needs of 10gb ethernet

Easily fixed with mATX by adding a NIC card... and often cheaper

 

[MarcParis]

Easily fixed with mATX by adding a NIC card... and often cheaper

Yes totally agree !

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
Cerberus-X late 2024 : Cooling the AMD Ryzen 7 9800X3D (19th December 2024)
-------------------------------------------------------------------------------------------------------------------------------

Let’s start with the harsh facts : My Noctua C14S, paired with 2 Noctua NF-A25x14G2 was not able to cool the AMD Ryzen 7 9800X3D @stock settings during AVX512 workload (IBT2.54 with a maximum 154.34W used) without thermal throttling!!!

The AMD Ryzen 7 9800X3D is given for a base TDP of 120W but it can boost till 160W.

Based on my former experience I was expecting my Noctua C14S to be effective till 170W but based on a monolithic & central design processor, like my Intel Core i5 12600K overclocked @5.2Ghz. But facts are clear, Zen 5 (like Zen 4) is based on very small chiplets, significantly harder to cool efficiently!

On top of that, I noticed major power peaks on the R7 9800X3D during gaming sessions. “Dragon Age: the Veilguard” can consume between 80W-120W during standard gameplay, but sometimes this is exceeding this amount significantly (~140W-150W). Of course I’m excluding shader compilation as it is also a power hungry task. Those power peaks were leading to distracting & irritating fan noise…ie not acceptable from my point of view.

That is where my quest to cool & tame the AMD Ryzen 7 9800X3D started..

I would target two main objectives :

• Explore alternative CPU coolers
• Undervolt / underclock the AMD R7 9800X3D

Best CPU Cooler (Cerberus-X)
I thought of 2 main competitors (both 120mm Tower style) of my beloved Noctua C14S inside the Cerberus-X :

• Thermalright peerless assassin 120 mini
• Noctua NH-D12L

Finally I decided to keep only the Noctua NH-D12L as “Thermalright peerless assassin 120 mini” fan is pretty mediocre in terms of noise and difficult to replace (due to fan clip)

Here are common parts :

• Thermal paste : Arctic MX-6
• Noctua Offset mounting hardware : NM-AMB13 (C14S) & NM-AMB14 (D12L)

Let’s compare the 2 cooling setups :

• Noctua C14S :
  • Orientation : Vertical (Heatpipes are towards top of the case)
    
    
  • Fresh Air Intake : From the top of the case
    
    
  • Hot Air Exhaust : Towards the side of the case (with NM-AMB13, the Noctua C14S is perfectly aligned with the side panel aeration)
    
    
  • CPU Cooler Fans : 2xNoctua NF-A14x25G2 placed as exhaust (perfect to exhaust all the hot air coming from the GPU)
    
    
  • Other Fans : None
  • Final setup look :
    
    
• Noctua D12L :
  • Orientation : Horizontal (I tested Vertical orientation, but it was requiring too much speed from the 2xbottom intake fans to cross the GPU)
    
    
  • Fresh Air Intake : From the vented side of the case
  • Hot Air Exhaust : Towards rear & top of the case
  • CPU Cooler Fans : 1xNoctua NF-A12x25R
    
    
  • Other Fans : 1xNoctua NFA9-25 (Rear) + 1xNoctua NF12x25 (Top)
    
    
  • Final setup look :
    
    

So is there any clear winner?
Well...well...here is a comparison from IBT 2.54 Very High profile (Intel Burn Test), assuming, similar peak power consumption :

Short answer between the Noctua D12L and the Noctua C14S, inside Cerberus-X, on a Ryzen 7 9800X3D (~155W TDP) :

• Pure Cooling performance : very similar on either CPU & GPU (RTX 4080 FE was also very similar)
• Cooling/Noise ratio : Advantage for the Noctua D12L as 140mm tends to be too noisy for me above 1000 rpm

At least, I've my own opinion on the debate between the 2 CPU coolers within the Cerberus-X. It could be important as I read some rather too optimistic review on reddit stating some 15-20°C advantage for the D12L over the Noctua C14S.

Here are some gameplays video using both CPU cooler, at stock settings :

• Noctua C14S :
  
• Noctua D12L :
  

Undervolting/Underclocking the AMD R7 9800X3D

Basically this step represents most of the time I've spent tweaking my setup. However I will spare you the boring approach of trials/errors..

I will try to sum-up what I've experienced :

• The most limiting factor for undervolting : Gaming. Basically undervolting PBO Curve Optimizer -30 was working fine with IBT 2.54 & Cinebench, but drivers were crashing while gaming.
• Best undervolting setup was using PBO Curve Shaper (new feature introduced with Zen 5), instead of Curve Optimizer
• Best undervolting setup : PBO Curve Shaper :
  • Min Frequencies : -30
  • Low Frequencies : -30
  • Med Frequencies : -30 (Max Threads performance, like Cinebench)
  • High Frequencies : -10 (Gaming)
  • Max Frequencies : -10
• Despite PBO undervolting, I'll still experienced audible power peaks while gaming, thus I decided to test also underclocking
• I tested 2 ways to underclock :
  • MSI Bios : TDP setup with power range
    • 120W-160W (Stock)
    • 95W-128W (no gaming performance loss)
    • 65W-88W (Gaming performance loss up to 2%, mostly equal)
  • AMD PBO / Limit / Manual / PPT : Put Wattage in W (some other MB could be in mW). Tested (& approved) at 95W
  • I recommend PBO PPT Limit as I experienced regular stutters while gaming with MSI Bios TDP Setup, especially at 65W-88W TDP
• Overall undervolting & underclocking (TDP limit) is leading to massive efficiency "Performance/Watts" improvements (up to +52%) versus stock parameters
  
  
  
  
  

As I was pleased with the overall results of undervolting/unclocking (look at this Cinebench 2024 Multi score at 1303 with a 95W TDP vs 1360 at best with 128W/PBO-Curve Optimizer -15) I decided to test a tiny CPU cooler, suited for mini ITX builds, the ID-Cooling IS-47-XT (47mm tall, 92mm slim fan, 95W TDP in Theory)

BONUS : ID COOLING IS-47-XT vs AMD R7 9800X3D

Here is the almighty ID Cooling IS-47-XT :

• 2 orientations possible ! (like most of Noctua CPU coolers)
  
  
• Good contact pressure
  
  
• Feeling as if it is missing something!
  
  
• Full setup using Noctua NFA9-14 PWM fan, 2500rpm Max
  
  
• Cooling Performance tested with a TDP of 65W-88W (Pass) and 95W-128W (heavy thermal throttle)
  
  
• Gaming Performance (TDP 95W-128W) : Very loud and thermal throttle
  
• Gaming Performance (TDP 65W-88W) : Quieter than the RTX 4080 FE, no thermal throttle!
  
• Idle noise : While at 65W-88W, the R7 9800X3D was basically quiet (not audible) even I needed to setup idle CPU temperature at 65°C (thermal balance was stable)

This test was not totally innocent as I wanted to test my R7 9800X3D in smaller cases than my Cerberus-X (like my N-ATX v2). Overall I'm impressed by both the ID Cooling IS-47-XT and the AMD R7 9800X3D, even undervolted/underclocked!
The AMD R7 9800X3D is shining by its incredible power efficiency!

I hope you enjoyed this article as much as I did.

More to come...hopefully still this year!

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
Cerberus-X N-ATX v2 : Testing the AMD Ryzen 7 9800X3D inside an ATX MB 15l case (23rd December 2024)
-------------------------------------------------------------------------------------------------------------------------------

After the successful testing with limited power limit on the AMD Ryzen 7 9800X3D inside the Cerberus-X (using PBO, power limit, PPT) I decided to test the full setup inside my SFFTIME N-ATX v2.

Why considering the N-ATX case?

• Not currently used in my living room
• Due to its size, it is fitting nicely behind my 49” screen (Position while testing my FormD T1)
• N-ATX is compatible with a 78mm tall cpu cooler. My current thermalright AXP120-67 is compatible with 120W TDP (based on AMD R7 5800X) based on my personal experiment
• N-ATX has pretty good GPU temperature and easy cable management
• N-ATX is compatible with mini ITX MB (tested with my minisforum BD790i)
• I’ve an active thread about this case : https://smallformfactor.net/forum/t...se-sfftime-n-atx-v2-complete-build-log.17997/

Here are the components I picked up :

• CPU : AMD R7 9800X3D
• MB : MSI X870 Tomahawk
• CPU cooler : Thermalright AXP120-67 with Noctua NFA12x25 Chromax fan (intake)
• Thermal paste : Arctic MX-6
• GPU : Nvidia RTX 4080 FE
• RAM : Corsair DDR5 2x24Go Vengeance 6200MTS
• SSD #1 : WD SN850X 2To
• SSD #2 : Samsung 990 Pro 4To
• Rear fan : Noctua NFA12x15 Chromax as exhaust
• PSU : Corsair SF750 platinum + custom cable for GPU (but suited for FormD T1, not the N-ATX v2..)

What about the CPU cooling?
Well I tested using PPT (PBO power limit) of 95W and 120W with optimized curve shaper

At 95W, the performance gap between the Noctua D12L/Cerberus-X and the Thermalright AXP12-67/N-ATX v2 is nil. However in terms of thermals the Noctua D12L/Cerberus-X is slightly better ~2-6°C

At 120W, the Thermalright AXP12-67/N-ATX v2 is capable of almost match performance of a stock R7 9800X3D. The thermal gap (+5°C for the D12L with a +23W additional to cool) is not super impressive for the Noctua D12L, does it mean it is bad? I would not say that the small chiplet design is clearly limiting the heat dissipation.

What about mix/gaming cooling?
Following the good results for the 120W TDP, I decided to try this setup first.


Here is a test with power limited to 95W :

Overall, GPU & CPU thermals are very similar to the Noctua D12L/Cerberus-X. Nothing new compared to what I already found.
Regarding the TDP of 120W I experienced the same disturbing power peaks as in the Cerberus-X. Thus I’ll continue my experimentations with a TDP of 95W.

I’ll continue my experimentations with something pretty new…stay tuned!

 

[MarcParis]

-------------------------------------------------------------------------------------------------------------------------------
CERBERUS-X : Testing MSI RTX 5080 OC (18th March 2025)
-------------------------------------------------------------------------------------------------------------------------------

This RTX5000 launch was just a nightmare…unfortunately it is still pretty bad.

Getting my hands on one RTX 5080 was the worst challenge ever in my, pretty long life, of IT geek. I was beaten by scalpers, I was scammed by fake scalpers…but I finally got one from official retailers…I definitely caught RTX fever that leads to bad bets/decisions.

It was even harder as not so many cards are compatible with SFF, basically 4, including the ghost FE model.
I gave up trying to catch one FE. Dislike former releases, Nvidia really screwed this launch like a champion!

Here is a RTX 5080 selection I built :

RTX 5080 Cards	Dimensions	U-ITX (306 x138 x 60 mm)	Cerberus-X (330 x 152 mm)	N-ATX ATX (67 x 388 x 163 mm) Mini ITX (82 x 388 x 163 mm)
Asus RTX 5080 Prime	306 x 126 x 50 mm	Yes	Yes	Yes
Nvidia RTX 5080 FE	304 x 137 x 40 mm	Yes	Yes	Yes
MSI RTX 5080 Ventus OC	303 x 121 x 49 mm	Yes	Yes	Yes
MSI RTX 5080 Inspire	288 x 112 x 50 mm	Yes	Yes	Yes
Gigabyte RTX 5080 Wind Force OC	304 x 126 x 50 mm	Yes	Yes	Yes
MSI RTX 5080 Trio OC	338 x 140 x 50 mm	Yes with 40mm offset panel	No	Yes
MSI RTX 5080 Suprim SOC	359 x 150 x 76 mm	No	No	Yes, with mini ITX only and no back fan
MSI RTX 5080 Vanguard	357 x 151 x 66 mm	No	No	Yes
MSI RTX 5080 Shadow	303 x 121 x 49 mm	Yes	Yes	Yes
Gainward GeForce RTX™ 5080 Phantom GS	331.9 x 145.6 x 69.9 mm	No	No	Yes, Mini ITX only
Asus RTX 5080 TUF	348 x 146 x 72 mm	No	No	Yes, Mini ITX only
KFA2 GeForce RTX 5080 1-Click OC	315 x 140.5 x 50.7mm	Yes with 10mm offset panel	Yes, barely	Yes
Gainward GeForce RTX 5080 Phoenix	331.9 x 133.1 x 60 mm	Yes with 30mm offset panel	No	Yes
Zotac GeForce RTX 5080 Solid OC	329.7 x 137.8 x 67.8mm	No	No	Yes
PNY GeForce RTX 5080 Triple Fan OC	330 x 60 mm	Yes with 30mm offset panel	No	Yes
Inno3D GeForce RTX 5080 X3 OC	300 x 120 x 50 mm	Yes	Yes	Yes
Asus ROG RTX 5080 Astral	357.6 x 149.3 x 76 mm	No	No	Yes, only mini ITX
Palit RTX 5080 gaming pro	331.9 x 127.1 x 60 mm	Yes with 30mm offset panel	No	Yes
Palit RTX 5080 gamerock	331.9 x 150 x 70.4 mm	No	No	Yes, only mini ITX
RTX 5080 GAMING OC 16G	340 x 140 x 70 mm	No	No	Yes, only mini ITX

In the end, I got one MSI RTX 5080 Ventus OC:





In detail, this MSI RTX 5080 Ventus OC is the model with 4 big heatpipes (the “PLUS” version seems to have 6 smaller heatpipes, do not expect any performance gap between the 2 models). There are 3 fans of ~90mm, directly controllable by the software Fan Control (2 fan controllers, 1 for the i/o fan and 1 for the middle/last fan). Fan speed range is approx between 1000 rpm to 2200 rpm. This card is pretty light, with just 1.1kg in total weight.

Most important : My RTX 5080 has 112 ROPS…a full chip!

I must admit I was a bit reluctant as few reviews/feedback were not so gentle, and I was not sure who was wrong…at least, now I know..

To test & benchmark it I used my beloved, cherished, almighty Cerberus-X!

Here is the full setup :

• CPU : Intel Core i5 12600K, stock
• CPU Cooler : Noctua C14S + NF-A14 (exhaust)
• MB : Asus Prime Z690-P Wifi
• RAM : 4x8go DDR4 GSkill Flare-X 3200MTS CL14-14-14-34
• GPU : MSI RTX 5080 Ventus OC
• PSU : Corsair SF750 + Cablemod 12vhpwr (3xType 4)
• SSD #1 : ADATA SX8200 Pro 2To
• Bottom Fans : Pair of Noctua NFA12x25 (Intake) + Spacer NA-IS1-12 chromax.black Sx2

Strangely enough, colors (silver+black) of the Asus motherboard are matching 100% the ones on the MSI RTX 5080..

One of the best selling points of getting an MSI GPU is that it is fully compatible with MSI Afterbuner, which remains the best software to tweak a GPU.

Please note that the MSI RTX 5080 OC does not have the possibility to increase the power limit. It is 100% and that’s the full maximum! Anyway I was never a fan of increasing the power limit.
Speaking of overclocking, I was able to apply a +300Mhz on the core (exceeding the 3000Mhz during load) and +1000Mhz on the GDDR7. These are still early numbers and I will continue playing with them later on.
At last, I did not apply any undervolting as it is impacting significantly the performance of the RT cores. I prefer to use DLSS4 mode to limit the power consumption.

Here are the benchmarks, comparing with my RTX 4080 FE and even my RTX 3090 FE!

Thanks to the “free” overclocking the RTX 5080 is a step up versus my RTX 4080 FE, especially with ray tracing.

But what about the thermals & noise of this MSI RTX 5080 Ventus OC card?

Let’s be honest, fans on the MSI RTX 5080 Ventus are miles away from Noctua’s ones, but they are at least decent. They have some parasite noise but overall they are OK. It is useful to say that “deshroudding” this card is easy & simple and is not removing the warranty. However I might wait for a retail 3D printed kit to test it properly.
Zero fan mode at idle is working fine too.

There is some coil whine but at a very acceptable level & occasional.

At last regarding thermals, within the Cerberus-X, with this very optimized airflow (tons of fresh intake air, quickly exhausted once heated up), the MSI RTX 5080 is running like a charm/champ! If I leave the original fan curve, the RTX 5080 is not exceeding 70°C under ~300W load.
Thus I decided to use a custom fan curve, limiting the fan speed to 50%, ie 1500rpm. With this setup the RTX 5080 is reaching 72-73°C under 300W load.

Here is a gameplay session in “The last of Us”


Time to conclude on this MSI RTX 5080 Ventus OC!

Let’s be simple, in an optimized airflow, this GPU is working very well. I’ve very little to complain about.
However, unlike the RTX 4080 FE, the heated air of the MSI RTX 5080 Ventus is not oriented. So within a case with little to no airflow, thermals & noise should be hurt significantly.

Stay tuned as I’m already testing it in my U-ITX.