Overclock.net › Forums › Graphics Cards › NVIDIA › NVIDIA Drivers and Overclocking Software › Ongoing Pascal latency problems - Hotfix doesn't work for everyone.
New Posts  All Forums:Forum Nav:

Ongoing Pascal latency problems - Hotfix doesn't work for everyone. - Page 32

post #311 of 1679
Thread Starter 
Nvidia rep posted this

'I am sorry for the inconvenience. A fix will not make it in time for the next driver so it will likely be included as part of the following driver. If a fix is ready and validated much sooner I will push for a hotfix driver to get it out to everyone as soon as possible. '


So not next driver...... but 2 drivers :/, unless they can make a hotfix
post #312 of 1679
Thread Starter 
More info from nvidia


'It varies by users configurations/drivers. If you see it on your system, swapping it for another 1080 card would give you the same DPC numbers. Our drivers are within Microsoft Windows specification for DPC but we are able to reduce it which should address this issue for users. '
post #313 of 1679
Quote:
Originally Posted by Dargonplay View Post

DRAM timings >>>>>> All in this regard.

Going from a 2000MHz C7 to a 3600MHz C16 would be a HUGE downgrade, this without counting the dozens of other timings, unless you are working with bandwitch sensitive applications, AKA not the OS or 99% of games out there.

Most RAM scaling games actually see more gains from frequency than latency AFAIK. c7 is EXTREMELY tight for 2000mhz though
Quote:
Our drivers are within Microsoft Windows specification for DPC but we are able to reduce it which should address this issue for users.

Awful specifications, then.
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
post #314 of 1679
Quote:
Originally Posted by Cyro999 View Post

Most RAM scaling games actually see more gains from frequency than latency AFAIK. c7 is EXTREMELY tight for 2000mhz though
Awful specifications, then.

I know, I tried getting to 8 for 2200MHz but no matter what, it was impossible, the G.Skill Flare does 2000MHz C7 though.

Also I've seen some dram benchmarks where frequency seems to give more performance in games, but then you see that they raise frequency but timings remains the same, thus lowering the Timing to frequency ratio as they increase DRAM frequency, which explains the performance gains.

They basically use 1600MHz with timings for 2666MHz, of course raising DRAM frequencies in these circumstance will yield tremendous amounts of performance but not because of frequency per se.
post #315 of 1679
Quote:
Originally Posted by Dargonplay View Post

I know it isn't so simple, trust me, almost two weeks trying to get third and 2ndary timings as low as possible.

Going from a C7 2000MHz you'd need a 4000MHz C14 RAM to match it, hence why I said 3600MHz C16 is a titanic downgrade... Because it is!

And responsiveness won't come from frequency but timings, and I don't mean plain C7 or C8 but relative timings to frequency, if you feel that change of ram had any positive effects it's probably just placebo, the frequency gain from 1600MHz to 1866MHz is just 16% while the timing performance loss from 9 to 11 is 22%.

Rocking a 2200MHz 9-11-10-1T-230 hell yeah!

I'm gonna need to see some proof for that. Admittedly this suite of tests were performed on an AMD CPU, but you can see that tRFC, CR, and tREF (refresh interval) typically have a much bigger impact than secondary and tertiary timings.

Your example is technically correct, but it doesn't show the whole story. I'll defer to this page: http://www.xbitlabs.com/articles/memory/display/haswell-ddr3_4.html

Bunch of graphs (Click to show)



Quote:
The charts are most illustrative. Increasing the memory frequency by 266 MHz turns out to be far more effective than lowering all timings by 3 to 4 cycles. Even when it comes to real-life latency, which is heavily influenced by timings, DDR3-1867 with rather high timings of 10-10-10-29 turns out to be better than DDR3-1600 with aggressive timings of 7-7-7-21. Comparing the effective bandwidth, DDR3-1600 is always inferior to its higher-speed opponent.

Summing it up, we can see that memory timings have become a negligible factor for modern computers, so you should first look at the clock rate of your DDR3 SDRAM whereas a low CAS Latency and other such parameters have but a small effect on actual performance. The same goes for overclocking. You should first try to make your DDR3 SDRAM work at higher clock rates and only then minimize your memory timings.

Using your example, ram running 1600 CL7 would need 2500 CL11 to match it. But as you can see 1866 CL11 already beats 1600 CL7 in everything except memory latency, and even then it only trails by 2.8% (54.4 ns vs 52.9 ns).

But if you're really that concerned about timings, I'd start working on that tRFC right away. wink.gif
Edited by magnek - 7/16/16 at 1:03am
post #316 of 1679
Quote:
Originally Posted by magnek View Post

I'm gonna need to see some proof for that. Admittedly this suite of tests were performed on an AMD CPU, but you can see that tRFC, CR, and tREF (refresh interval) typically have a much bigger impact than secondary and tertiary timings.

Your example is technically correct, but it doesn't show the whole story. I'll defer to this page: http://www.xbitlabs.com/articles/memory/display/haswell-ddr3_4.html

Bunch of graphs (Click to show)



Using your example, ram running 1600 CL7 would need 2500 CL11 to match it. But as you can see 1866 CL11 already beats 1600 CL7 in everything except memory latency, and even then it only trails by 2.8% (54.4 ns vs 52.9 ns).

But if you're really that concerned about timings, I'd start working on that tRFC right away. wink.gif
Quote:
Originally Posted by magnek View Post

I'm gonna need to see some proof for that. Admittedly this suite of tests were performed on an AMD CPU, but you can see that tRFC, CR, and tREF (refresh interval) typically have a much bigger impact than secondary and tertiary timings.

Your example is technically correct, but it doesn't show the whole story. I'll defer to this page: http://www.xbitlabs.com/articles/memory/display/haswell-ddr3_4.html

Bunch of graphs (Click to show)



Using your example, ram running 1600 CL7 would need 2500 CL11 to match it. But as you can see 1866 CL11 already beats 1600 CL7 in everything except memory latency, and even then it only trails by 2.8% (54.4 ns vs 52.9 ns).

But if you're really that concerned about timings, I'd start working on that tRFC right away. wink.gif

I wont prove you wrong right now, I will in a later date, when it isn't almost 6 AM, but yes, everyone should work on their tRFC, considering I have it down to 230




I'd say that quite literally there's nothing to improve anymore, 220 wont even POST, 225 gives Memtest86 errors, and 229 fails at 95%, but thanks for the advice, I'd appreciate if you help me out on the second set of third timings though, there are like 20 different timings that I could preset with Boundaries A and B, I just don't even understand how the ratio between those two affect the third timings beyond the fact that it does.

Good night thumb.gif
Edited by Dargonplay - 7/16/16 at 3:28am
post #317 of 1679
Quote:
Originally Posted by Dargonplay View Post

Yes, if those are idle readings then you will have audio problems and worse game stuttering.

I updated the OP so people can easily understand their readings with LatencyMon.

Hi Dargonplay

Maybe update OP with the nvidia feedback form

here it is

http://surveys.nvidia.com/index.jsp?pi=6e7ea6bb4a02641fa8f07694a40f8ac6
post #318 of 1679
I got my 1070 G1 Gaming yesterday and I have no issue with latency or micro sttuter. It is running even smoother than my old 970 G1. smile.gif
Gaming
(10 items)
 
  
CPUMotherboardGraphicsGraphics
i7 4790k Gigabyte Z97X Gaming 3 Gigabyte GTX 1070 G1 Gaming Gigabyte GTX 970 G1 Gaming 
RAMCoolingOSMonitor
Kingston HyperX Fury 16gb 1866mhz Corsair H100i Windows 10 Pro Dell 27" S2716DG 1440p 144hrz G-Sync 
PowerCase
XFX 850w TS Corsair 400R 
  hide details  
Reply
Gaming
(10 items)
 
  
CPUMotherboardGraphicsGraphics
i7 4790k Gigabyte Z97X Gaming 3 Gigabyte GTX 1070 G1 Gaming Gigabyte GTX 970 G1 Gaming 
RAMCoolingOSMonitor
Kingston HyperX Fury 16gb 1866mhz Corsair H100i Windows 10 Pro Dell 27" S2716DG 1440p 144hrz G-Sync 
PowerCase
XFX 850w TS Corsair 400R 
  hide details  
Reply
post #319 of 1679
Quote:
Originally Posted by cookieboyeli View Post

So much misinformation in this thread doh.gif

Over 100uS current Interrupt to DPC latency is BAD. In fact, over 50uS is not good either. Though it is certainly not unacceptable.

This is how a system should be at idle:
LatencyMon Idle Usage!

Remember, a 1000Hz mouse relies on your system being able to process the input every 1ms, aka every 1000uS.

If you have regular levels above about 100uS or that bounce all over the place and get spikes you're going to be introducing jitter and stutters not only on your mouse movement's update rate but jitter and stutter on your on your audio and video rendering and presentation chain.

Then end result is a massive degradation in system FLUIDITY (kind of like what using vsync does but with your mouse and audio).

It will make tracking targets harder and screw up your aim. You may realistically not NOTICE it happening, but it IS still happening, and once you fix it you will then notice how it ISN'T happening. smile.gif

The same issues is described here where other factors negatively influence a mouse's polling rate, except the cause here is primarily from something else:
http://www.blurbusters.com/faq/mouse-guide/


I've made a few half-assed guides on lowering overall Interrupt (DPC/ISR) latency. Including those 500uS+ spikes from Nvidia drivers..
Some of it is in this thread. It's incomplete as hell though: http://forums.guru3d.com/showthread.php?s=05b8aa692c06608317d3a63235d17495&t=406260

You may not be able to get rid of them entirely, but you can certainly reduce their frequency significantly! biggrin.gif I may only get a spike every few minutes now.

My average idle DPC latency is 2uS

Work towards this idle usage:

(In this old picture I had frequency set to drop at idle as I was still testing).

I also don't need to disable TONS of power saving features and idle at 200watts to do this, I idle at 66w and have most power saving features enabled actually. Two that you should change for sure though are disable C7 states, and set processor minimum frequency to 100% in power options. If your system is configured correctly you will have 0.1 to 0.7w idle power increase from doing that. thumb.gif
I recommend keeping Nvidia power usage at optimized instead of max performance, in all my testing it works flawlessly at lowering idle power usage without causing problems during load like balanced does. It's as if Nvidia knows what they're doing. tongue.gif

Depending on your system and mouse's performance, your polling rate might be all over the place, varying wildly from 400-1600Hz and causing major spikes. You can test your mouse's performance with Mouse Movement Recorder in the MarkC Mouse Fix:

If anyone needs help or advice on optimizing DPC latency let me know, I spend about 30 hours a week on teamviewer helping people for fun. smile.gif

EDIT About DPC latency under load! I understand that is important too. However, if you've lowered the idle DPC latency to the absolute minimum your system can physically achieve, you have already done 95% of what you can do to lower the load DPC latency to IT'S minimum. smile.gif

Thx for the info some more useful stuff to add to my latency list thumb.gif
post #320 of 1679
Quote:
Originally Posted by mrr9 View Post

I started getting some stutters in Witcher 3 recently, and here's why (I have a 780 Lightning) :




I've disabled the Realtek HD audio driver long back since I have an external dac+amp. I removed the Killer interface thingy, so I only have the network driver. I'm playing with pagefile (On, off, auto). But nothing is helping.

On the bright side, I think I have the highest score in this thread (17,281)....I win

Great score i thought i had a good score
but hats off too you

you clearly have the lead smile.gif
New Posts  All Forums:Forum Nav:
  Return Home
Overclock.net › Forums › Graphics Cards › NVIDIA › NVIDIA Drivers and Overclocking Software › Ongoing Pascal latency problems - Hotfix doesn't work for everyone.