PDA

View Full Version : TANE interface with the NVidia etc drivers



PEV
October 7th, 2015, 10:05 PM
Thanks for including me here. Hopefully I will have something to offer.

A question for starters, What is happening to make the TANE interface with the NVidia etc drivers more robust?. Despite much fiddling on my part my system still fails with driver not responding.

I am not at home at present so I can't give you the exact specs for my system.

WindWalkr
October 7th, 2015, 10:21 PM
A question for starters, What is happening to make the TANE interface with the NVidia etc drivers more robust?. Despite much fiddling on my part my system still fails with driver not responding.

Please post any separate topics into their own threads.

thx,

chris

oknotsen
October 8th, 2015, 01:20 AM
Please post any separate topics into their own threads./Moderator mode
Moved it, so done.

WindWalkr
October 8th, 2015, 03:25 AM
What is happening to make the TANE interface with the NVidia etc drivers more robust?. Despite much fiddling on my part my system still fails with driver not responding.

For starters, try the internal test build available to trainzdev members (see the google docs) to see if that has any impact. Beyond that, there are three real possibilities:

* Some driver issue specific to your card's chipset. We're collecting data on what chipsets are having difficulties to see if anything stands out. If we narrow it down to a reliable repro on a particular chipset, then we can hopefully "just" purchase that card, repro it with different test code until we narrow down what the driver is having issues with, and then work around it. This is time consuming but usually possible. You may be able to work around the issue locally by using different drivers.

* A hardware fault on your specific card (or elsewhere in your system.) Unfortunately a relatively common occurrence, and nothing we can do to help if it turns out to be this.

* A bug in our code that will affect all GPUs for a given usage.

The only way we can really tell the difference is to collect enough data to see if there any patterns. Since most nVidia cards don't have problems, we can likely rule out the third one. If we can narrow it down to a specific chipset then we can review that possibility. Otherwise, if just a small number of people scattered at random across the chipsets have problems, we have to assume some kind of hardware fault on those machines.

chris

PEV
October 8th, 2015, 05:26 AM
Thanx Chris...

I have been back-loading earlier versions of the drivers with some success. but the problem is not beaten yet.

As you say it may be a hardware fault.. we'll see...

andi06
October 8th, 2015, 05:38 AM
My experience has been one of regular crashes up to and including build 78571, but the HF2 build hasn't crashed yet. This might correspond to an issue specific to my card.

Re the Dev builds: 78602 is virtually unusable since you can't access the surveyor lists. 78571 was generally OK.

WindWalkr
October 8th, 2015, 06:44 AM
Andi: just to be sure that we're on the same page, PEV is talking about GPU driver failures specifically, not just any old program crash.

chris

andi06
October 8th, 2015, 07:48 AM
So am I. ---------------

WindWalkr
October 8th, 2015, 08:12 AM
So am I. ---------------

Which GPU are you on, again?

chris

andi06
October 8th, 2015, 08:27 AM
NVIDIA GeForce GTX660, driver is 355.82
Prior to 78602 I was getting CTDs with no Crashdump and blue screens about every half an hour.
I haven't used 78602 much except for CM and Preview, but I haven't had a crash to desktop (for any reason) since HF2.

JCitron
October 8th, 2015, 10:39 AM
I'm running an EVGA GTX780Ti and have had some crashes...

I did notice that there's a correlation with their Precision-X utility and crashing so I stopped using it. Precision-X lets the user adjust the fan speed and power for overclocking, though I did nothing more than adjust the fan speed upwards to keep the card cool. I will continue to monitor this (pun intended) and report here. These utilities grab interrupts and make other underlying tweaks to the drivers so who knows what they are affecting. In most cases there's little problems. TS12's stuttering got worse with Precision-X running and perhaps the utility is affecting T:ANE.

In addition to the video card utilities, some motherboard companies supply an all-in-one utility such as ASUS. This motherboard utility also includes support for an ASUS video card with this utility allowing the same kinds of tweaks as the separate EVGA Precision-X. I was clued on to this by SuperFudd and his crashing, which he reported in the T:ANE Bugs thread. Once he stopped using his utility, he had no more crashes.

So I would check here first and if there is nothing else loaded except for the driver then I'd move on to the more complex stuff of the GPU and video card its self. :)

John

ianwoodmore
October 8th, 2015, 03:43 PM
I have dual GTX780ti. and they do not get overly hot. Normally 37 deg C Max 50degC during Commit.
64 bit system will use all memory it can find. I have 32GB quad channel RAM, plus pagefiles on every drive slightly larger than max physical RAM.
During testing in earlier TANE builds I found I was getting dropouts when memory commit hit a value greater than physical memory plus whatever pagefile WIN 10 had chosen.
The main cause was that the drive with TANE on it DID NOT HAVE A PAGEFILE.
Note that in earlier WIN 10 builds TANE memory commit was not releasing when TANE was closed down. I was getting over 100GB figures after a lengthy database repair or validation on ~350,000 assets. Not an issue in latest WIN 10 with normal TANE commit at 1-2GB and max 5GB.

Similarly, %CPU went to ~100% as each asset was committed even with an i7 CPU.
Some of the background processes running can be hogging CPU cycles.

Two processes in particular were notorious.

WmiPrvSE.exe which in earlier WIN 10 builds used to chew up to 35%CPU continuously. Restarting this process would drop it down to 2-3% but if you restarted WIN10, or shutdown and reloaded WIN10, it was back again.
LCore.exe which is Logitech Gaming Software used about 9%CPU instead of a more normal 1-2%.

Neither of these causes problems in WIN10 insider build 10547.

TANE causes SearchIndexer.exe to work hard during some activities such as startup and database rebuild. This may chew up 5-6%CPU but should drop back to near zero at other times.
With CM having just searched for all faulty assets my current status is 1-5%CPU total and 16%RAM Total with TANE.exe on 1,353,484KB of commit memory and zero hard faults (no pagefile being used).

These figures relate to a top end grunty PC with Intel CPU, oodles of RAM and massive pagefile availability and the latest nVidia 358.50 graphics driver.
I run 4 monitors at all times at 1920x1080 100Hz Async and get 60-118FPS on quite a few routes and sessions. with very little jitter or stutter.
I'm currently running Trainzdev build 78602. I often have multiple TANE windows open simultaneously.

I haven't had a CTD in months, not since increasing pagefiles.

My other PC with AMD Phenom 11 CPU, a single GTS530 GPU, nVidia driver and 8GB RAM is hopeless for TS12 or TANE.

In summary, for me this works:

Latest WIN10 OS with generous pagefiles on all TANE related drives.
Latest nVidia graphics driver. DX12 for some cards but not yet for earlier.

Extract from Anandtech report on nVidia 358.50

Meanwhile on the Windows 10 front, this driver release finally resolves the issues utilizing SLI and the NVENC encoder at the same time, meaning ShadowPlay and GameStream now work with SLI setups. Support for stereo rendering under DirectX 12 with an SLI setup has also been added, making this the first time we’ve seen stereo rendering under DirectX 12 addressed. On the other hand Fermi owners hoping to find WDDM 2.0 support in this driver release will be disappointed, as a quick check finds that Fermi cards continue to use the WDDM 1.3 path under Windows 10.

Latest TANE development build, even with warts and all. I'm never contemplating going back to TS12 or other legacy Trainz builds.

i7-3930K CPU not the latest but still very effective multiple core threading.
Dual GTX780ti GPUs. With latest graphics driver may be able to test SLI.
ASROCK MOBO.
SSDs RAID0 c:\ drive

Factors affecting the above data

I've downloaded all of the DLS hosted assets, with no assets from other sources other than builtin and some DLC. Total installed 266,194.
Still await balance of ex-TS12 builtins to be uploaded to DLS.
Still await SnC and Murchison 2 DLC in TANE compatible form.

I've upversioned 55% of all assets in my TAD to TB 3.7, and a further 32% to TB 4.2. Most of the remainder are builtin, base, or payware that are locked, or assets such as aliased assets that cannot be upversioned without obsoleting them.

I'm still seeing ~12,000 missing dependencies reported by CM, but I'm suspicious that CM may be calculating differently to my understanding. For example, if I upversion an asset that is a dependency it sometimes results in MD increasing even after allowing time for validation to take effect.

I've shunted most of the problematic assets that I can't currently fix into Disabled mode (~1,000). Will review these later.

I've repaired most of the defective assets to TB 3.7 validation compliance.
I still have a total of ~3,500 Faulty to solve, but only 64 out of 86,000 at TB 4.2 so far.

There are 109,000 Warnings (mainly less than TB 3.8 warnings), and some/most of those need LOD solutions or resolution of the >500 poly lowest mesh issue.


Overall, I'm pleased with progress. Like other Trainzers I wish we could solve all bugs instantaneously but that ain't going to happen. I hope the above results of hundreds of hours of research and trials can help others to dramatically reduce those mysterious CTDs.

PEV
October 17th, 2015, 07:24 PM
Further to NVidia driver failures, I suspect that I have a fault in my graphics card as it has crashed with a couple of other programs in the last day or so.

I shall buy another card tomorrow and let you know the result.

Thanks for the help and encouragement on this.

rumour3
October 18th, 2015, 02:34 AM
I updated my NVIDIA drivers (GTX860M) to the latest ones a few days ago and my system became horribly unstable- T:ANE would freeze the system, but other fairly simple programs were also doing the same thing, requiring a four second press on the power button to shut down. In the end, I removed both the drivers and the Geforce experience utility and let Win10 reinstall what it thought was the best driver- 354.04, and all is now well again. T:ANE running smoothly here.

R3

andi06
October 18th, 2015, 02:52 AM
I had a similar experience with frequent blue screens and was on the point of changing my card (GTX 660) Then magically one day everything suddenly became stable after one of the earlier dev builds. Haven't had a video card crash since and my current drivers (355.82 on Win 10) seem fine.

JCitron
October 18th, 2015, 11:43 AM
I too have had good luck with the latest driver. The BSOD crashes were due to changes in the DirectX and Direct3d implementation in Windows 10. There was a bit of a storm about this on the Microsoft TechNet and Insider forums regarding this issue.

PEV
October 31st, 2015, 08:38 PM
I'm pleased to report that the replacement and upgraded graphics card seems to have fixed all my problems. Also video quality is significantly better.

Thanks for your suggestions on this.