Fermi Cards - Powersavings Driver Issue (GTX460 only or all Fermis)

The 259.47 improved the DPC latency. It is mainly below 100us, but sometimes it jumps to ~1000us and once so far it jumped to 26000us. However, there is no audio crackling at all and no more jerky windows when they are being moved.

BTW, there are still slowdowns in MPC-HC with HD video, using DXVA. Can somebody check that?

 

Not here, MPC-HC, 1080 mkv, running foobar with asio at the same time and other stuff, latancy stays below 150us

 

I have just installed my Gainward GLH Gtx 460 in XP, with the Geforce 258.96 WinXp Whql drivers.

Before (with a gtx 260 and ?? drivers) I had no DPC latency spikes, now I have 8000ms spikes about 1/20 second apart.

Mobo = P35DQ6 Gigabyte..
Maybe the sub-P45 problem exists for real, wonder when, if ever, it'll be fixed?

 

My God, I was so upset today (I didn't change my sig yet ) when I saw this:
EVGA GTX460 1GB SC EE with 258.96 WHQL

View attachment 4556

but with 259.47 WHQL

View attachment 4557

Me happy again.

Thanks BetA13 for your posts. :thumbup:

 

I have the same problem and still spikes with the latest (259.47) drivers.. extremely annoying and no answers from nvidia or in my case gigabyte.

 

Hi all what is a good DPC Latency range to be in, i don't have a Fermi but this thread did make me check my Latency and i am nearly always above 500 and often hitting over 1000.

 

Well if you look at DPC checkers graph scale you see <500 is green, 1000 is yellow and 2000 becomes red. No rocket science

 

No, it's not rocket science, but looking around on google, most people are getting less than 100, many even in single digits, so 500 often peaking over 1000 seems rather high to me, and may explain why sometimes i get pops coming through my speakers.

 

U want to stay below 500 without any big spikes happening. If your getting spikes consistently then you may have a driver issue. I do not recall my evga gtx 260 have any issues with DPC. I believe it was always below 500 maybe even 100.

 

Yeah the Memcopy issues could play into here easy to test that would be looking @ the transfer speeds (i really advice people to check that with CUDAZ http://cuda-z.sourceforge.net/ for Nvidia or or other DX9/10 bandwidth test for ATI) to find bottlenecks though nvidia implemented some tricks to lower the bandwidth for Flash Playing for example via the Driver.
This is still a issue and people should try to understand that it's not only important how fast the GPU is in MHZ but also how their system is tuned in Bandwidth speed between GPU/CPU memory
I saw hundreds of people in Adobes Forum that have problems playing back 1080p Video via Flash in Full Framerate (with relatively good systems) and they don't understand that Adobe can't fix this
I did my own tests of this issue and it's a entire new thing how to tune your PC it changed pure FSB and CPU Mhz aren't the most important anymore (though many tweak their systems this way and maybe even lowering Memory MHz and latencies to the higher CPU Speeds, absolutely wrong thing todo these days if you want to work efficiently with a GPU) but CPU/GPU Bandwidth has become a more important factor especially when you use a OS that renders stuff on the GPU now (Vista,Win7,Mac OSX,Linux VDPAU) and more applications are moving to the GPU every month. Currently we gonna see Web Browsers Moving.

This test here is becoming more and more important for Nvidia Cards

I guess the problem was actually the Memory Latencies values where to low in the power saving C-State tough that's hard to figure out as Memory stuff on Nvidia cards is still not completely reverse Engineered. This could also fix a lot more issues now especially Flash Video Playback on those 460 Cards in low power modes. Though they introduced a complete new service for the Driver and that seems strange for a simple Latency change (deamonu.exe> Nvidia Settings Update Manager , could be though that this has a entirely different background apart from the Latency Power State issue.

Yeah Nvidias Driver team is top notch that's something where Nvidia is really ahead currently though ATI seems to improved also in those regards, and btw it seems they took my ranting sirius http://forums.nvidia.com/index.php?showtopic=165957, at least they didn't yet disabled any of the Performance Counters for older Nvidia Cards on Windows XP Drivers so everyone now can see the GPU, Memory and Video Decoder Counters not only G200 users anymore, without needing todo the overclock trick

PS: It would be really nice if someone could update these tests
http://www.youtube.com/watch?v=ToFgYylqP_U
http://www.youtube.com/watch?v=4zzKDHHevxs

with the new Nvidia Drivers on Win7 SP1

 

Damn good post Cru_N_cher thank you !

 

Im right that latency are a big issue to consider: Bus, Memory, IRQs.

are possible to know all nvidia features about power model and their P-States? Im sure that not.

i dont believe this DPC latency issue will be a bandwidth speeddown anyway. One thing is a HD Video playback (can to require a "high" bandwidth) and other very diferent is only Windows 7 Aero feature. moreove, even for video playback, nvidia have specific P states: P10 for SD and P8 for HD content (at least on fermi).




About deamonu.exe? Well, im waiting for 260.xx release. Maybe deamonu.exe will be a stranger trick or maybe will be used for other task or both. Im sure that we can discover this very soon.


About YT test, With latest 258 drivers + W7 SP1 or W7 + some newer GDI/D2D/DW patches, the results are better, but CPU are a little high yet

 

After some testing today i have figured out that the Jmicron Sata controller on my Motherboard was the cause of my High DPC Latency and not the Nvidia drivers.

 

How'd u figure that out?

I'll know for sure it's the driver for me when I uninstall, run dpc checker then install a new driver. I have already checked that my hpet is disabled by default, disabled on board sound, removed usb devices and updated my mobos bios. So far nothing has fixed my dpc latency. I do not recall having any dpc latency issues with my gtx260 in the exact same system as my gtx460 so that leads me to believe it is driver/460 related. Only thing that was updated in my system was my video card.

 

i'm another statistic on high (8000) dpc in idle mode
Video Card: Gigabyte GTX460 OC 1G
Mobo: MSI P55A-G55
OS: WinXP SP3 (32-bit)
nVidia Driver: 258.96

edit1: any updates from nvidia/gigabyte?

 

I uninstalled the Nvidia drivers and still had High DPC Latency, so i played around with some settings, and it was a setting for the JMicron controller that made the latency drop and stay well below 100.

 

cool, so what are the settings you changed and how you do it?

 

The problem must have started when i updated my motherboard bios last, the new bios for some reason would lockup at detecting drives with AHCI enabled on my JMicron Sata controller so i had to run it in IDE mode, but for the hell of it yesterday i went back to the previous bios to see if it was the bios causing the high DPC, but the old bios made no difference, but i decided to keep the old bios though as AHCI worked on that one, and sure enough switching the AHCI on the JMicron controller cured my high DPC Latency.

But it seems my case was not the Nvidia drivers after all.

 

what bios are you using? and do you have ahci turned on or off?

 

Well I had problems with dpc only since installing my new video card. Even when I had my old mobo bios with my gtx260 there was no issues with dpc latency that I can recall. I have updated my mobos bios but that did not help with dpc latency. I'll just have to wait for a new beta/whql from nvidia then i'll do some more testing.