Hi! I am new here, although I am a happy user of Afterburner since three years ago. I recently replaced my 960 with a brand new 3070 (Asus TUF OC), which works fine. I am using Afterburner 4.6.3 beta 5, and I have defined custom fan curves and a moderate undervolting curve, with which I am achieving great scores in 3DMark, complete stability and reasonable temperatures (average 63 degrees in TimeSpy, with a room temperature of 23). I haven't touched any power or thermal limits, all standard: 100% and 83 degrees. I have found some weird things, though, when playing Fallout 4 these days (not a "heavy" game). Sometimes, the Afterburner logs show sporadic spikes of "Thermal limit" or "Power limit". They have no effect on performance, but it's weird... For example: 1) Thermal limit=1 when GPU temperature is actually 49 degrees 2) Power limit =1 when there is a spike to 114 or 135 Watts (before and after the spike, it's 50 Watts) These spikes are very brief, just one sample in the Afterburner log. What puzzles me is: why? If the thermal limit is 83, and the power limit is around 230 Watts, why should the boolean flag go to 1 even if the actual values are much below these thresholds? Is there some filtering on the temperature and power values, so that they appear to be fine while the "raw" value causes the driver to report an "over threshold" condition? Or is it simply a weirdness of the current drivers (461.40)? (I can send the recorded log file to anyone who wants to see it)
Same thing with the TL on a 3080 Strix OC so I'd say that it's either Asus Ampere thing or just Ampere in general. PL though is different since 3080 runs on PL most of the time.
Well, yes, I was mentioning Fallout 4 (capped at 60 FPS) because it's an old game and thus it requires only little GPU power (on average 50 Watts), with occasional spikes to around 130... which is still much lower than the 3070 power limit. And yet, during those spikes the "power limit" flag goes to 1. Of course with more recent and heavier games, all maxed out and with no FPS caps, the card stays almost constantly on power limit.
yea, many reviewers on youtube said they noticed heavy spikes on ampere cards. That's why even for a 3080 it's recommended to have double the PSU power, to catch those spikes and not experience crashes.
Yes, I knew about the power spikes and I have seen some graphs made with a fast power meter/oscilloscope. Evidently, and that is what my post was about, Afterburner simply logs the wattage / temperature that the driver is reporting. And in case of the wattage, it's somehow time filtered (by the driver) so these power spikes don't show. But internally, the card seems to be well aware of them, and the "power limit" flag is reported accordingly. Still, this does not explain the "thermal limit" spikes when average temperature is around 50 degrees. While, as an engineer, I can understand that power can have an instantaneous spike, temperatures have a much lower dynamic. I find it really hard to believe that a card that is running at 50 degrees actually jumps to above 83 for less than a second, and then immediately returns to 50. Physically impossible.
What would be interesting to know is do other AIB cards show the same behavior or is this an Asus thing only?
AFAIK most cards don't have a Thermal Sensor inside the GPU package but measure Temperature via Amperage flowing trough or other electrical means. That would easily explain your temperature spikes. P.S The new Ampere cards peak high as frack.... https://www.igorslab.de/en/when-the...red-and-evaluated-together-basics-practice/2/
Great explanation. This agrees with the fact that I see those temperature spikes when the clock is being aggressively ramped up (=current is pulled).
Could also be a difference between edge/junction temperatures. Edge is (traditionally) reported to the end user. In these instances it could be the junction (hot spot) temperature spiking that triggers the TL.
I had thought of the "average temperature" vs "junction temperature" (junction would indeed be faster and hotter), and that is why I posted my question here: surely the developer of Afterburner knows in detail how these cards work.
Reporting the limit at 100 ms intervals doesn't mean that that is the internal precision of the measurement or limiting; Could be just that thermal limiting took place for at least one measurement period sometime during the last 100 ms, so it's flagged.
There are no temperature spikes, only the TL limit being activated. This seem to correspond to load actually decreasing (like the start of a 2D loading screen) so there is no reason for a temperature spike.
A funny observation: I have been playing some other games, and the temperature spikes do not show up when playing very heavy stuff like Control (which has an intensive ray tracing implementation). The card runs at a rather stable 64-65 degrees, it's almost constantly power limited as it should be (230 W) since the game is power hungry, but no thermal limitation at all, not even for an instant. It really seems that those instances of TL at low loads and lo0w temperatures are just a "phantom".
Hi Folks, i registered just to comment here. i am experiencing the same as OP, temp limit spikes in afterburner on load change (rdr2 for example when loading screens come up, some benchmark runs when they finish, etc) and sometimes during regular gameplay of rdr2 card is a msi rtx 3090 suprim x and have heard of other users with rtx 3080, rtx 3060 having similar behavior this temp limit spike can be seen in Afterburner, Gpuz Thrm perfcap reason and HWinfo Performance limit thermal flag as yes. all temps (gpu, mem tjunction are well in line and below throttling threshold) for me this started with drivers newer than 456.x when i switch back the drivers it goes away, with updated it appears. edit: do you see those spikes in Gpuz as perfcap limit thrm as well blipping when a load change occurs?
Yes, in GPU-Z too. Well, at least it seems it's "normal" and not only a thing that happens on Asus boards. Thanks!
if you are up to it try the 456.x drivers , i ultimately went ahead (461.40) and not use them due to issues with rdr2
When doing some undervolting tests I have noticed phantom power limit spikes at instances when the underlying voltage-frequency curve changes, such as in this example superposition run: At the red mark, you can see a curve change happening as the temperature rises above about 53 degrees C. The frequency changes from 1590 MHz to 1620 MHz, and at the same time there's a power limit spike. Note that there is no obvious power change, and the card is clearly not power limited since the load is in the 130-140 W range. It doesn't happen all the time, as can be seen at the blue mark where the frequency jumps up to 1635 MHz due to another curve change as we go past 60 degrees C. (And, yes, since I'm limiting the card to just 750 mV the target frequency goes up when the card gets hotter, as the curves seem to flatten more than being lowered when temperature rises.)
Here's another run with both power and temp limits shown, and both spike simultaneously two times - exactly when the underlying curve changes at the red and blue markers.
Good analysis, and that's exactly what I was seeing. I haven't checked again with the 461.72 drivers, though...