SystemCooling.com compares CPU stress utilities

[MikeC]

It's a good comparison of various software utilities used to stress processors, kind of a summary of the things we at SPCR went through before settling on CPUBurn as our reference CPU / heatsink / system load utility. Looks like we made the right choice, according to SystemCooling.com

[Ralf Hutter]

Looks like we made the right choice, according to SystemCooling.com

Whew, that's good to know!

And look, they even came up with a 2.52°C difference in the load temps of Prime95 vs. CPUburn, right in between the "2-3°C" that's often quoted by SPCR. They must be doing something right!

[Edward Ng]

Anybody care for some Power Angel results?

System Specs:
DFI LAN Party UT nF3 250Gb
ClawHammer at 1.9volts and 2.6GHz
two 512MB sticks of PQI Turbo PC3200 (TCCD)
one SP1614N
6800GT at 430/1180 and 1.5volts when in 3D
two CSP-750 MarkII pumps
two 120mm Nexus fans
OCZ PowerStream 420 PSU

Idle:
AC Voltage during test: 117.6 Volts
Current being drawn: 2.35 Amps
Watts AC power consumed: 182 watts
VA figure: 278

Prime95, In-Place Large FFTs Torture:
AC Voltage during test: 117.1 Volts
Current being drawn: 3.83 Amps
Watts AC power consumed: 306 watts
VA figure: 450

CPUBurn, K7:
AC Voltage during test: 116.7 Volts
Current being drawn: 4.38 Amps
Watts AC power consumed: 351 watts
VA figure: 511

RTHDRIBL, 1024x768 Planetary Demo:
AC Voltage during test: 117.0 Volts
Current being drawn: 4.43 Amps
Watts AC power consumed: 357 watts
VA figure: 522

The VA figures are proof that APFC is very much useful; look at how much extra VA is drawn over watts due to the lack of PFC on my OCZ PowerStream. It should come as no surprise that even with the circuit breakers upgraded in our apartment, when it's summer time and the air conditioning is active, and I've got several machines on at once, power often clicks out. This happens particularly often when the refrigerator compressor and/or the microwave is activated, as I share my breaker with the kitchen. It's a total wonderment to me that none of my equipment has yet to take damage from the outages.

*Knocks on wood*


This data makes more sense if you have an idea of what the DC load may be on the PSU.

Going by these results from this thread:

GPX Mode (75.1 watts load):
AC Voltage during test: 119.0 Volts
Current being drawn: 1.54 Amps
Watts power consumed: 117 watts
VA figure: 183
Power Factor: 0.64
Resultant Efficiency (75.1/117) = 65.2%

G Mode (137.6 watts load):
AC Voltage during test: 118.5 Volts
Current being drawn: 2.45 Amps
Watts power consumed: 191 watts
VA figure: 290
Power Factor: 0.65
Resultant Efficiency (137.6/191) = 69.1%

As well as Mike's data from his PowerStream 470 review, one can derive some simulated efficiencies and educated guess load figures, based upon power drawn.

Here's what I got:

Idle:
Watts AC power consumed: 182 watts
Estimated DC Draw at ~68% efficiency (182*0.68) = 123.8watts

Prime95, In-Place Large FFTs Torture:
Watts AC power consumed: 306 watts
Estimated DC Draw at ~75% efficiency (306*0.75) = 229.5watts

CPUBurn, K7:
Watts AC power consumed: 351 watts
Estimated DC Draw at ~78% efficiency (351*0.75) = 273.8watts

RTHDRIBL, 1024x768 Planetary Demo:
Watts AC power consumed: 357 watts
Estimated DC Draw at ~78% efficiency (357*0.78) = 278.5watts

-Ed

[Edward Ng]

Okay, some interest was presented to me for numbers that are more...down to Earth, so to speak.

Specs for the second test system:

desktop 3000+ NewCastle
DFI LAN Party UT nF3 250Gb
XP-120, no fan (fully reliant on case exhaust fan and PSU fan)
one 256MB stick of cheap RAM
Seagate ST340014A in an unpowered Cooler Master CoolDrive 6
SeaSonic Super Tornado 400 w/APFC
one AcoustiFan AF120CT case exhaust fan; automatically turns on at certain temperatures (I'll indicate this)
passively cooled PCI GF2mx graphics card


Results for, "Fully Automatic," in the BIOS (2009MHz@~1.5 volts)

Idle (exhaust fan is not spinning):
AC Voltage during test: 118.9 Volts
Current being drawn: 0.59 Amps
Watts AC power consumed: 69 watts
VA figure: 70
Power Factor: .99

Prime95, In-Place Large FFTs Torture (exhaust fan is spinning):
AC Voltage during test: 118.6 Volts
Current being drawn: 0.99 Amps
Watts AC power consumed: 116 watts
VA figure: 117
Power Factor: .99

CPUBurn, K7 & High Priority (exhaust fan is spinning):
AC Voltage during test: 118.6 Volts
Current being drawn: 1.13 Amps
Watts AC power consumed: 134 watts
VA figure: 136
Power Factor: .98


Results for Maximum Underclock/volt (800MHz@~0.75 volts)

Idle (exhaust fan is not spinning):
AC Voltage during test: 118.7 Volts
Current being drawn: 0.42 Amps
Watts AC power consumed: 48 watts
VA figure: 49
Power Factor: .98

Prime95, In-Place Large FFTs Torture (exhaust fan is not spinning):
AC Voltage during test: 118.8 Volts
Current being drawn: 0.46 Amps
Watts AC power consumed: 53 watts
VA figure: 54
Power Factor: .98

CPUBurn, K7 & High Priority (exhaust fan is not spinning):
AC Voltage during test: 119.0 Volts
Current being drawn: 0.46 Amps
Watts AC power consumed: 53 watts
VA figure: 54
Power Factor: .98

As we can see, the higher the consumption of the machine, the greater the difference in power draw between Prime95 High Heat Torture versus High Priority K7 CPUBurn. This was particularly apparent in the really heavily overclocked test system from the previous post, but it shows even here.

I'm still in the process of testing different undervolts with this system. By my guess, at the current pace, I can run this machine stock speed at only about 1.25volts, which should cut down on heat quit a bit over stock. Those results are slowly accumulating in this thread.

-Ed

[GlassMan]

They never make clear which prime95 test they used. There are 3 torture tests, small fft's. large fft's and blend. The cpu temps during the small fft's test is 2 to 3C higher and much more consistant than the large fft's test, which produces more system heat less cpu heat. If you graph the blend test you can see the cpu temps rise and fall, tighten and loosen as the size of the fft's change.

[JohnMK]

Yes, please use small FFT's to produce the most heat. I've done testing on my Hyperthreading system showing with two instances of Prime95 going (small FFT torture) my CPU temps stabilize at 60C, which is the same as when I run two instances of cpuburnp6.

LargeFFT will leave the CPU idling (from the POV of the CPU, not the OS necessarily) too much to produce maximal heat load.

[Edward Ng]

I do pose one small question, though...

In all cases where Large FFTs would fail to pass stability testing, would the Small FFTs test also fail? Remember that I am also testing the stability of the undervolts or overclocks at the same time, which is the whole point of using Prime95 for me; heat and stability testing.

After all, if I want pure heat, without regards to stability, there's always CPUBurn.

-Ed

[halcyon]

Good question Edward!

I have no done that much testing between large and small FFT to know that for sure, but I do know that on many systems's large FFT seem to fail more often (or faster).

The thing with Prime95 stability testing is that failures due to overclocking/undervolting/overheating/other hardware failures are not systematic.

Sometimes one can go with 8 hours straight without a single failure in Prime95.

Then during another test it fails within the first 30 minutes.

This is of course a sign of instability, but the sporadic occurences can make it more difficult to spot.

Ideally one should heat test (say cpuburn k7) and stability test (prime95 several runs of several hours separately).

regards,
halcyon

[JohnMK]

It would seem to me that when your goal is to stress stability, you need to run both tests. Might as well run all three. Large FFT's would probably stress the northbridge and the system RAM where the small FFT test would not. I can imagine any flaws that exist there might be more readily brought to the surface with the large FFT test, since they do tend to overflow the L1 & L2 caches.