How is it improved stability at a specific overclock if it throttles by 150Mhz? I think the term "stable" is incorrectly used to describe that situation.
 
Here's my thing. If I have cards that get boost clocks around 1243 and after a while when 70c is reached they drop to about 1230Mhz and stick there for the duration of my game and I have zero artifacts, no crashes, or errors at all. What will this do for me? It seems, if feniks is to be believed, that I will lose performance only to gain a lower TDP usage number. Why would I want this? Also, I have never seen my memory clock speed change when I reach a boost clock throttle point.
 
The other thing is, why does base clock matter? My cards never run at base clock in games. They are always boosting way beyond what the base clock is set to.

dredd

How is it improved stability at a specific overclock if it throttles by 150Mhz? I think the term "stable" is incorrectly used to describe that situation.

Here's my thing. If I have cards that get boost clocks around 1243 and after a while when 70c is reached they drop to about 1230Mhz and stick there for the duration of my game and I have zero artifacts, no crashes, or errors at all. What will this do for me? It seems, if feniks is to be believed, that I will lose performance only to gain a lower TDP usage number. Why would I want this? Also, I have never seen my memory clock speed change when I reach a boost clock throttle point.

The other thing is, why does base clock matter? My cards never run at base clock in games. They are always boosting way beyond what the base clock is set to.

 
well, it seems to me your core throttling is related to temperature (thermal throttling), try keeping the cards below 70-72C threshold at all cost all the time and it won't happen (mine doesn't exceed 64C under load so far).
however as you noted, the "stability improvements" were achieved by lowering the actual TDP under load by means of more aggressive power throttling (reduces core speed under load in 13mhz increments until TDP is brought down to "safe" zone) ... so is stability improvement working this way? yes it does, but it sacrifices a little bit of performance at same time too ...
 
mind that each app loads the gpu in a different way, I found the Test #1 of 3Dmark11 to be the most stressful for Keppler and here I could see deepest core power throttling vs actual tdp (while heaven3.0 is like nothing to it and zero power throttling occurs).
 
the 150MHz drop is my bad, I read it wrong from the graph (screenshot is a tad blurry on the black background of graph, minimum value is 1150Mhz), but I will double check it later again in precision itself and mouse over the lowest dip to read the lowest core speed (when most of power throttling occurred). it could have been between 50 and 100Mhz drop though, still quite significant.
 
I am not saying this is bad per se, actually it does help stability, because the card doesn't spike TDP to unsafe/unstable zone anymore under great stress (which could result in a crash), so this is a good thing in this meaning, but at same time the max boost clock is reduced ... but then since we cannot up the voltage to compensate, I would say this is the way it is with Kepplers ... probably best if this issue is left ignored.

Yeah thermal throttle but these cards are way too loud to force them to stay under 70c all the time. I am stable even at high TDP usage, no artifacts, bluescreen, freezing, forced restarts, or crashing of any kind. 
 
Unstable TDP doesn't sound right to me. If it was running at unstable levels I would crash, but I can't make it happen at the clocks I have set. 1243Mhz max boost 6772Mhz memory and +145% power target. 
 
What I'm wondering is if someone never crashes, why would they want TDP throttle increased and performance decreased?
 
All these limits are BS. Throttle at 70c...pathetic IMO. 80c is a better number...what is Nvidia doing?

Just wondering if I would gain anything by doing this to my 670 FTW's....
 
If I'm understanding feniks correctly, in a roundabout way they just increase stability by lowering clocks sooner than before?
 
If that's the case that's not increasing stability at all. It's just lowering the clocks. But maybe I didn't understand feniks correctly.

OK, I re-run the test, this time with my favorite Adaptive VSYNC enabled (my score went up in 3dm11 by over 120 points) and the power throttle I was talking about dipped by 59MHz (from 1242MHz to 1177Mhz) on Test#1 in 3dm11. during Test#2 it dipped down to 1202Mhz (40MHz lower than max 1242Mhz). it didn't do it right away, only after some time passed. maxed GPU temp reported was 61C, max TDC reached was 126.2% (used to be close to 133%) with Power Limit of 122%.
 
generally, it doesn't bother me greatly, because the GPU load (stressful for Keppler or non-stressful) depends on the app.
 
will run it through Metro 2033 benchmark on highest detail and see how deeply it throttles down, that would give an idea what to expect while gaming with current OC.
 
screenshot taken after 3dm11 run in Performance mode (adaptive vsync enabled):


Uploaded with ImageShack.us

thebski

Just wondering if I would gain anything by doing this to my 670 FTW's....

If I'm understanding feniks correctly, in a roundabout way they just increase stability by lowering clocks sooner than before?

If that's the case that's not increasing stability at all. It's just lowering the clocks. But maybe I didn't understand feniks correctly.

 
that's tricky, because the power throttling depends on the kind of load (heaven load means nothing while 3dmark11 is a different story), in some apps it will nicely hold the core speed steady throughout the test, because actual TDP was lower (e.g. no more than 120%), in others it will start power throttling because TDP tries jumping way overboard ...
 
 so far I see it does both, power throttles easier and earlier and does it at deeper levels than before ...
 
I am not sure if I like it myself or not ... works stable though, just that semi-automatic core speed (throttling) is a tad annoying ...

huh! the good news is that it did NOT power throttle even once during 3 passes of Metro 2033 benchmark ... held a steady core speed of 1242Mhz, max TDP reported was at 119.7% ... those Kepplers are very intriguing haha!

one problem however ... my GPU temp sensor disappeared from the task bar even though it is enabled in Precision X ... and I can't make it show up again ...

Well, my flash went without a problem thank gawd. I guess i should go stress it out or something. Thanks EVGA!

Macros

iGamer4tv yes your card's BIOS has been updated correctly.

Thank you so much man!