Fan Testing Loads - Suggestions, please!

[TPCEA]

Do these seem like reasonable loads?

Load types (with startups, except #1):

1. Control: BIOS
2. Light: Asus Armoury Crate, Corsair iCUE, HWInfo64
3. Medium: The above plus Firefox with 11 tabs and Adobe Acrobat Reader
4. Heavy: HWInfo64, Steam & 3DMark's Speed Way Benchmark
5. Ultra: All of #3, Steam & 3DMark's Time Spy Extreme Benchmark

If you think these loads are not adequate or need balancing, please let me know which and how. I also have PC Mark, although I'd prefer not to use the complete or extended tests unless there's no other option. I could also use Prime95 if that would be a good option. I'm considering a 6th level (extreme), so would love input on that.

 

Separately from these, I will also conduct airflow and noise tests.

[TeraSeraph]

You might be better off limiting your CPU to certain wattages in bios then running aida64 or prime95. This will give you the most consistent results.

[PDifolco]

They are consistent, but you're mixing CPU and GPU loads, which have different effects on temps

If you're only looking at CPU cooler fans only use CPU loads, same for GPU

A good GPU Only intensive benchmark is MSI Kombusor, and it has a built in CPU burner that loads CPU to 100% 

[TPCEA]

3 minutes ago, TeraSeraph said:

You might be better off limiting your CPU to certain wattages in bios then running aida64 or prime95. This will give you the most consistent results.

Are you referring to an extreme load? I'll have to learn how to do that. I'd most likely get prime95 at this point since I've put out too much money for this project already.

 

I forgot to get those fans I asked about yesterday, so the discounts+rebates are gone. Oh well.

[TPCEA]

5 minutes ago, PDifolco said:

They are consistent, but you're mixing CPU and GPU loads, which have different effects on temps

If you're only looking at CPU cooler fans only use CPU loads, same for GPU

A good GPU Only intensive benchmark is MSI Kombusor, and it has a built in CPU burner that loads CPU to 100% 

It's no accident that I'm not using just CPU/GPU testing. That is the reason I use 2 CPU censors, a mobo sensor, a VRM sensor and the GPU (slot?) sensor. Does that help?

[TPCEA]

6 minutes ago, micha_vulpes said:

If you are testing fans on coolers, I would also toss in a real world use case :
Something that runs maybe half the cores to simulate a modern game,  And Something that runs on all the cores and tends to leverage the entire chip, rather than getting hung up on a single part like the math co processor nodes For the life of me I cannot think of what we call these now. FPU?

• CB R23 I like because it has pauses when it resets. You can sometimes get core instability dropping from high load to idle instantly like it does
• Converting Blue Ray Source content to a super high quality mp4 is a pretty typical "high stress" end user workload representative of what average home users might do.

Things like p95 SFFT are great for testing thermal limits, but I do not think its a real world use case. I have also had CPU's pass 24+ hours of Small FFT, or Mixed FFT, Only to crash 10 minutes into a folding@home workload or Cross compiling software where there are tonnes of fast, repeated starts and stops on random threads as it clears qued tasks.

 

For GPU testing oddly I like two older benchmarks on loop

• Heaven Benchmark
• Valley Benchmark

These are both clearly old, BUT at least on the last 3 cards I have had they actually seem to push the cards thermally harder than SuperPosition or TimeSpy with a lot more coil "crunching" and generally hotter core temps ( my a4000 only hits 81C on Super Position regardless of settings, but hits 90C on heaven). Heaven is particularly hard on the card and driver stack in my experience when it transitions to the first dimly lit interior scene, even high end cards drop frame-rate hard here and often stutter, bad vram overclocks will often start to get the flashes and random colored pixels at this spot too.

I don't mind real-world testing so long as it's automated and isn't going to cost me anything. The thousands I've sunk into this project sting already.

 

The other day, I ran prime95 on my computer for several minutes with only a DC FK120 on the left side of my U12A just to see what happened. Apparently, that wasn't a great idea because my Internet turned off (mobo overheated?). I had to turn off my PC and let it cool off for a bit...

 

I've just installed Heaven, Valley and SuperPosition basic versions to see what happens. I'll try to remember to update later.

[PDifolco]

28 minutes ago, RevGAM said:

It's no accident that I'm not using just CPU/GPU testing. That is the reason I use 2 CPU censors, a mobo sensor, a VRM sensor and the GPU (slot?) sensor. Does that help?

Thing is the CPU and GPU temps are not independent, a hot GPU makes the ambient higher so impacts the CPU, and vice versa, even if everything is watercooled (as the parts themselves heat up, I did realize that when I had such a setup)

Plus temps also depend on the cooler, case etc, so any benchmark can't be really transposed to other users and use cases

[TPCEA]

4 minutes ago, PDifolco said:

Thing is the CPU and GPU temps are not independent, a hot GPU makes the ambient higher so impacts the CPU, and vide versa, even if everything is watercooled (as the parts themselves heat up, I did realize that when I had such a setup)

Plus temps also depend on the cooler, case etc, so any benchmark can't be really transposed to other users and use cases

I understand this, but the only way to overcome these issues is to do lab testing, which is so devoid of real-life that the numbers are even less meaningful - they're idealized to the extreme because of controlling for virtually every variable.

 

Additionally, I have neither the time, space, engineering background, mathematical ability, money nor equipment to engage in such testing. I'll leave that the Hardware Busters, GN, LTT, HWCooling, Quazarzone (where applicable) and others to create idealized numbers that don't match up to ANY setup at all.

[TPCEA]

SuperPosition benchmark on 1080p Medium (default), with only a DC FK120 on the right side of the U12A set to turbo, and all case fans set to minimum.

[PDifolco]

4 minutes ago, RevGAM said:

I understand this, but the only way to overcome these issues is to do lab testing, which is so devoid of real-life that the numbers are even less meaningful - they're idealized to the extreme because of controlling for virtually every variable.

 

Additionally, I have neither the time, space, engineering background, mathematical ability, money nor equipment to engage in such testing. I'll leave that the Hardware Busters, GN, LTT, HWCooling, Quazarzone (where applicable) and others to create idealized numbers that don't match up to ANY setup at all.

Ok, that makes sense, if you want to test your hardware and cooling

Now about case fans (that seeming to be your subject), my experience is that if you put 3 decent front fans and one back, all at around 1200 rpm for quietness, to create a good airflow your'e good and won't get much by adding 6 more

[TPCEA]

Heaven benchmark (much lower quality, to my eyes, than SP), defaults except on Ultra.

 

I was typing with it still on the results screen (Ultra) and boy was that slowing down my computer, unlike SP. My typing speed was way slower than reality...Perhaps 2/3 to 1/2!

[TPCEA]

10 minutes ago, PDifolco said:

Ok, that makes sense, if you want to test your hardware and cooling

Now about case fans (that seeming to be your subject), my experience is that if you put 3 decent front fans and one back, all at around 1200 rpm for quietness, to create a good airflow your'e good and won't get much by adding 6 more

I know. But, again, I'm asking if the loads I've chosen are appropriate and balanced for what I'm doing. Also, what would be good for Extreme?

[TPCEA]

Valley, ultra, everything else the same. Also lagged typing. Not much improvement in the visuals. I'd like to see SP-level benchmark outdoors to see if they actually improved the graphical quality. All 3 have enjoyable music.

[PDifolco]

9 minutes ago, RevGAM said:

I know. But, again, I'm asking if the loads I've chosen are appropriate and balanced for what I'm doing. Also, what would be good for Extreme?

Yeah I've said they're ok

For Extreme I'd say run CBR or Prime95 + Kombustor GPU "stress test", will put everything at 100%

[TPCEA]

8 minutes ago, PDifolco said:

Yeah I've said they're ok

For Extreme I'd say run CBR or Prime95 + Kombustor GPU "stress test", will put everything at 100%

Thanks! Should I have anything other than HWInfo64 running as well, just to really tax it?

[PDifolco]

2 minutes ago, RevGAM said:

Thanks! Should I have anything other than HWInfo64 running as well, just to really tax it?

Don't think it'll change much as CPU will be @100% with CBR, if you divert some CPU % usage to something else total will still be 100  

[TPCEA]

1 minute ago, PDifolco said:

Don't think it'll change much as CPU will be @100% with CBR, if you divert some CPU % usage to something else total will still be 100  

@TeraSeraph@micha_vulpes

Any thoughts?

[TPCEA]

Cinebench R23 Single Core:

[TPCEA]

Cinebench R23, multi-core:

 

 

Please note that Cinebench incorrectly identified my Windows as 10. I have 11 Home.

 

[TeraSeraph]

2 hours ago, RevGAM said:

Are you referring to an extreme load? I'll have to learn how to do that. I'd most likely get prime95 at this point since I've put out too much money for this project already.

 

I forgot to get those fans I asked about yesterday, so the discounts+rebates are gone. Oh well.

No, go into your bios or XTU and limit the CPU wattage, you can make your computer artificially throttle with a stress test to say 50W, 75W, 100W, and 150W rather then layering various computer loads. It will be more consistent.

[TPCEA]

12 minutes ago, micha_vulpes said:

I really do not know why this is the case, maybe because the older API's cant parallelize as well, or are less efficient at scheduling the render thread but they just seem to really tax it differently than SP does

Well, all I can say is that the differences were real, and seem to reflect the age of the software, except for the graphical quality of H & V seemed almost the same. I find it interesting that Valley was less taxing than Heaven, but not really surprised because it doesn't present as wide a variety of surfaces and reflectivity as Heaven. The raindrops were poorly rendered and disappeared instead of staying on the "lens" or sliding down, and some of the foliage didn't respond to the rain.

                         Avg FPS    CPU Temps        Mobo Temps   VRM Temps   GPU Temps

Heaven            442.4        45-68, 47-70           36-41              46-55           52-77

Valley               239.7        39-63, 44-62           34-35             50-51            ~56?-66

SuperPosition 226.36       35-50 & 41-60         31-36             39-43           48-74

CR23               N/A             39-86, 38-99          31-32              38-63           41-45

 

Pretty interesting, though, that SP had the least impact overall on temps, and second most on the GPU. I don't find it surprising that CR23 didn't really impact the GPU or mobo, so I'm not sure if this really qualifies as extreme. That said, I did forget to exit some programs, including FF.

[TPCEA]

29 minutes ago, micha_vulpes said:

I don't think there are many scenarios like this real world either. Usually if you are building a large code base or something you only do n+1 or n+2 threads, where n is your CPU threads. Normally running 2n high use high priority threads just won't change it over n+1

If I follow your meaning...I'm doing this many loads because I, ideally, want the data to be useful for people with different usage levels, from the casual user doing very little all the way up to heavy-duty art and AI creators.

 

I don't understand your last sentence. The increase from Ultra to "Extreme" via CR23 is interesting. I want to show a graph here, if I can just use that data.

[TPCEA]

Something is wrong. I cannot get it to make a graph showing the two data sets in a line graph. It's probably my own stupidity.

[TPCEA]

3 hours ago, PDifolco said:

They are consistent, but you're mixing CPU and GPU loads, which have different effects on temps

If you're only looking at CPU cooler fans only use CPU loads, same for GPU

A good GPU Only intensive benchmark is MSI Kombusor, and it has a built in CPU burner that loads CPU to 100% 

Which Kombustor and which prime95 settings?

[TeraSeraph]

31 minutes ago, RevGAM said:

Something is wrong. I cannot get it to make a graph showing the two data sets in a line graph. It's probably my own stupidity.

Nope, it's a flaw in excel's programming. No one has bothered to fix it for 6 years.