Thread for LTT Labs Test Suggestions

[Thorarin]

I'd love an update on the HDMI cable test, adding more current offerings.

 

The list is over two years old now and doesn't contain anything by your friends UGREEN for example. They are fairly prominent in Europe, while Infinite Cables for example is just not a thing around here. It would be nice to see if anything has improved in the area of cables actually meeting specs.

[m9x3mos]

The framework 16 is going out to reviewers soon and hitting mass production using liquid metal pad. A review on usage of liquid metal pad in laptop would be amazing. 

[Colin1374]

Sorry if this is the wrong thread to post this in, but a labs review of the Astro A50X headphones would be dope. It may not serve well as a full LTT video though, but maybe a short circuit video backed by labs data. It seems to have a unique setup with three HDMI ports on the base, and an easy input switcher on the headphones themselves at $399, so it would be interesting to know what metrics and quality you get with that price. 

[__Dreamer__]

Itek Monitor GGF - 28
Went on sale quite a lot of times and seams to be the cheaper 28 monitor 60hz that is good enough

is supposed to be 10bits but no one tested it nor verified the picture quality
despite that as soon as went on sale from 300 to 150 is was immediately out of stock so a lot of people bought it

would be really cool to see a review of that

[ToboRobot]

It sounds like we are coming full circle and review content delivery may not be video first forever.

Monetizing that data/content of such a platform has always been the challenge and why video became dominate.    

The Labs content could be monetized by paywalling certain features like a fully interactive website for paid users vs statically generated data and charts.  Or it could be time based, where paid users get the data the instant it was published, and the free tier gets access to regular weekly or monthly data dumps.

Either way, will labs be a benefit included with peoples floatplane subscriptions, will it be included in a new Floatplane Premium Tier or will it be a separate thing? 

[greentumaru]

How about 510 vape batteries. I just had one explode on me. I called the store to tell them they were selling dangerous goods but they just offered me half off the next battery I purchased. I declined and got something hopefully safer.

[JudgeFudge821]

I could not find another topic on this, so I'd like to see a beginner's guide to RAM and specifically how the different frequencies actually impact the gaming experience.  Similar to graphics card reviews, is there value in RAM kit reviews/tables.  Can we compare like setups where RAM is swapped out for various tests?

[NoBrainer]

Hey I think this would be the perfect lab test for Linus. 
 

what is the actually needed radiator size to cool x amount of heat generated in watts?

 

This should be a relatively straightforward mathematical calculation that we can test for with known variables:

1. static pressure and airflow of the air going through the radiator. 
2. ambient temperature of 22c and In house summer temperatures
3. radiator thickness with same volume. Will a slim (1.18 liter volume) 420x140x20mm radiator with 19FPI(fins per inch) to a (1.18liter volume) 140x60mm radiator with 10FPI
4. fin density
5. heat plate to wattage/cm2 so we know the cold plate actually can transfer the heat. 
6. liquid flow rate through the loop keeping it as simple as possible to prevent bottlenecks 
7. 100% copper vs 100% no copper system
8. Why would we ever want to have a low delta T temperature? Aren’t we interested in what is the maximum amount of heat the radiator can transfer from the chip>water> air. Meaning the higher the water temperature is compared to the ambient the more efficient the heat transfer is. 

we always talk about “rule of thumb” but what actuall basis does this actually have? Isn’t it completely made up out of the eather 20 years ago that some random dude said and just became dogmatic truth with zero evidence for it. 
 

A common rule of thumb that has been passed around the online forums is that you should dedicate 120mm (or 140mm) of radiator space per component that is to be cooled, with an additional 120mm (or 140mm) of radiator space if you are going to be overclocking.  Or So a rule of thumb would be 180 watts per 120mm rad using 72cfm fans. that I can find as early as 2007. 
 

the radiators below are all copper. 

AcNex= Alphacool NexXxoS full copper radiators

XSPC TX=XS-PC TX Ultrathin Radiator

HWlabs GTS=Hardware Labs Black Ice Nemesis GTS
 

Spoiler Radiator Volym(liters)  Dimension(mm)  FPI(Fins per inch) AcNeX HPE-20 140mm 0.48 liters 20*140*171  19 AcNeX HPE-20 240mm 0.65 liters 20*120*270 19 AcNeX HPE-20 280mm 0.87 liters 20*140*311 19 AcNeX HPE-20 360mm 0.94 liters 20*120*390 19 AcNeX HPE-30 420mm 1.97 liters 30*144*455.5 18 AcNeX HPE-30 140mm 0.75 liters 30*144*173.5  18 AcNeX HPE-30 240mm 1.01 liters 30*124*271.5 18 AcNeX HPE-30 280mm 1.36 liters 30*144*314.5 18 AcNeX HPE-30 360mm    AcNeX ST-30v.2 140mm AcNeX ST-30v.2 240mm AcNeX ST-30v.2 280mm AcNeX ST-30v.2 360mm AcNeX ST-25 120mm AcNeX ST-25 240mm AcNeX ST-25 360mm XSPC TX 120mm XSPC TX 240mm XSPC TX 360mm HWlabs GTS 120mm HWlabs GTS 140mm HWlabs GTS 240mm HWlabs GTS 280mm HWlabs GTS 360mm HWlabs GTS 180mm   EKWB Q S120mm EKWB Q S140mm EKWB Q S240mm EKWB Q S280mm EKWB Q S360mm 1.46 liters   0.76 liters 1.02 liters 1.37 liters 1.47 liters 0.46 liters 0.73 liters 1.06 liters    0.40 liters 0.71 liters 1.02 liters 0.62 liters  0.78 liters 1.09 liters 0.96 liters 1.80 liters 1.16 liters 0.64 liters 0.80 liters 1.11 liters 1.41 liters 1.58 liters 30*124*391.5    30*143*177 30*124*275 30*144*316 30*124*394,5 25.5*120*151 22.5*120*272 22.5*120*272   20.5*125*171 20.5*125*278 20.5*125*398 29.6*133*158 29.6*153*172 29.6*133*278 29.6*153*312 29.6*133*398 29.6*180*217   30*130*165 30*145*185 30*130*185 30*145*325 30*130*405 18   15 15 15 15 15 15 15   22 22 22 16 16 16 16 16 16   20 20 20 20 20

Some different fans as example.

Spoiler

Fans 120mm-200mm   120x120x25  NF-F12 industrialPPC-3000 PWM          NF-F12 industrialPPC PWM   Volume( liters)   0.36liters   700-3000 ROM 109.89 cfm 7.63 mmH2O 43.3dB 450-2000 RPM 71.69CFM 3.94 mmH2O 29.7dB   140x140x25 NF-A14 industrialPPC-2000 PWM          NF-A14  industrialPPC-3000 PWM         Thermaltake Toughfan Pro 14  0.49 liter  500-2000 rpm 107.4 cfm 4.18 mmH2O 31dB   800-3000 rpm 159 cfm 6.58 mmH2O 41dB     500-2000 rpm 119.6cfm 3.57 mmH2O 31dB   180x180x38         Dynamic X2 GP-18         Prisma AL-18 PWM   1.23 liter 1 200 rpm 153.7 cfm 2.26 mmH2O     300-1 200 rpm 44.9 – 153.7 CFM 0.4 – 2.26 mmH2O 16 – 35.4 dB 300-1200RPM 43.1 – 146.8 CFM 0.44-2.5 mmH2O 14.9-37.4dB   200x200x30  NF A-20 PWM 1.2 liter 350-550-800 rpm 59.33-86.5 cfm 0.51-1.08 mmH2O 10.7-18dB

 

people say it’s not an exact science, but how can that be the case? We know exactly how big the radiator for a car needs to be to maximize the heat it can handle without the engine overheating, we know exactly how big radiators for air conditioning need to be to move x amount of heat. 

just take this video of Linus cooking an RTX 3090 doing apparently completely fine with 120mm radiator with 350 W TDP

So with the info from EKWB if the Power Dissipated for the let’s say the HWlabs GTS 140mm radiator with 10°C Coolant Delta (W/10C ΔT)is about 130W/10C ΔT at 1300RPM push only fan. 
 

This would mean if the ambient temperature is 22c and the water loop is 10c warmer at 32c that

would that mean at double the delta T temperature Capacity W/20K

 

Power Dissipated for 10°C Coolant Delta (W/10K ΔT)

If we double the water temps to (W/20C ΔT) the cooling capacity would be about 260W/20C ΔT, meaning the temperature is above 40°C Coolant Delta (13 W/1C ΔT) and the water temperature is above 60c and everything should be well under the thermal limit as we saw in the video( unless the pump/plastics can’t handle 60c+). 
 

practically entailing that everyone have an extremely oversized radiator compared to the maximum headload they need to handle, and might just sabotage their loops by having multiple radiator in series that makes it less efficient and obstructing the water. 

and we can see that Hardware Labs uses a delta of 25c, about 1.8GPM and 65CPM per fan. 

Or this apparent RTX4090 and 7950x3D using AlphaCool 140mm v2 radiator to cool everything 

https://smallformfactor.net/forum/threads/complete-s4max23-worlds-smallest-4090-build-brickless-5l-s4mini-4090fe-7950x3d-800w-water-cooled.18499/

 

that was also replicated earlier with the 3090

https://smallformfactor.net/forum/threads/s4max-brickless-s4m-w-3090-fe-and-r9-5950x-800w-5l-water-cooled.11277/

 

we can read at EKWB description of radiator performance. 
https://www.ekwb.com/blog/radiators-part-2-performance/

 

and looking at the xtremerigs evaluation almost 10 years ago with 1850 RPM fans( don’t know the CFM/static pressure they get close to 500w cooling capacity with 360mm radiators

 

and we can also look at the 280mm radiator performance and the 140mm radiators

performance down below 

Spoiler

 

 

https://www.xtremerigs.net/2015/02/11/radiator-round-2015/5/

 

And at the 280mm radiators showing their performance as well

https://www.xtremerigs.net/2016/02/15/hardware-labs-nemesis-280gts-radiator-review/5/

https://www.xtremerigs.net/2015/11/19/ek-coolstream-ce-140-radiator-review/5/

 

 

[NoBrainer]

4 hours ago, NoBrainer said:

Hey I think this would be the perfect lab test for Linus. 
 

what is the actually needed radiator size to cool x amount of heat generated in watts?

 

This should be a relatively straightforward mathematical calculation that we can test for with known variables:

1. static pressure and airflow of the air going through the radiator. 
2. ambient temperature of 22c and In house summer temperatures
3. radiator thickness with same volume. Will a slim (1.18 liter volume) 420x140x20mm radiator with 19FPI(fins per inch) to a (1.18liter volume) 140x60mm radiator with 10FPI
4. fin density
5. heat plate to wattage/cm2 so we know the cold plate actually can transfer the heat. 
6. liquid flow rate through the loop keeping it as simple as possible to prevent bottlenecks 
7. 100% copper vs 100% no copper system
8. Why would we ever want to have a low delta T temperature? Aren’t we interested in what is the maximum amount of heat the radiator can transfer from the chip>water> air. Meaning the higher the water temperature is compared to the ambient the more efficient the heat transfer is. 

we always talk about “rule of thumb” but what actuall basis does this actually have? Isn’t it completely made up out of the eather 20 years ago that some random dude said and just became dogmatic truth with zero evidence for it. 
 

A common rule of thumb that has been passed around the online forums is that you should dedicate 120mm (or 140mm) of radiator space per component that is to be cooled, with an additional 120mm (or 140mm) of radiator space if you are going to be overclocking.  Or So a rule of thumb would be 180 watts per 120mm rad using 72cfm fans. that I can find as early as 2007. 
 

the radiators below are all copper. 

AcNex= Alphacool NexXxoS full copper radiators

XSPC TX=XS-PC TX Ultrathin Radiator

HWlabs GTS=Hardware Labs Black Ice Nemesis GTS
 

Reveal hidden contents   Radiator Volym(liters)  Dimension(mm)  FPI(Fins per inch) AcNeX HPE-20 140mm 0.48 liters   20*140*171  19 AcNeX HPE-20 240mm 0.65 liters 20*120*270 19 AcNeX HPE-20 280mm 0.87 liters 20*140*311 19 AcNeX HPE-20 360mm 0.94 liters 20*120*390 19 AcNeX HPE-30 420mm 1.97 liters 30*144*455.5 18 AcNeX HPE-30 140mm 0.75 liters 30*144*173.5  18 AcNeX HPE-30 240mm 1.01 liters 30*124*271.5 18 AcNeX HPE-30 280mm 1.36 liters 30*144*314.5 18 AcNeX HPE-30 360mm    AcNeX ST-30v.2 140mm AcNeX ST-30v.2 240mm AcNeX ST-30v.2 280mm AcNeX ST-30v.2 360mm AcNeX ST-25 120mm AcNeX ST-25 240mm AcNeX ST-25 360mm XSPC TX 120mm XSPC TX 240mm XSPC TX 360mm HWlabs GTS 120mm HWlabs GTS 140mm HWlabs GTS 240mm HWlabs GTS 280mm HWlabs GTS 360mm HWlabs GTS 180mm   EKWB Q S120mm EKWB Q S140mm EKWB Q S240mm EKWB Q S280mm EKWB Q S360mm 1.46 liters   0.76 liters 1.02 liters 1.37 liters 1.47 liters 0.46 liters 0.73 liters 1.06 liters    0.40 liters 0.71 liters 1.02 liters 0.62 liters  0.78 liters 1.09 liters 0.96 liters 1.80 liters 1.16 liters 0.64 liters 0.80 liters 1.11 liters 1.41 liters 1.58 liters 30*124*391.5    30*143*177 30*124*275 30*144*316 30*124*394,5 25.5*120*151 22.5*120*272 22.5*120*272   20.5*125*171 20.5*125*278 20.5*125*398 29.6*133*158 29.6*153*172 29.6*133*278 29.6*153*312 29.6*133*398 29.6*180*217   30*130*165 30*145*185 30*130*185 30*145*325 30*130*405 18   15 15 15 15 15 15 15   22 22 22 16 16 16 16 16 16   20 20 20 20 20

Some different fans as example.

  Reveal hidden contents

 

Fans 120mm-200mm   120x120x25  NF-F12 industrialPPC-3000 PWM            NF-F12 industrialPPC PWM   Volume( liters)   0.36liters   700-3000 ROM 109.89 cfm 7.63 mmH2O 43.3dB 450-2000 RPM 71.69CFM 3.94 mmH2O 29.7dB   140x140x25 NF-A14 industrialPPC-2000 PWM          NF-A14  industrialPPC-3000 PWM         Thermaltake Toughfan Pro 14  0.49 liter  500-2000 rpm 107.4 cfm 4.18 mmH2O 31dB   800-3000 rpm 159 cfm 6.58 mmH2O 41dB     500-2000 rpm 119.6cfm 3.57 mmH2O 31dB   180x180x38         Dynamic X2 GP-18         Prisma AL-18 PWM   1.23 liter 1 200 rpm 153.7 cfm 2.26 mmH2O     300-1 200 rpm   44.9 – 153.7 CFM   0.4 – 2.26 mmH2O   16 – 35.4 dB 300-1200RPM 43.1 – 146.8 CFM 0.44-2.5 mmH2O 14.9-37.4dB   200x200x30  NF A-20 PWM 1.2 liter 350-550-800 rpm 59.33-86.5 cfm 0.51-1.08 mmH2O 10.7-18dB

 

 

 

people say it’s not an exact science, but how can that be the case? We know exactly how big the radiator for a car needs to be to maximize the heat it can handle without the engine overheating, we know exactly how big radiators for air conditioning need to be to move x amount of heat. 

just take this video of Linus cooking an RTX 3090 doing apparently completely fine with 120mm radiator with 350 W TDP

So with the info from EKWB if the Power Dissipated for the let’s say the HWlabs GTS 140mm radiator with 10°C Coolant Delta (W/10C ΔT)is about 130W/10C ΔT at 1300RPM push only fan. 
 

This would mean if the ambient temperature is 22c and the water loop is 10c warmer at 32c that

would that mean at double the delta T temperature Capacity W/20K

 

Power Dissipated for 10°C Coolant Delta (W/10K ΔT)

If we double the water temps to (W/20C ΔT) the cooling capacity would be about 260W/20C ΔT, meaning the temperature is above 40°C Coolant Delta (13 W/1C ΔT) and the water temperature is above 60c and everything should be well under the thermal limit as we saw in the video( unless the pump/plastics can’t handle 60c+). 
 

practically entailing that everyone have an extremely oversized radiator compared to the maximum headload they need to handle, and might just sabotage their loops by having multiple radiator in series that makes it less efficient and obstructing the water. 

and we can see that Hardware Labs uses a delta of 25c, about 1.8GPM and 65CPM per fan. 

Or this apparent RTX4090 and 7950x3D using AlphaCool 140mm v2 radiator to cool everything 

https://smallformfactor.net/forum/threads/complete-s4max23-worlds-smallest-4090-build-brickless-5l-s4mini-4090fe-7950x3d-800w-water-cooled.18499/

 

that was also replicated earlier with the 3090

https://smallformfactor.net/forum/threads/s4max-brickless-s4m-w-3090-fe-and-r9-5950x-800w-5l-water-cooled.11277/

 

we can read at EKWB description of radiator performance. 
https://www.ekwb.com/blog/radiators-part-2-performance/

 

and looking at the xtremerigs evaluation almost 10 years ago with 1850 RPM fans( don’t know the CFM/static pressure they get close to 500w cooling capacity with 360mm radiators

 

and we can also look at the 280mm radiator performance and the 140mm radiators

performance down below 

  Reveal hidden contents

 

 

https://www.xtremerigs.net/2015/02/11/radiator-round-2015/5/

 

And at the 280mm radiators showing their performance as well

https://www.xtremerigs.net/2016/02/15/hardware-labs-nemesis-280gts-radiator-review/5/

https://www.xtremerigs.net/2015/11/19/ek-coolstream-ce-140-radiator-review/5/

 

 

some useful documents that wasn’t easy to find. 

 

https://myengineeringtools.com/Thermodynamics/Air_Cooled_Heat_Exchanger_Design.html

ATS_White_Paper_Calculating-loads-liquid-cooling.pdfAquacoolEstimator.XLSHardwareLabsEstimator.XLSXSPCEstimator.XLS

[Kilrah]

On 3/9/2024 at 4:25 PM, NoBrainer said:

people say it’s not an exact science, but how can that be the case?

It is technically an exact science, but in most consumer-available parts the ratings necessary to do the calculations are not precise enough or outright missing. E.g. your case has a mesh to not suck dust in or a fancy front panel, but then you have zero insight on how that's going to impact the airflow. Could be just about anything. Behavior of the rad and attached fans could be known, but Interaction between multiple fans in the case is "complex CFD simulation" territory.

[NoBrainer]

15 hours ago, Kilrah said:

It is technically an exact science, but in most consumer-available parts the ratings necessary to do the calculations are not precise enough or outright missing. E.g. your case has a mesh to not suck dust in or a fancy front panel, but then you have zero insight on how that's going to impact the airflow. Could be just about anything. Behavior of the rad and attached fans could be known, but Interaction between multiple fans in the case is "complex CFD simulation" territory.

Well it should be fairly straightforward to construct a rough estimate that is accurate enough for 99% of use cases when adhering to a few known control variables if we understand the impact they have. 
 

it could be constructed first as an external radiator configuration where there an unobstructed airflow from the air entering the radiator and exiting the radiator that can be matched with the same result when inside a case. 
 

We don’t even need to use a CPU or GPU for the tests, but just provide the same amount of watt/cm2 to equivalent to the desired cooling configuration that is needed to be simulated. 

 

the fan types aren’t even important as long as we can make sure the amount of 

specify airflow in m³/h and the static pressure mm H₂O is consistent 

especialy when the impact of having fans in series or parallel is enormous. 
 

Essentially trying to pin down the barebones most effective way to use the radiator(when knowing its thermal capacity) in a system. Trying to build the system around the cooling needs. 

 

because something is seriously wrong if the truth of the matter is that anything above a single 30mm thick 240mm, 280mm or 180mm radiator is extremely overkill even for overclocking and a temperature delta of 25c/ΔT above ambient. 
 

that means most water cooling systems( excluding aluminum AIO as they are less capable compared to copper radiators) are doing something extremely wrong and breaking their systems with wrongly designed systems around false community “common sense wisdom “

 

showing that having a radiator in the bottom towards the floor, having a radiator in the front with an obstructed esthetic part are doing little to nothing at all to cool the system and just wasted money 

 

do we need to be 100% accurate? Unlikely but having a guide that is 95% accurate is way better than using a rule of thumb that hasn’t been anchored to reality for 20 years. 

 

 

 

[Francy Alinston]

 

This is a video I saw a few days ago it's about an HDMI/DP port connector on a device (Xi3) 

[RollinLower]

26 minutes ago, Francy Alinston said:

 

This is a video I saw a few days ago it's about an HDMI/DP port connector on a device (Xi3) 

i second this one! that port is awesome and should be on all GPU's and monitors IMO.

[Kilrah]

9 hours ago, NoBrainer said:

if the truth of the matter is that anything above a single 30mm thick 240mm, 280mm or 180mm radiator is extremely overkill even for overclocking

I mean more area is always going to be useful, 3 fans spinning slowly and quietly instead of 1 or 2 running faster for the same dissipation etc so not exactly a waste depending on your criteria...

Usually the bottleneck will be the CPU->block->coolant transfer, maybe the coolant flow. On my 360 AIO at 300-ish W I can reduce both pump and fan speed to 70% with no notable difference, coolant around 40°C, the CPU to coolant path is the limit. 60K difference between CPU and rad, 20K between rad and ambient, no question there. Pump between 70% and 100% is 1K difference on rad temp so it's sufficient. Fans at full get the rad about 2K lower and that allows for about 10W more but it just further emphasizes the block as the bottleneck.

 

 

[majorbigpain]

Hey Team,

 

I have a tool that I would love to see you guys check out. It is called the Creaform Metrascan. It is used for Metrology and uses Lasers to scan surfaces for various inspection tasks. It is the only software I have come across that has a bar for remaining Ram usage and vram usage. After scanning a part for about an hour the 64 gb of ram on my laptop is used up. Then I import it into Polyworks to interpret the data for an inspection report. A plus for you guys. Metrascan and Polyworks are both Canadian companies. 

[NoBrainer]

On 3/12/2024 at 9:20 PM, Kilrah said:

I mean more area is always going to be useful, 3 fans spinning slowly and quietly instead of 1 or 2 running faster for the same dissipation etc so not exactly a waste depending on your criteria...

Usually the bottleneck will be the CPU->block->coolant transfer, maybe the coolant flow. On my 360 AIO at 300-ish W I can reduce both pump and fan speed to 70% with no notable difference, coolant around 40°C, the CPU to coolant path is the limit. 60K difference between CPU and rad, 20K between rad and ambient, no question there. Pump between 70% and 100% is 1K difference on rad temp so it's sufficient. Fans at full get the rad about 2K lower and that allows for about 10W more but it just further emphasizes the block as the bottleneck.

 

 

Well AIO tend to be worse as it’s aluminum. But I would say over kill in the sense that everything above x won’t contribute much at all, even if it’s nice to have it won’t matter for the TDp capabilities of the system. . If we have dB floor we want to stay under it’s an easy way to draw the line of overkill, and the CpU/GPU temperatures just need to have enough TDP capacity to saturate it as close as the thermal limit as possible. If the cpu is 70c or 40c doesn’t have any meaningful output as long as the heat gets removed and doesn’t cook other computer parts. 
 

with lab data you should be completely fine running a 140mm radiator and using 3 fans as forced exhaust for the case while forcing air to go through the radiator. 

[ThePCNerd129]

You should make a tier list of all the m.2 and test their speeds and boot times

[zbigfish]

We've seen a little of the Honeywell PTM7950, but I'd like to see a head-to-head comparison between it, Thermal Grizzly KryoSheet, and liquid metal.

[PPush]

Hi all,
my first post here, so I hope I guessed the right section.
 

For those unfamiliar with it, LCZero is a deep neural network-based chess engine and, I know I'm asking a bit, but it would be really useful to have a benchmark on it.

That way, even for those who are not professional chess players like myself, it would be possible to assemble a relatively high-performance computer at low cost.
 

The software is completely free and open source and even already has a built-in command-line bechmark.

[191x7]

5 minutes ago, PPush said:

Hi all,
my first post here, so I hope I guessed the right section.
 

For those unfamiliar with it, LCZero is a deep neural network-based chess engine and, I know I'm asking a bit, but it would be really useful to have a benchmark on it.

That way, even for those who are not professional chess players like myself, it would be possible to assemble a relatively high-performance computer at low cost.
 

The software is completely free and open source and even already has a built-in command-line bechmark.

Or they could add, at least, the Fritz Chess benchmark tool that comes with Fritz Chess. 

[PPush]

On 4/6/2024 at 9:07 PM, 191x7 said:

Or they could add, at least, the Fritz Chess benchmark tool that comes with Fritz Chess. 

The interesting thing with LCzero is that the neural networks are mainly designed to run on the GPU, especially the larger ones that obviously give more accurate moves

Edit.
Just to add some information. LCZero has several backends, obviously CUDA but also ONNX-DirectML.

[SaperPL]

I know it's not exactly right place for this suggestion, but I don't see a dedicated feedback thread started and maintained by the LMG staff, so I'm hanging it on the "providing value to the community thing" stated in the opening post. Maybe there should be a separate dedicated thread to the process feedback.

 

Seeing some GPU test articles pop up on the Labs page, I would like to suggest figuring out a unified physical measurements of the card system with some reference drawings. And a corresponding way of checking how much space is there in the case/chassis GPU compartment.

 

The importance of it is that it's such a mess that we don't know how the card is actually measured on vendor's product page if there's no drawing showing it. And even within one vendor site we get cards that are measured differently - some are measured from the beginning of the PCI bracket (it's fingers) to the top of the card, and some will be measured from the pci-e connector PCB begging to the card top. Then again it's the same question whether the bracket is counted in the card length.

 

On the thickness side it gets tricky as well - you can measure overall thickness and call it a day, but it'll be safe assuming that all cards are offset from the back of PCB the same amount of space, so full 5mm of allowed clearance at the back. But what if the card's cooler retention mechanism doesn't need those 5 mm for spring tension screws and thus the card is sticking out at the back less than 5 mm, so let's say 2 or 3 mm. Then such 40mm "2 slot" card may actually not fit correctly next to the another extension card that uses all that 5mm of space or in a mini-ITX/sff case where card thickness is limited.

 

Another issue is that how much space is needed for the power cables. Some cards will have PEG sockets/connectors recessed to the reference height pci-e add-in card PCB, but some oversized - taller PCB cards will have them placed higher than that which affects how much room the card needs beyond the actual height of the card. And it gets even more important with those new 12-pin power connectors with their limited bend angles - having a number that shows how much space you actually need for your card in the case would be good.

 

From my perspective, the MVP approach would be to include a drawing showing how you are measuring those dimensions on the questionmark tooltips next to the parameter name and sticking to a unified way of measuring the cards and cases, but if you're going to such lengths of CT scanning the cards, then maybe it would make sense to have a process that shows the exact dimensions of the cards on their photos/scans.

 

Finally - you don't have an exact vendor product ID on the test page nor a link to the vendor's product page, just the product name, and that's a problem - cards can have different dimensions of the cooler between variants even to a point where v1 will be a 2.5 slot card and v2 will be 2 slot card. It will be important for other components such as SSDs and memory kits where one product name can have different SKUs with different components internally, and may be varied between distribution regions. For example 16GB G.Skill RipJaws V 3200 MHz CL16 kit would have up to 5 different SKUs according to this community research on ryzen memory compatibility from back in the day: https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

 

[Karavusk]

I originally wrote this as an email but I guess this is the place where I should be posting this:

 

Hello labs team

I was at the Caggtus LAN in Germany around a week ago and I was playing on some free play PCs for a bit which were around there. The mice they had were really nothing special and kinda on the bad side (gaming mouse reaper 400) but definitely usable. BUT.... THE... KEYBOARDS!! Please help me validate my sanity. PLEASE. 

 

I am mostly playing LoL at a relatively high level (master in euw) and I am really not that much of an elitist with peripherals even though I enjoy having high end stuff. When I played on this keyboard... oh my god. The urage exodus 800, a 100€ mechanical keyboard from Hama (MediaMarkt garbage brand, you know the ones that sell you a simple DVI cable for 30€) under their gaming brand.

The switches felt insanely bad and I have no idea how they made actual mechanical blue switches feel this bad. My Temu keyboard I paid 3€ for felt way better than that. Sadly the feeling wasn't the worst part. Please labs team you are my only hope! Tell me why the hell this keyboard felt like it was adding a delay of like 60ms. I was playing with a ping at or under 10ms constantly and it was fine with a different keyboard. What the hell did they physically do to fuck up this badly? IT IS A WIRED KEYBOARD! HOW?!

 

For the sake of my sanity (and to validate my excuse for throwing the game) can you please look into this? Just the latency and not even a full test would be way more than enough. I even advertised my screwdriver and backpack a bit at the LAN

 

Thanks you very much for your help<3

[Xemor]

Dear Labs Team,

 

I'm not sure if this has been suggested before, but the internet is in desperate need of an objective set of machine learning benchmarks for a wide variety of graphics cards. Ideally for training and inference, it doesn't have to be perfect as there is little competing information on the internet so anything would be an improvement. Ideally training a well known NLP model for some number of steps, and doing some amount of inference with the model and seeing how long it takes for a variety of graphics cards. There's a rapidly growing audience of people interested in which graphics cards are the best for machine learning, which is horrifically underserved by the current tech media landscape.

[igormp]

1 hour ago, Xemor said:

Dear Labs Team,

 

I'm not sure if this has been suggested before, but the internet is in desperate need of an objective set of machine learning benchmarks for a wide variety of graphics cards. Ideally for training and inference, it doesn't have to be perfect as there is little competing information on the internet so anything would be an improvement. Ideally training a well known NLP model for some number of steps, and doing some amount of inference with the model and seeing how long it takes for a variety of graphics cards. There's a rapidly growing audience of people interested in which graphics cards are the best for machine learning, which is horrifically underserved by the current tech media landscape.

That's a hard one because there's no common stack between the GPUs, like you have DX or Vulkan for games or other graphical applications.

The differences between each vendor's API makes the results not really comparable between one another, and in the end Nvidia is still the de facto winner in this area until the others manage to catch up.