257.15 Beta: DPC Latency issues (like ever)

Experience with 9600M GT on (bootcamped) Macbook Pro 17" running Windows 7 64-bit (same as 32-bit, same as Vista):

1 a) Even the latest 257 Beta driver still does not solve the problem of high DPC latencies in 2D (low power) mode. DPC spikes reach well over 1500 us, which in turn can lead to audio-dropout (especially for professional audio users). Neither Performance nor Low-Power 3D modes show these problems.

The same was happening on Vista (while on XP it always ran in "Performance 3D" mode which isn't that much better a solution). I reported this via NVidia temporary "Feedback" form more than a year ago. :bang:

1 b) No more DPC spikes are happening when performance modes are switched (clock-rates changed). Maybe that was already fixed with an older version already though.

2) Like the current WHQL driver with the 257 Beta the GPU is *always* running in Performance 3D mode (maximum clock-rates, heat and power-consumption) when an external display is attached. This did not happen with older version drivers. :3eyes:

 

I thought we were done hearing about DPC. I'm sorry but, what a joke. What a pointless thing to obsess and worry over.

 

Sorry, but you obviously don't know what you are talking about. With NVidia's DPCs taking as longer 1.5 ms you can forget about using small audio buffer sizes without audio dropouts.

And since audio applications usually don't stress the GPU strong enough to make it permanently switch to 3D mode it is a very real and serious issue! And with the extreme spikes happening whenever clock-rates where switched you even couldn't use large audio buffers to be save. At least the later problem seems to be fixed now (took at least 1.5 years :banana.

The only solution upto date is to use Rivatuner to enforce a permanent 3D mode (usually Low Power 3D mode as a compromise). But obviously this is a rather weak workaround since you have to reboot everytime you either need more performance or less heat/power-consumption.

It's simply very bad coding from NVidia's side. As far as I remember according to Microsoft's tech documents no DPC should occupy the CPU longer than around 120 us. My personal experience is that anything below 300 us is good enough.

 

Somethings wrong with your laptop gpu then because my gpu downclocks out of 3d mode all the time.

 

Concerning the downclock I have a suspicion that it is connected to HDPC. It *never* downclocks when my projector is connected via HDMI, but I *sometimes* got it to downclock (like after a fresh driver installation) with my 23" monitor via DVI.

It always downclocks when using the single inbuild display (nothing connected externally).

 

As soon as you will have 2 monitors connected card will NEVER go out of 3d preformance mode (also in case of multigpu card, gpu that handles monitors wont downclock, otherone will)... its intentional and also in some release notes i read from nvidia few months ago. About DPI latency, i agree when i was doing tests about dpc latency nvidia drivers are FAR infront of other any other drivers. I see numbers like nvidia having like 75% of total DPC latency and things like that... i was also geting spikes up to 1500 during gameplay and everything under stress, but since i dont have any dropouts or slowdowns i dont care.

 

SoLoR read my mind.
Be honest with me now, how long have you actually considered worrying over DPC latency with Nvidia display drivers?

 

I don't understand the question. For as long as I am using a 9600M GT on Vista/W7 I have to worry about DPC latencies. And on XP I would have to worry about heat/power-draw.

This laptop isn't just for gaming and office, it's for making music. So everytime I want to do that I have to enforce a fixed power-mode via Rivatuner and reboot. :micro:

 

Well first off, your using a macbook.....
Second, try using the latest full bootcamp package 3.1.0.1, I recall hearing how the bootcamp runtime files from previous versions had dpc issues.

 

Excuse me, but what is the argument? I know about many problems with the Macbook with and without Windows, but none of these should in any way affect the NVidia driver's 2D DPC behavior.

Thanks, but unfortunately I'm already using these and the KBDMGR.EXE related DPC (and other) problems are a different case (the v3 version BOOTCAMP.EXE fixes these problems.

This is a definitive NVidia driver problem that can be worked around by using any non 2D performance mode (the Microsoft Standard VGA driver also fixed the clock-rates to non 2D ones and thus also fixes the problem).

 

Have you tried using the nvidia laptop drivers? Or just the desktop display drivers/

 

Why are you using 257.15 Beta drivers for your laptop? That driver is for Nvidia 4xx cards.

 

hm... does anyone actualy read release notes this day? you have 2 versions of 257.15 drivers, one are named geforce/ion meant for 6xxx - 4xx series. Other ones are 257.15 Verde release meant for notebook series... Im not sure who gives people idea that 256+ series are fermi only...

 

Right, like for example the 195.XX drivers came out when the 200 series was the de facto, but people still used it on the 9k and 8k series' for their improvements.

Also, Eini, I guess what I'm getting at is, are you really experiencing losses in audio/video, or are you simply complaining because you see the DPC program tell you that it is high?

 

It's a real issues with real and experienced drop-outs (not only by me but other people as well). Long DPCs (CPU being occupied by a single non-audio driver) always lead to dropout problems with professional audio usage (small audio buffers). NVidia and Apple know about the issue since at least december 2008 when I reported it to them.

http://www.rme-audio.de/forum/viewtopic.php?id=4320

 

No problems what so ever:


I doubt that its NVIDIA's fault.

 

You do recognize that you are using a different card, aren't you? And are you even sure that your card is dynamically changing clock-rates at all (you can use Rivatuner to check that for most Nvidia cards)?

The lengthy DPCs mostly happen when there is some load on the GPU (like scrolling some webpage). Like I wrote before, doesn't happen in any 3D mode, only 2D (and formerly happened badly whenever a mode/clock was switched, but that seems fixed now).

 

I know why it would make for concern, but if you don't have any problem with audio and video playback and streaming then ignore what DPC you have.

 

I do have problems with audio and so have many others! I wrote that before.

At 64 samples audio buffer the audio-driver has to stream audio-data every 1.45 ms, at 128 samples audio it's every 2.9 ms, 256 samples every 5.8 ms.

With Nvidia driver DPC latencies going towards 2 ms and happening at random times it's easy to understand that working without audio drop-outs is not possible.

 

Well, 256+ drivers are coded for Fermi. Works like charm and improved raw performance on 4xx cards.
In addition Nvidia modified ini file adding all cards, but with less performance improvements and some bugs.

But if you think otherwise I'm OK with that. Cheers.