Production The First - Monsterlabo High Performance SFF case & Fanless CPU / GPU

[yeahe - ML]

Wow, great thing! I tried and downvolted 1080 to 900mV.
Firestrike ultra stress test:

• at default settings: 55C and ~1750 frequency at the end, 28.11 avg fps
• undervolted: 52C and ~1830 freq at the end, 28.41 avg fps

Spoiler

You may had taken the power consumption of the GPU ?

 

[yeahe - ML]

Good news I have seen for CPU. The New Ryzen 9 3900 with a TDP of 65W for 12 cores / 24 threads !!

 

[Nord1ing]

You may had taken the power consumption of the GPU ?

No, i will check it then i'll try to optimize it more.

Good news I have seen for CPU. The New Ryzen 9 3900 with a TDP of 65W for 12 cores / 24 threads !!

Since AMD boost to around same clock for x and non-x models, isn't this TDP too low compared to max power consumption?...
at 28:14:

 

[yeahe - ML]

I need to get more into that !!
My first testing session this morning on Firestrike extreme / 22C ambiant / 1080FTW

1/ Standard configuration
Furmark 9570 points
Max GPU power is 199W -Average 138W
Average voltage 0.91V
Average freq 1074Mhz

2/ MSI afterburner with clock at 1700Mhz (below standard curves)
Furmark 9130 points
Max GPU power is 168W -Average 111W
Average voltage 0.86V
Average freq 1074Mhz

--> I need to check why in standard mode frequency is lower than when I try to downclock ...
--> I will go even below & complete with CPU.
--> I tried to use EVGA software. I was just able to decrease performance at same power

 

[yeahe - ML]

No, i will check it then i'll try to optimize it more.


Since AMD boost to around same clock for x and non-x models, isn't this TDP too low compared to max power consumption?...
at 28:14:

I hope we will keep the same range of TDP increase. Maybe a 40 more would keep us below 100W. That would be really nice for 24 threads.
The TDP is a difficult number to play with and may be only realistic at stock clock.

 

[Eduard]

You may had taken the power consumption of the GPU ?

I will do and post soon before/after undervolting results.What I remember from my previous tests is that my 1080Ti reference, was boosting itself to 1870MHz, but it was fluctuating a lot. Consumption was 240-250 Watts. At 1.1 V - also fluctuating. Don't remember the temps, but the difference was huge. After setting up the curve on 1810 MHZ with 0.9V the clock became stable and max consumption was reduced to 190 Watts in Heaven. And I lost like 30 points in performance (like 3550 => 3520). So the gain is huge.

Here I set up the curve much lower than the standard boost, in order to limit more the consumption as 250W is too much for the case.

 

[yeahe - ML]

I will do and post soon before/after undervolting results.What I remember from my previous tests is that my 1080Ti reference, was boosting itself to 1870MHz, but it was fluctuating a lot. Consumption was 240-250 Watts. At 1.1 V - also fluctuating. Don't remember the temps, but the difference was huge. After setting up the curve on 1810 MHZ with 0.9V the clock became stable and max consumption was reduced to 190 Watts in Heaven. And I lost like 30 points in performance (like 3550 => 3520). So the gain is huge.

Here I set up the curve much lower than the standard boost, in order to limit more the consumption as 250W is too much for the case.

Ok understand. I will make some test.
But what I'm wondering is why we can't play directly on the voltage like for CPU. He we have a curves that goes like to the limit for voltage.
I will check if it keeps the minimum voltage only.


1/ Standard configuration
Furmark 9570 points
Max GPU power is 199W - Average 138W
Average voltage 0.91V
Average freq 1074Mhz

2/ MSI afterburner with clock at 1700Mhz (below standard curves)
Furmark 9130 points
Max GPU power is 168W -Average 111W
Average voltage 0.86V
Average freq 1074Mhz

3/ MSI afterburner clock at 1600Mhz
Furmark 8840 points
GPU - 14°C compare to standard
Max power 155W

4/ At 1200Mhz
Furmark 7250 points
Max GPU 91W - Max voltage 0.69C

Still to go deeper into this test

 

[Nord1ing]

So my results are next for a firestrike ultra stress tests run:

1. Baseline (power settings at defaults):

   • 	max	avg
     Voltage, mV	1000​	941.96​
     Temp, C	56​	52.36​
     Power, W	196.197​	175.48​
     Clock, MHz	1847.5​	1773.60​
     FPS		28.36​

2. Down-volting+OC to 900mV at 1914 MHz

   • 	max	avg
     Voltage, mV	900​	897​
     Temp, C	53.0​	48.9​
     Power, W	178.5​	163.4​
     Clock, MHz	1923.5​	1909.2​
     FPS		28.79​

3. Down-volting+OC to 900mV at 1949 MHz

   • 	max	avg
     Voltage, mV	900​	900​
     Temp, C	53.0​	49.5​
     Power, W	178.0​	164.6​
     Clock, MHz	1949.0​	1939.7​
     FPS		28.96​

4. Down-volting+OC to 880mV at 1911 MHz

   • 	max	avg
     Voltage, mV	894​	883​
     Temp, C	51.0​	48.1​
     Power, W	169.9​	156.2​
     Clock, MHz	1911.0​	1899.8​
     FPS		28.76​

 

[yeahe - ML]

So my results are next for a firestrike ultra stress tests run:

1. Baseline (power settings at defaults):

   • 	max	avg
     Voltage, mV	1000​	941.96​
     Temp, C	56​	52.36​
     Power, W	196.197​	175.48​
     Clock, MHz	1847.5​	1773.60​
     FPS		28.36​

2. Down-volting+OC to 900mV at 1914 MHz

   • 	max	avg
     Voltage, mV	900​	897​
     Temp, C	53.0​	48.9​
     Power, W	178.5​	163.4​
     Clock, MHz	1923.5​	1909.2​
     FPS		28.79​

3. Down-volting+OC to 900mV at 1949 MHz

   • 	max	avg
     Voltage, mV	900​	900​
     Temp, C	53.0​	49.5​
     Power, W	178.0​	164.6​
     Clock, MHz	1949.0​	1939.7​
     FPS		28.96​

4. Down-volting+OC to 880mV at 1911 MHz

   • 	max	avg
     Voltage, mV	894​	883​
     Temp, C	51.0​	48.1​
     Power, W	169.9​	156.2​
     Clock, MHz	1911.0​	1899.8​
     FPS		28.76​

Ok same behaviour. But that's nice to see how you win some watts and degrees by keeping performance.
I just don't understand as when I'm doing myself why at stock the clock is not higher than when we down clock.

 

[Nord1ing]

Ok same behaviour. But that's nice to see how you win some watts and degrees by keeping performance.
I just don't understand as when I'm doing myself why at stock the clock is not higher than when we down clock.

As I recall correctly nvidia's boost take into consideration thermal and power limits. Default curve for the same frequency go well beyond official tdp of 180W.

 

[Eduard]

Ok same behaviour. But that's nice to see how you win some watts and degrees by keeping performance.
I just don't understand as when I'm doing myself why at stock the clock is not higher than when we down clock.

As you probably know, there is a silicon lottery. 2 same GPU can perform very differently. So in order to make all GPU work out of the box the manufacturer sets the standard curve like for lower performer GPU. Otherwise, some worse GPU will crush with a standard curve.
So depending on your luck, there are big chances that you GC can perform better if you tweak the curve (lowering voltage creates bigger temp headroom).

So our goal here is to find the real needed voltage for a given clock (most times it is lower than standard curve sets). For your tests, don't lower the clock, but find the lowest possible voltage that your GC can be stable maintaining 1774MHZ for example. In this way, you will not lose any performance, but will gain in consumption and thermals.
If you are lucky like Nordl1ng you can even overclock the GPU and lower the voltage in same time =)

 

[microclue]

Hi! I purchased the second-hand monsterlabo unit, and nicolas asked me to share my experience building it. Before I get into that, I'd like to say that their packaging was extremely sturdy and personal, and no major flaws with the second-hand unit.

These are my specs:

Gigabyte Z390i
Intel I7-9700k
32GB ram
EVGA GTX 980 ftw acx 2.0 (reused from my previous build)
Corsair SF450 + additional atx cable
140mm fan (random one on clearance at my local store, NOT from noctua)

It was quite difficult to build starting from scratch - the videos were very informative but it was lacking some gotchas (like changing the back panel to the 3D printed one before starting the build). And initially, I was getting high thermal throttling because the gpu and cpu block weren't aligned correctly (the only "minor" flaw). However, nicholas and elisa were very helpful and responsive to my issues, and we were eventually able to get it working!

I do not receive any throttling at stock speeds, with my gpu and cpu hovering around 60-65 C under load. Unfortunately, it doesn't seem to perform well under some minor gpu overclocks, probably due to the VRMs overheating, so I'll need to purchase the vrm heatsink once it's available. (I have not tried cpu overclocking but it does full turbo just fine.)

Attaching an image of my setup! At this link... https://photos.app.******/U6HvkfDmxvnk43qv7

Thank you, this is a well thought out case design!

 

[yeahe - ML]

@microclue Many thanks !!

 

[Eduard]

Hi! I purchased the second-hand monsterlabo unit, and nicolas asked me to share my experience building it. Before I get into that, I'd like to say that their packaging was extremely sturdy and personal, and no major flaws with the second-hand unit.

These are my specs:

Gigabyte Z390i
Intel I7-9700k
32GB ram
EVGA GTX 980 ftw acx 2.0 (reused from my previous build)
Corsair SF450 + additional atx cable
140mm fan (random one on clearance at my local store, NOT from noctua)

It was quite difficult to build starting from scratch - the videos were very informative but it was lacking some gotchas (like changing the back panel to the 3D printed one before starting the build). And initially, I was getting high thermal throttling because the gpu and cpu block weren't aligned correctly (the only "minor" flaw). However, nicholas and elisa were very helpful and responsive to my issues, and we were eventually able to get it working!

I do not receive any throttling at stock speeds, with my gpu and cpu hovering around 60-65 C under load. Unfortunately, it doesn't seem to perform well under some minor gpu overclocks, probably due to the VRMs overheating, so I'll need to purchase the vrm heatsink once it's available. (I have not tried cpu overclocking but it does full turbo just fine.)

Attaching an image of my setup! At this link... https://photos.app.******/U6HvkfDmxvnk43qv7

Thank you, this is a well thought out case design!

Hi, thank you for shearing.
Did you tried to stress test your CPU? I also have 9700k, but it is overheating under prime95 for exemple. I had to limit the TDP in order to keep the temps in check.

By the way, thanks to Nicholas et Elisa! They helped me a lot also!

 

[microclue]

Hi, thank you for shearing.
Did you tried to stress test your CPU? I also have 9700k, but it is overheating under prime95 for exemple. I had to limit the TDP in order to keep the temps in check.

By the way, thanks to Nicholas et Elisa! They helped me a lot also!

Yes, I did stress test the cpu! For small FFT prime95, the CPU does throttle. It does not in the balanced stress test. If you pay attention to the wattage usages though, it begins to throttle >200W which is way beyond the specifications of the case. Small FFTs are a weird test since it uses the AVX2/AVX512 extensions heavily. It's not a typical use case unless you're during scientific computing imo, so I'm not worried about that.

If it throttles in the balanced test, then you might have had the same problems as me though...

 

[yeahe - ML]

Hello all,

If you have few seconds for us you can vote for us here:

Un PC très puissant, totalement silencieux et customisable | Innovatech

Les PC sur le marché sont de plus en plus performants et de plus en plus énergivores. Pour éviter la surchauffe de certains composants, les ordinateurs doivent aussi être équipés d'un dissipateur de chaleur. Pourtant, le refroidissement est souvent relégué au second plan et n'intervient qu'en...

www.innovationswallonnes.be

Sorry that's in french, but it's in the RED BOX

We are participating to a conference on which they will reward the most innovative company in the local area.

We may be not the most innovative but we have a lot of nice people that are following us Many thanks

 

[Eduard]

Yes, I did stress test the cpu! For small FFT prime95, the CPU does throttle. It does not in the balanced stress test. If you pay attention to the wattage usages though, it begins to throttle >200W which is way beyond the specifications of the case. Small FFTs are a weird test since it uses the AVX2/AVX512 extensions heavily. It's not a typical use case unless you're during scientific computing imo, so I'm not worried about that.

If it throttles in the balanced test, then you might have had the same problems as me though...

Wow, this is not at all the same results that I had. My i7-9700k reaching 100C and starts to throttle after 1-2 mins at 150W (not even necessary to turn on AVX). So I had to limit the consumption at 120W (in bios) max and in this way it reaches 70-75C max with 20cm fan @ 500RPM.I reassembled the system like 4-5 times getting the same results, so not a contact issue... It is frustrating...

 

[yeahe - ML]

@Eduard & @microclue Yes there is something wrong !!! We need curves to decide

120W - 70C @500RPM seems for me to be in the good range when system is closed.

That would be nice to compare to what @microclue has in same stress test at same power consumption.
Normally you should'nt be able to go over 160/170W at CPU level but as you will be close to CPU temp. limit, little improvement can have quite a big impact.

@Eduard, in some case the chip IHS is really bad (concave), and you can easily loose +10°C/15°C just into the grease.

In the same way, have you seen that the cooling of the i9 9900KF is better than 9900K ? Just because of spreading inside is better as there is more space due to unactive iGPU

 

[Eduard]

1.@Eduard, in some case the chip IHS is really bad (concave), and you can easily loose +10°C/15°C just into the grease.
2.In the same way, have you seen that the cooling of the i9 9900KF is better than 9900K ? Just because of spreading inside is better as there is more space due to unactive iGPU

1. Hm... Gamers Nexus made several tests grinding IHS, but never got better results than -2C. Do you have any links to tests with the difference of 10-15C?
2.How there can be a difference in cooling if there is no stractural difference (if descreet GPU used)? F version has a desactivated iGPU and normal version has activated, but not ACTIVE iGPU. This should result in exactly same results.

 

[Nord1ing]

Wow, this is not at all the same results that I had. My i7-9700k reaching 100C and starts to throttle after 1-2 mins at 150W (not even necessary to turn on AVX). So I had to limit the consumption at 120W (in bios) max and in this way it reaches 70-75C max with 20cm fan @ 500RPM.I reassembled the system like 4-5 times getting the same results, so not a contact issue... It is frustrating...

Interesting...
I am waiting to upgrade my rig to 8C ryzen 3700x. I will post my thermal results.
P.S.:
Temps in WoT EnCore bench

Spoiler: YouTube