I probably couldn't have seen that because I always keep my Windows settings at Adjust for Best Performance.

With my 660 supporting the display, and only having one fan, this was easy to test:
Core clock at 1071 (stock for this GPU)
GPU Power 91%
Temperature 60°C
GPU usage 90%
Fan speed 60%

Changed Windows to adjust for best appearance:
Dropped Fan speed to 30%

Watched GPU Power rise from 91% to 94% as the temperature rose (and power usage at the socket go up a few Watts).
After 5min temperature rose to 75°C
Power dropped back down to 91% (suggesting a clock drop)...
GPU usage remained at ~90%
Core dropped to 1058 (just one step of 13MHz)

Opened up the Calculator (scientific mode with history), and kept a few things open on the desktop, but didn't notice any lag when trailing the calculator about.

Did a GPUZ render test in a small window, but again no lag. Even in full screen the render test only showed a few lines and was reasonable for a 660.
With temps and clocks back to normal I did another full screen render, and didn't see any discernible difference.

So, the clocks do drop a bit ~70°C, but how this impacts on system responsiveness is system specific. Again, I think that the CPU is key here; the PCIE controller is on the 3rd generation i7's but not the older LGA1366 i7's.

Note that suspending these WU's is still causing a driver restart.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Keep in mind that the lag that I had was very very subtle. It wasn't so much as a visual lag; it felt more like the window being moved, was heavier to move. Refreshes were just a tick slower than expected, movements weren't 100% snappy.

Like I said, it was a night/day difference to me, but could be imperceptible by most other users.

The drop in clock speed by 13 MHz / 26 MHz should result in 1% / 2% longer crunching times for each "atomic" work package and hence increase the time between screen refreshes by about this much.

However, I could imagine this resulting in larger lag if the additional crunching time pushes the overall "time per frame" just beyond the screen refresh interval, i.e. 16.7 ms at 60 Hz. Quite a corner case and I'm not sure if this would reduce screen refreshes straight from 60 Hz (assumed) to 30 Hz, or if it would "just" introduce the occasional skipped frame (micro stutter).

Or you're just so hyper sensitive that you can easily feel 1% differences here.

MrS

---

Scanning for our furry friends since Jan 2002

Interesting analysis.

I do not believe the screen is changing from 60 Hz to any lower number. However, I do believe there is a lag in the response.

If I read things correctly, then.... if there is a "1 refresh lag", it would show up as a 16.7 ms lag. And if there is a "2 refresh lag", it would show up as a 33 ms lag.

I believe it to be the case that I can and am perceptually sensitive enough to be feeling that lag. You mention "easily feel", and I wouldn't go that far. It's very very subtle, and I often cannot "easily feel" the differences when doing direct comparisons. But, over the long term (like using each for 30 minutes), I am positive I can feel a difference.

I hope that makes sense.

Not sure if we're talking about the same thing.. it wouldn't be the screen refreshing at less than 60 Hz, but that the same frame buffer would be displayed multiple times (well, twice, otherwise things would be really choppy).

On the other hand: delaying frames (i.e. increasing input lag) would make the cursor and window feel "heavy", but wouldn't make the display of quick movements any less smooth.

My current display has a fair share of input lag (otherwise it's brilliant :) and when I used it together with a wireless mouse I could also feel the lag of the cursor, things always felt heavy. In this scenario I would likely have felt any additional lag badly. Now that I'm using a cable mouse again things are nice & smooth and it would probably take significant additional lag for me too start to feel it again.

And an area where I think I'm way more sensitive than most is slight stutter in videos, especially visible when slow camera pans are performed over a landscape with lot's of detail. On regular TFTs the picture seems to make small jumps at ~25 Hz (movie frequency). So far I haven't been able to detect this with 100+ Hz screens.
Totally unrelated to the current discussion, but IMO interesting nevertheless ;)

MrS

---

Scanning for our furry friends since Jan 2002

I'd encourage you to do some temp-testing, assuming you have a Kepler GPU, using Precision-X. If the GPU temp goes above 70*C, watch the clock. If the clock decreases to below base+maxboost, let it run that way for a week (you may have been running that way already). THEN... Use Precision-X to set a fan curve that does NOT let the temp go above 70*C (I have it set to go full-fan at 68*C). Then, using that cooling-curve with a clock of base+maxboost, run that for a week, to see if it makes a difference in the feel for you.

It made a difference for me.

The default fan profile already keeps my GTX660Ti below 70°C. Anyway, I just reduced it manually and let it reach 71 - 72°C. No downclocking (1.228 GHz, 1.175 V, max turbo +50 MHz). Yet.. when I moved the GPU-Z windows around quickly it did seem better at the lower temperature.

I think a double-blind test and some statistic would be needed here.

MrS

---

Scanning for our furry friends since Jan 2002

Do you have a Kepler GPU? If so, try letting it get up to 75*C, I bet it surely downclocks by then. Note: This is completely safe.

Do you have a Kepler GPU? If so, try letting it get up to 75*C, I bet it surely downclocks by then. Note: This is completely safe.

Hi Jacob. I have 3 Kepler GPUs (650 Ti) but to get them to 75C I'd have to put the computer in the oven :-)
Right now they're running at low fan speeds with temps of 40C, 42C & 44C (91-93% usage).

Wow.
It's quite easy for me to get any of the cards above 70*C using the standard GPU cooling profile, possibly because I'm also running 8 CPU tasks, or maybe possibly because my GPUs are pre-overclocked from eVGA, not sure. I'm surprised you can keep those GPU temps so low under load. I have to run Precision X to set a custom fan curve to keep the temps below 70*C.

Wow.
It's quite easy for me to get any of the cards above 70*C using the standard GPU cooling profile, possibly because I'm also running 8 CPU tasks, or maybe possibly because my GPUs are pre-overclocked from eVGA, not sure. I'm surprised you can keep those GPU temps so low under load. I have to run Precision X to set a custom fan curve to keep the temps below 70*C.

Jacob, that must be a world record for quick replies. I'm sending it in to Guinness. OK, maybe I'll just drink a Guinness instead. Remember my cards are the 650 Ti, not as powerful as yours. They're all in slot 1 in Antec 300 cases with either 4 or 5 fans, paired with AMD GPUs in slot 0. They're OCed (above the already factory OC) currently at +100 core, +350 memory (1084, 6010). They're all the MSI PE/OC with the large open fan/HS. I've had other MSI cards with this cooling scheme and it's impressive, unfortunately I think MSI may have just discontinued it. Unfortunately, as I was looking for more sales so as to add to the collection:

A review and commentary on the cooling scheme:

http://www.rwlabs.com/article.php?cat=&id=727&pagenumber=17

Yes, my GTX660Ti is just as much a Kepler as your's ;)
Joke aside, I just tried it again and this time it downclocked at 72°C. Yay, my GPU's not broken! BTW: I made sure to get one with an excellent cooler, it's the KFA² with factory OC.

MrS

---

Scanning for our furry friends since Jan 2002

:) I made sure to get one that's as pre-factory-overclocked as possible.... Then, after I put it into my system, I put another GPU in the slot right above it, to block the fan. THEN... I put a 3rd GPU on top of that, to block the 2nd GPU's fan.

And so the whole system, quad-core fully loaded with 8 CPU tasks... and 3 GPU's running full speed... generates a crap-load of heat :)

:) I made sure to get one that's as pre-factory-overclocked as possible.... Then, after I put it into my system, I put another GPU in the slot right above it, to block the fan. THEN... I put a 3rd GPU on top of that, to block the 2nd GPU's fan.

And so the whole system, quad-core fully loaded with 8 CPU tasks... and 3 GPU's running full speed... generates a crap-load of heat :)

Here you go:

http://www.youtube.com/watch?v=yZnhgnYA_UA