Furmark+IntelBurnTest (simultaneously) fail (always furmark crash)

[londiste]

Is stress testing something that has gone out of vogue or something? Stress test by definition is a test designed to assess how well a system functions when subjected to greater than normal amounts of stress or pressure. If the system passes stress test, you can be quite sure it will take any normal stress and be fine.

 

[EarthDog]

It is not for testing clocks. It is a stress test to test power consumption and temperatures.

But, it is, for many(most), to stress test for STABILITY (see the OP). Annnnnnnnnnd, NVIDIA says not to test power with this in the first place.

EDIT: And if we are getting the same power as testing with 3DM FS, then why would we want to have something more stressful just pound on the limiter for the same result? That doesn't remotely make sense to me.

So, is there an answer to my question for those who use this, like the OP, to find stability in a system when its HUNDREDS of MHz off from running clocks? Can we admit it isn't good for that at minimum?

 

[londiste]

What does frequency have to do with stability on stock clocks? Limits do, as well as behaviour around them.
By your argument, Furmark should be too easy a test as frequency is off to a low side.

EDIT: And if we are getting the same power as testing with 3DM FS, then why would we want to have something more stressful just pound on the limiter for the same result? That doesn't remotely make sense to me.

Why not? And why would it be more stressful when you just said it is the same power as 3DM FS?
If you look at the Tomshardware test results Furmark is a bit heavier on memory compared to 3dM FS which explains the small difference in power consumption.

 

[qubit]

Is stress testing something that has gone out of vogue or something? Stress test by definition is a test designed to assess how well a system functions when subjected to greater than normal amounts of stress or pressure. If the system passes stress test, you can be quite sure it will take any normal stress and be fine.

In principle you're correct. However, consumer kit is built down to a price and hence robustness isn't always what it should be, especially with budget components and if the components are old. Hence, it's possible that these stress tests can damage them. Therefore, something that doesn't stress a system to its limits is a more prudent option.

 

[EarthDog]

What does frequency have to do with stability on stock clocks?

I should't have to answer this.. you know why... and I have said why, quite consicely and clearly multiple times already.

By your argument, Furmark should be too easy a test as frequency is off to a low side.

Que? How is it an 'easy' test when clocks that are HUNDREDS of MHz lower and notably lower voltage 'easy'?

Brother, when testing a CPU overclock, do you test at the clockspeed you run at or towards the base clock? Why would a GPU be any different? How is testing at 1680 MHz .8xxV the same as testing running clocks of 1950 MHz 1.05V?

I simply don't know what else I can say here... the logic just B slaps me in the face on this one, LOL!

 

[londiste]

Que? How is it an 'easy' test when clocks that are HUNDREDS of MHz lower and notably lower voltage 'easy'?

Are you really trying to tell me lower clocks and lower voltage is a more stressful test than higher clocks and higher voltage?

Brother, when testing a CPU overclock, do you test at the clockspeed you run at or towards the base clock? Why would a GPU be any different? How is testing at 1680 MHz .8xxV the same as testing running clocks of 1950 MHz 1.05V?

Furmark is not for testing (gaming) clock speeds. It is not meant for it, it is not good for it. I think I have written this several times already.

 

[EarthDog]

Are you really trying to tell me lower clocks and lower voltage is a more stressful test?

No. How did you extrapolate that from what I have been saying.

I am asking a couple of things (that seem to be refused to be answered............)

1. How is it the SAME test for STABILITY when one is using (in my case) 300 MHz less clocks and notably less voltage using the same power as running the clocks at ACTUAL boost? Right? How is testing at 1650 Mhz 0.8xxV the same as testing running clocks of 1950 MHz 1.0xV?
2. Why do we test CPUs at the clocks they run at as a norm/standard, but its OK to run a GPU hundreds of MHz less and call that OK?

EDIT: YOU say its for power consumption... yet, we see the vast majority of users, like the OP, run it as a stability/stress test. Isn't that one (of a couple) points to run a stress test is to see if it is stable where you are at, be it stock or overclocked??? Come on now....how many times have you seen me say, in this forum alone, not to run this to test stability???

 

[londiste]

I am asking a couple of things (that seem to be refused to be answered............)

1. How is it the SAME test for STABILITY when one is using (in my case) 300 MHz less clocks and notably less voltage using the same power as running the clocks at ACTUAL boost? Right? How is testing at 1650 Mhz 0.8xxV the same as testing running clocks of 1950 MHz 1.0xV?
2. Why do we test CPUs at the clocks they run at as a norm/standard, but its OK to run a GPU hundreds of MHz less and call that OK?

This started with stress testing at stock.
1. How do you want to define stability? Clocks - again, especially if we are talking about running things at stock - are only one part of the problem. And the frequency range is guaranteed in spec. Limits are a much bigger problem, especially with GPUs.
2. Without OC - and going beyond the spec - you are likely to test a contemporary CPU at hundreds of MHz less than maximum boost. It will run into power limit.

EDIT: YOU say its for power consumption... yet, we see the vast majority of users, like the OP, run it as a stability/stress test. Isn't that one (of a couple) points to run a stress test is to see if it is stable where you are at, be it stock or overclocked??? Come on now....how many times have you seen me say, in this forum alone, not to run this to test stability???

There is no such thing as guaranteed clock speed for a GPU. Has not been for years.
When I play games, my GPU frequency is anywhere between about 1820 to 2040 MHz. Clock depend on how much power GPU/card pulls (which Furmark does test) and what temperatures it gets (again, which Furmark does test). It does not test for maximum possible frequency. Or stability of that frequency in every situation. No single test does.

 

[EarthDog]

This started with stress testing at stock.
1. How do you want to define stability? Clocks - again, especially if we are talking about running things at stock - are only one part of the problem. And the frequency range is guaranteed in spec. Limits are a much bigger problem, especially with GPUs.
2. Without OC - and going beyond the spec - you are likely to test a contemporary CPU at hundreds of MHz less than maximum boost. It will run into power limit.

There is no such thing as guaranteed clock speed for a GPU. has not been for years.

1. Stability is defined as being able to run at your given clockspeed...stock or overclocked.
2. My CPUs boost to their all core/thread clocks and single thread clocks without issue in P95 with AVX or AIDA64 with AVX/AVX-512.

RE: No such thing as a guaranteed clock, you are absolutely right. You'll notice earlier I specifically said "GENERALLY" when talking about clocks in my post. The point there is when gaming (or testing with 3DM FS), clocks are generally A LOT higher than when testing in Furmark. They tend to stabilize around a general clockspeed once temperatures stabilize. This clock value is WELL over the RATED boost one sees in GPUz. When running Furmark, the clocks instantly drop to BELOW BASE BOOST clock and stock voltage... it approaches the BASE CLOCK (because it is doing what it can to fit within TDP). As I said, in my examples, the 2080 I have runs around 1950 MHz gaming or 3DM FS... but when I bench furmark, it runs in the mid 1600's with a lot less voltage.

I just don't see the point in testing something, for power or stability, when for furmark it is literally POUNDING on the limit and CONSTANTLY trying to stay below TDP. It is quite simply apples and oranges.

 

[londiste]

1. Stability is defined as being able to run at your given clockspeed...stock or overclocked.

By that definition no current CPU or GPU is stable at stock

I just don't see the point in testing something, for power or stability, when for furmark it is literally POUNDING on the limit and CONSTANTLY trying to stay below TDP. It is quite simply apples and oranges.

That is exactly what every other load on the same GPU does. They will simply manage to stay at a higher clock rate thanks to lower ALU occupancy.

 

[EarthDog]

By that definition no current CPU or GPU is stable at stock

Sorry? What? What are you saying here????? That isn't remotely true...?!!!

 

[londiste]

Sorry? What? What are you saying here????? That isn't remotely true...?!!!

What exactly did you mean by given clock speeds?

1. Stability is defined as being able to run at your given clockspeed...stock or overclocked.

 

[EarthDog]

That is exactly what every other load on the same GPU does. They will simply manage to stay at a higher clock rate thanks to lower ALU occupancy.

You missed the part where I said it "generally" settles to a 'stable' clock it seems... my clocks on the GPU when gaming hover around 1950 MHz.. could be a bin or two higher, could be a bin or two lower... that is how these cards work, spot on. What Furmark does though, is hit so hard against the limit that it has to lower clocks HUNDREDS OF MHz as well as lowering voltage in order to fit within that same power envelope. Clearly the card is working MUCH harder to maintain TDP/Temps with Furmark than it is with gaming. Why bother using it if the power is the same but one just bangs off the rev limiter harder and with greater losses?

What exactly did you mean by given clock speeds?

For a CPU, it is whatever you set it at, being stock or overclocked. For a GPU, see above... due to how these works, it is difficult to say EXACTLY 1950 MHz (where my card runs)... but you don't think for stability and stress testing that running hundreds of Mhz lower isn't actually testing your clocks? All that tells me is Furmark is putting so much stress on the GPU that it can't come close to running its 'given'/actual clockspeeds or voltages.

Also, see my screenshot from above. Furmark is on the Left, Unigine Heaven on the right. Fan set at 55% manually, power limit at stock. That is the 4K UHD bench (does the same at 2560x1440 as well).

 

[Shambles1980]

"given" is a word to generalize something to its specific state...
for example if i were to say..
These second gen i5 cpus can all run fine at their given clock speeds..
you could then check what each cpu's individual speed was. "if you wanted to know the specific speed"
No one expects people to state the stock speed of every variation of an i5 2nd gen in a sentance of that manner.

 

[londiste]

Also, see my screenshot from above. Furmark is on the Left, Unigine Heaven on the right. Fan set at 55% manually, power limit at stock. That is the 4K UHD bench (does the same at 2560x1440 as well).

Heaven runs completely into voltage limit?

 

[trog100]

i currently have my 2080ti set to a max power limit of 75% furmark hits its power limit (boost) at about about 900 mhz.. with the max power limit set to 125% the boost goes up to around 1750..

furmark wont do the slightest harm.. its all now controlled in the bios just like the max power usage is.. both a modern cpu and gpu are now pretty much idiot proof out of the box..

you are living in the past earthdog my friend.. time you dropped this vendetta against poor old furmark.. he he

trog

 

[EarthDog]

you are living in the past earthdog my friend.. time you dropped this vendetta against poor old furmark.. he he

It came from the RTX guide... that isn't so far past its not relevant.

 

[Shambles1980]

i still say just run occt psu stress test rather than ibt + furmark. its prety similar but has more benifits.

 

[Vayra86]

This started with stress testing at stock.
1. How do you want to define stability? Clocks - again, especially if we are talking about running things at stock - are only one part of the problem. And the frequency range is guaranteed in spec. Limits are a much bigger problem, especially with GPUs.
2. Without OC - and going beyond the spec - you are likely to test a contemporary CPU at hundreds of MHz less than maximum boost. It will run into power limit.

There is no such thing as guaranteed clock speed for a GPU. Has not been for years.
When I play games, my GPU frequency is anywhere between about 1820 to 2040 MHz. Clock depend on how much power GPU/card pulls (which Furmark does test) and what temperatures it gets (again, which Furmark does test). It does not test for maximum possible frequency. Or stability of that frequency in every situation. No single test does.

GPU Boost 3.0 is stock for Nvidia cards. They get a base clock and GPU Boost does the rest - in both directions: it will also throttle if your kit gets too hot.

Every decent stress test does test for maximum possible frequency, in fact, within the limitations of GPU Boost 3.0. A good stress test can simulate a real world load, and in those, the GPU will find equilibrium between temps, highest possible clock rate and voltage. Furmark however is not a real world load, and driver/BIOS contains flags to make sure GPU Boost 3.0 does specifically NOT what it is supposed to do. How? Simple: you get a hard lock on voltage, one of the key variables for GPU Boost to work proper. You simply cannot use the whole clock/voltage curve that is set at stock for these cards, regardless of temperature and regardless of actual power usage at the wall.

So while there is no guaranteed clockspeed for an Nvidia GPU, there is a guaranteed GPU Boost 3.0 behaviour, and Furmark presents a situation where that behaviour is... adjusted. It is also the only stress test that manages to do this, its the odd one out so we can split hairs about frequency and varying clocks, but that is not the underlying cause for what you see in Furmark.

 

[EarthDog]

Heaven runs completely into voltage limit?

Yes, but not typically that hard... lol. I was running something custom. Here is a better SS. First is a full Heaven Extreme run (using Hwbot wrapper), then 4K UHD, then 2560x1440.

Obviously, both are hitting the power limit, but CLEARLY one is slamming into it with reckless abandon causing the clocks to drop precipitously down hundreds of MHz just to fit within TDP while the other is pretty damn stable throughout...a bin or two like I said above.

 

[londiste]

Furmark however is not a real world load, and driver/BIOS contains flags to make sure GPU Boost 3.0 does specifically NOT what it is supposed to do. How? Simple: you get a hard lock on voltage, one of the key variables for GPU Boost to work proper. You simply cannot use the whole clock/voltage curve that is set at stock for these cards, regardless of temperature and regardless of actual power usage at the wall.

Err... anything to back up that claim?
Furmark will simply run into power limit. Perfectly normal with GPU Boost 3.0. Most stress tests will run into power limit in the exact same way.
AMD has similar power management system. I think it is currently called PowerTune.

Obviously, both are hitting the power limit, but CLEARLY one is slamming into it with reckless abandon causing the clocks to drop precipitously down hundreds of MHz just to fit within TDP while the other is pretty damn stable throughout...a bin or two like I said above.

Try Firestrike? Shadow/Rise of Tomb Raider? Witcher 3? They'll all slam the power limit in the exact same way. With somewhat varying frequencies. And no, they probably won't go down that low.

 

[EarthDog]

Err... anything to back up that claim?

What claim... that it drops precipitously due to slamming off the power limt? The screenshot is the proof. Look at what Heaven does and look at furmark...

Furmark will simply run into power limit. Perfectly normal with GPU Boost 3.0. Most stress tests will run into power limit in the exact same way.

Indeed. Except FM hits the cards so hard, that it lowers clocks HUNDREDS of MHz/voltage tenths, just to have it fit within the the power envelope. I'm not sure I can post any more proof that what we see right here. What does that graph tell you if it doesn't show FM hitting the power limit a lot harder?

Try Firestrike? Shadow/Rise of Tomb Raider? Witcher 3? They'll all slam the power limit in the exact same way. With somewhat varying frequencies. And no, they probably won't go down that low.

The won't come close to that low...they'll be close to those running clocks! I've tested this before... but will be happy to provide another screenshot though the burden of proof doesn't lay on these shoulders... look here in a couple mins......

I can also run SOTR and show the same thing if you like.......

 

[londiste]

What claim... that it drops precipitously due to slamming off the power limt? The screenshot is the proof. Look at what Heaven does and look at furmark...

The claim was that there are flags in driver/BIOS specifically for Furmark, locking voltage.

Indeed. Except FM hits the cards so hard, that it lowers clocks HUNDREDS of MHz/voltage tenths, just to have it fit within the the power envelope. I'm not sure I can post any more proof that what we see right here. What does that graph tell you if it doesn't show FM hitting the power limit a lot harder?

Again, Furmark is not a test for clock speeds.
Firestrike will flat out be power limited all the way.

Stop with the clocks already. I get it. Furmark runs at a lot lower clock speeds. I have never argued that. We knew that from get-go. Everyone knows.
It runs into power limit. So does everything else we mentioned here.
Well, not Heaven apparently. Which makes it a good test for high clocks but not for much else and these clocks won't be representative of a game any better.

edit:
Interesting. Did you run Furmark at 2160p and 8xAA? When I try this I get about 90% of power limit and 2040MHz (dropping to 2012 by 76C) :/
Furmark should be most evil with 720p and no AA

 

[EarthDog]

Stop with the clocks? Sure thing.. one last image (after your suggestion again note) to show that 3DM FS runs high clocks while also tickling the power limit.. SOTR does the same thing, but again, clocks are hundreds of MHz higher...

Flat out power limited...it tickles it.................yet reaches typical gaming clocks and isn't pounding on the power limit requiring the clocks and voltage to drop in order to maintain it.

Yeah, great stress/stabilty/power test...

EDIT: In the end, everyone will do what they want. But after seeing what I have been seeing, reading what NVIDIA and AMD call it and suggest not to use it. I don't see this application as terribly useful compared to others that are out which show the same things, and actually test at proper clocks. I get the 30K foot view of clocks do not matter, but people use this as a stability/stress test man... it fails miserably at that because it cannot seem to test proper clocks. There is stress testing your GPU and there is this... you/users don't have to agree, but every time I run across people saying its OK, I'm throwing this information out to let users make the decision for themselves. I'm not a lemming, but I will follow this.

 

[MrGenius]

Meanwhile...@der8auer is running Furmark and Prime95 simultaneously, for hours on end, to test stability for his signature pre-overclocked systems being sold @caseking.de.