strctr_k's troubleshooting thread

An area for people to discuss Scope related problems, issues, etc.

Moderators: valis, garyb

User avatar
garyb
Moderator
Posts: 23248
Joined: Sun Apr 15, 2001 4:00 pm
Location: ghetto by the sea

Re: strctr_k's troubleshooting thread

Post by garyb »

where do i get my info?
from almost 10 years of SonicCore support and about 7 years of forum life before that. i get my info from clients.

there are no absolute lists of working/not working gear. this is because there are too many variables involved to tell you anything absolutely. in addition, neither SC, nor i myself have the time or resources to test EVERY bit of hardware that folks may use.

sure!!!
set up things as you please. it's yours after all.
you are the one asking questions.

personally, i don't think that you are using an nvme for any specific reason. i don't know. because i don't know, i might ask a question, or make a statement. it's not a personal attack. i certainly am not against the use of such devices. you are wrong when you say it's the same as any other drive. there are many different data pipelines, and they all behave differently.

as to Holger's email, you can always email SC, but truthfully, he won't know either. nobody knows why you have stuttering. it's not a normal problem. what else is running while this happens? what vstis are running? are vsti samples stuttering or is the audio that is inputted stuttering? i would expect stuttering to be the result of timing issues, but momentary cpu overruns would do it. again, there is nobody who can give you a specific answer without a lot of questions.

the thing that is deadest about the masterverb thread is that almost ALL newer motherboards give good results. 10 years ago, this was not the case.

*edit*
after re-reading the original post, it looks to me that there is something(an app trying to update, or?) that is eating resources at various times. if you can isolate what it is, maybe with the details page in the task manager while the problem is occurring, then you could solve the problem. thinking about it this way, i doubt it's the drive. is Windows able to update? i usually start my machine 20 minutes before using it to allow all the apps and Windows to check for updates.
strctr_k
Posts: 16
Joined: Fri Mar 20, 2020 12:26 pm

Re: strctr_k's troubleshooting thread

Post by strctr_k »

i used to troubleshoot on this forum too btw (and i burnt out), and i'm quite familiar with the masterverb thread actually, since i posted at least one setup in it. it's meant to measure pci throughput on different combinations of motherboards and scope cards by trying to load as many masterverb (back when pro wasn't included for everybody)/masterverb pros until the pci capacity limit is reached. this is a test for pci cards that doesn't make as much sense on the xite since it has its own on-board memory for reverbs and things requiring memory.

this, classically, is done into an *empty project*, but you yourself usually do the test with a mixer and physical outputs loaded and connected, which could be faulty since the mixer might have bits and bobs that occupy pci lanes or whatever. i doubt it would change anything in practice since masterverbs use a fair amount of pci lanes per unit, and it's mosty delays on synths and other things that use some (and not much,) but i can assure that the original test, still to this day, is done in an *empty project* just to make sure nothing else in the project uses said pci lanes and get more uniform results.

for example, i used to get lower count and higher ulli with a pulsar1 in the system, but much better performance without it in the system (luna2 + scope pro). well, 8 vs 12-13 ish, and ulli below 7? was mostly not happening with a pulsar1.

been doing this for 2 decades, are you sure you have been attentive though? =P (this is just a quip and not meant to emotionally attack you)

the email address i used for my older account is gone, so i can't log into it.

i'll leave it at that.

what made your post condescending? seriously. what makes a device work? overloading doesn't mean cpu overloading? slot sharing can cause clicks? wait what? i know. the way garyb framed it, it clearly meant the processor being overloaded, not the bus (since THAT was what I was wondering about with the nvme nosebleed), and you questioning this, and explaining... i'm a fully trained in dsp/embedded systems engineering, i don't like to prance around about it at all, but please don't assume i have no clue what is going on. i've tried to hint that i knew what i was doing without chest-thumping, but seriously.

i'll go a little further: i've reviewed the documentation sent with the scope7/scopesdk7 just hours ago, and according to this specific documentation, there sure as hell (sorry) isn't TOO MANY VARIABLES to consider. nvme drives certainly aren't mentioned, nor on the website. i can quote the document verbatim if you want, but meh.

my original post wasn't ambiguous: i've moved BACK into a working xw-4600 system, from an almost-but-not-quite-working core i5 system. why would i ask a question about a working system? it works. it still does! true, i didn't specify it specifically without ambiguity, but i was trying to debug the not-quite working system. you comitted no error, and your input is appreciated even if i complained it wasn't immediately useful to me!

i've recorded over 72 hours of 20hz sine waves the last 2 weeks, tweaking one thing, testing, one other thing, testing for 3 to 6 hours... do i have the necessary rigour (canada, british spelling) to figure things out you think? i've thought i was click-free at least 3 times so far over the last months, hence the fact that i might inquire into this a little forcefully and am a little annoyed. wait i must have forgotten to unplug a usb device! no i had no usb device plugged in, but usb devices being plugged in change nothing to the results: almost no clicks, but clicks.

meanwhile, the information about nvme drives in setups, impossible to find. i'm not sonic core, i don't have access to all the internal stuff and can only use the public information on the forums. i would have thought that, since garyb "saw" some "more advanced systems than mine working fine", using just the scope7 pdf doc tweaks, he would have been able to chip on wether nvme was a factor or not.

if so many systems work, it should be easy to just list them, specifically, motherboard, processor, drive, win10 verision etc, like bud nicely did, without it being a problem! just like it used to be! win10 different because it's a service? no thanks, it's still an os based on windows nt installed on a local drive. post the exact version, or post absolute proof that it doesn't matter (-> nobody tried all combinations, hence...). but yes mine is always updated. stated so. asking the exact win10 version still happens in every non-scope click-debugging thread (way over 200, on reaper, gearslutz, focusrite, nvidia, ni etc forums) i've ran into lately while researching this btw. hardly a hostile or irrelevant question, it's a fundamental troubleshooting question still!

this forum used to be all about sharing specific setups that worked well since it used to be very touchy, and after spending *months* debugging this, maybe some more trust from you and garyb that i'm not a complete and total imploding imbecile would have been appreciated. "all x-y chipsets work" just doesn't cut it. know working setups? POST THEM. END OF STORY. that way i can infer, along with everyone else, what works and doesn't, just like back in the day. factually, not ALL recent setup work, and this is *factual*, from public information!

btw the google docs is nice but doesn't list drive types etc. sorry for pointing this out again like this matters!

now as for my problem i wish to solve, i don't really know. the xw4600 works. i can live with it. i can buy replacements for a decade. i want to record, not debug. the pci setup is more than enough for my need. i reall wish it would work in the newer machine, but how much more time do i feel like investing or wasting into this?

but like, do you or sonic core want me to invest in xite? do you or garyb or sonic core want me to buy more plugins and encourage developers? should i post the wavelength plugins not in ah peter drake's archive for completeness, and/or dig out more stuff? should i waste time learning the sdk? i've donated to your paypal to keep this forum running too, was happy to do so. do you think i feel like recommending this amazing platform that nothing has equaled yet in 20 years with this sort of weird treatment? do you want me to make sure every artist and producer i discuss with never gets close to any sonic core product ever again? because it would absolutely break my heart to do so (and i have major ethical reasons against it), but i can garantee you i can play that game too if you want, and i can garantee it would be legal as commeting on potentially broken hardware in private is 100% within my rights. the thing is we'd ALL lose here.

folks, i'm not your enemy. i'll test the nvme thing myself if it'll help other people, but please don't claim ALL motherboards with pci slots work out the box with 2 tweaks, but not know wether nvme m.2 causes problem when ALL those motherboards HAVE those m.2 slots and everyone is going to want to use them, especially on "more advanced systems." i'll buy an extra ssd, clone the nvme, remove it, put my cards back into my newer system and report here if necessary, if it helps someone, just one person! just ask nicely if you or garyb don't know for sure.

i'll repeat, you did nothing wrong, didn't need to apologize, and please don't feel bad (nor garyb, but whatever dood) for this. i have never, ever seen a weird one-click-per-hour thing like this, ever, on any system, and am totally dumbfounded. there are not "many potential causes" to this as personally, i never seen anything so sparse. plus, the scope7 pdf are clear, add irq sharing, what else? my 2 other asio drivers (including the ni komplete audio 6 driver not updated since 2106) on this same system work flawlessly on recording and playback, and it's clear i wanted to know wether nvme drives could cause this type of behaviour with regards to pci-over-pcie, with maybe some bios part i hard overlooked. 100% clear question, zero ambiguity. i never asked to go through a full troubleshooting session start from scratch, start with my specific question, then i'll ask more questions with regards to troubleshooting.

i've had this pci setup running smoothly for 15 years without needing tech support from you, garyb, or anyone on more than one system/setup, so mind that i don't feel that vulnerable to the "we might help you, we might not, your choice" type of rhetorics, if it's ok, not to mention i've helped develop some of the knowledge you use to troubleshoot, if it's also ok. please don't behave as if you own or control this knowledge and that you've developed it exclusively.
User avatar
garyb
Moderator
Posts: 23248
Joined: Sun Apr 15, 2001 4:00 pm
Location: ghetto by the sea

Re: strctr_k's troubleshooting thread

Post by garyb »

there are not "bits and bobs" from the mixer that would make trouble. mixers don't really use the PCI bus, but reverbs do, hence an empty project.

"dood"? :lol:

there aren't really any magic bullets app for this, as most systems don't use resources in the way that Scope does. you are correct about this.

the m2 connection has been fine for other users.

windows 10 should always be connected to the internet.

no, it's not so easy to make the list that you want. the two people involved need to do many more important things. you are about the only guy who has said that a particular motherboard doesn't work in quite some time, which makes me say "almost ALL current motherboards work". heck, even super rare XEON scalable processors have worked for my man Bill Goldstein without issues, but that was with an XITE. since the end of socket 1155, motherboards just haven't been a problem, unless they are bad out of the box, which happens...



i'm not sure why or if you are heated, or think that i am heated.
much ado about nothing, i guess.
strctr_k
Posts: 16
Joined: Fri Mar 20, 2020 12:26 pm

Re: strctr_k's troubleshooting thread

Post by strctr_k »

garyb wrote: Thu Nov 05, 2020 4:10 pm you are wrong when you say it's the same as any other drive. there are many different data pipelines, and they all behave differently.
ok listen. i'm not wrong. this is exactly why i was worried, it's new tech. i never said it wasn't different from other drives. i said IT WAS NOT AN EXTREME TWEAK. there is no tweak involved, that's all. plug and play, install, it's rolling, no os complaints. i'm specifically worried the new tech isn't quite ironed out yet and might cause trouble! it's not sata, it's on the pcie bus, hence my worry. stop infantilizing me please!
garyb wrote: Thu Nov 05, 2020 4:10 pm as to Holger's email, you can always email SC, but truthfully, he won't know either. nobody knows why you have stuttering. it's not a normal problem. what else is running while this happens? what vstis are running? are vsti samples stuttering or is the audio that is inputted stuttering? i would expect stuttering to be the result of timing issues, but momentary cpu overruns would do it. again, there is nobody who can give you a specific answer without a lot of questions.
no vsti, nothing loaded in reaper, just 2 channels configured for recording, no usb devices other than keyboard/mouse, nothing running other than reaper and scope, windows 10 always updated, no momentary cpu overruns, no shared irqs, no conflicts, no usb wireless (i take it off for audio, on for web/updates.) the scope project has asio2 in and out modules, 2 channels, control room or modular for the sine wave, and nothing else. no ds wave modules loaded, nor configured. removed onedrive and cortana, processexplorer only lists sgmbroker, servicehost, and almost nothing else on that front, everything looks normal on that front. no cpu spikes, reaper recording maxes at 2-3%, scope uses almost none. i didn't turn off anything risky, only, specifically things. power stuff all seems ok, usb and pci always on, cpu at 100%, and in device manager, i turned off "allow system to power off this device" for the usb hub/roots when available, and the usb wireless dongle. i turned off some telemetry stuff too, but nothing marked as risky. tried with the on-board intel graphics, and the radeon 580rx. both have perfect latencymon performance (sub-300us at all times,) measured with the asio driver loaded, and reaper recording, for over 3 hours. i've only tweaked the bios thing that relate to c-steps, cpu power throttling, etc, as per the scope7 document and best practices. i can save and load different bios setups and tested a few. tested both with and without processlasso running. cpu, memory, voltage all at normal settings in the bios, i didnt touch those. no xmp for the memory.

i'm trying to think what else i did. tried reinstalling drivers, etc. but i think that's all. it's pretty similar to what i do on win7, with onedrive and cortana (and i think yourphone.exe) added, but no fancy tweaks. ctfmon.exe turned off maybe?

the stutter is in the recording from scope asio2 dest module (i also hear it, i think but one stutter per hour is hard to debug) into reaper. tried recording on both the nvme and normal magnetic spinning drive (i prefer those for recording,) both had clicks.

i had similar issues with the ni ka6 asio and virus ti asio drivers (both connected through usb,) and they took a lot of work to get working, but now they work flawlessly (4+ hours) in the same conditions as above. i did the tests with scope WITHOUT those 2 devices connected.

i tried the 3 pci cards in the xw4600, and can confirm that recording in similar situations (same reaper version, same scope version and modules configuration) is also flawless (3+ hours), so i have excluded old cards dying for the most part. i have also loaded the dsp/pci stuff as much as i could on the xw4600, and all 3 cards can load and run plugins to the max without hiccups.

now here is where the plot thickens: i also tested with all usb devices (midi and audio) connected, usb wireless plugged , fully loaded scope project, ds wave modules loaded, 8 stereo channels in reaper, playing a game with usb mouse/keyboard going nuts, a few cpu cores loaded (not all of them 100%, gta v would cause issues i think) and getting spikes, gpu loaded and active, and the behviour was identical: not less clicks, not more clicks. always pretty sparse, at random occurences, rarely more than one an hour, sometimes 2 hours clean without click, but no pattern i could identify that would let me pinpoint to something. identical behaviour! i tested the ni ka6 driver in the same conditions and it worked well, even with the game running, no clicks! but i can't even affect the behaviour of the scope systei have never seen this type of behaviour, ever, and never ran into the many causes that might cause this.

it looked like something happening at a lower OS level, like the ssd trim thing mentioned, but nothing i managed to affect or identify in any way, for good or bad, once the rest of my system worked well.

and no, it's not normal behaviour. i've never seen this, on any scope pci systems, non-scope systems, audio systems i've built for other people, other people's scope system, other people's non-scope audio system, mac audio latop (used the virus and ni ka6 on it too), pc audio laptops, linux/openbsd boxes, raspi bitperfect audio playback system, or anything else. i've never had to record sines 3+ hours in a row for 12 days to figure out if i had clicks. i've never seen a single asio buffer repeated without drop-outs, underrruns, and things happening within the first minute. i've never had to spend 2 days writing a python script because scrolling through 3 hours or recordings slowly to spot the click became unbearable. sometimes 2 in a row in 2 minutes, then 2 hours clean! this ain't i-have-cpu-spikes or i-have-irqs-conflicts type of situation. even my first k6-300 system with via chipset (!) never was this much trouble. i've been clear on this, but you chose not to believe me, which is ok i guess.

now, in my particular case, an h310 board with the normal tweaks DIDN'T yield a stutter-free system, and i'm not the only one, so i'm sorry if you felt attacked. we can do this as adults, without questioning or assuming why i have an nvme drive or not. re-read what you wrote, you clearly ridiculed me for both why-are-you-using-an-nvme-drive-it's-useless, and trying to have one system for all, which was clearly unecessary. plus implying the problems was my fault for a mistweak or an error in configuration on my part. 100% garantee i went through every possibility about 20 times, slowly testing each one, except the one i couldn't test, more than a dozen time.

if you knew what would cause this problem, you'd state the possible causes, unambiguously, not say nobody knows =P

btw the core i5 is housed in an infinitevortex rackmount case, i still have some stickers if you want some.
strctr_k
Posts: 16
Joined: Fri Mar 20, 2020 12:26 pm

Re: strctr_k's troubleshooting thread

Post by strctr_k »

garyb wrote: Thu Nov 05, 2020 4:55 pm the m2 connection has been fine for other users.

windows 10 should always be connected to the internet.

no, it's not so easy to make the list that you want. the two people involved need to do many more important things. you are about the only guy who has said that a particular motherboard doesn't work in quite some time, which makes me say "almost ALL current motherboards work". heck, even super rare XEON scalable processors have worked for my man Bill Goldstein without issues, but that was with an XITE. since the end of socket 1155, motherboards just haven't been a problem, unless they are bad out of the box, which happens...

i'm not sure why or if you are heated, or think that i am heated.
much ado about nothing, i guess.
you said, clearly, ALL motherboards, not most =P it's in the other section on this forum if you need to re-read what you wrote with your fingers!

there has been no post on the masterverb thread between november 2016 and may 2020, so... recent posts... yeah. i can link a few other threads with win10 issues on newer boards, and older also, but ok.

microsoft have stated that win10 could be unconnected from the internet permanently in canada to respect privacy laws, so i'm not sure why you say that. they say it's fine and doesn't break things. maybe it does break things and they just haven't been testing properly? have they called you for help? you in the nt kernel team son?

if the m2 connection was fine, why not say so! it's what i wanted to know. thanks! you don't have to read the rest or what i wrote since reading more than a couple (haha) lines is obviously too much to help someone who won't buy what those two guys are so busy working on to bother with their old and current stuff. i get it! hopefully bill goldstein and those 2 other artists listed on sonic core's site (apologies to them) will buy the whole production. you could even get those thousands of flawless systems' owners to write a little bio that you could add to the website so it looks like someone (besides us lowly bottom-feeding pci-using forum dwellers trying to make things work, customer since 1999 in my case) is actually interested in using the platform?

later! or not =P
User avatar
Bud Weiser
Posts: 2684
Joined: Tue Sep 14, 2010 5:29 am
Location: nowhere land

Re: Masterverb Test Thread

Post by Bud Weiser »

strctr_k wrote: Thu Nov 05, 2020 5:16 am
ah that might be it. it's pcie/m.2 nvme x4, on the core i5, and yes it's the boot drive.
Again ...

watch 18:00min - 20:50min

https://www.youtube.com/watch?v=GUsLLEkswzE

He´s talkin´ about a laptop, but the difference is pretty obvious and he´s talkin´ about difference of laptop/desktop anyway.

So,- AVOID to make a NVMe x4,- which IS a PCIe (not SATA) drive,- to be the boot/system drive in a "realtime" audio DAW system.

:)

Bud
User avatar
valis
Posts: 7306
Joined: Sun Sep 23, 2001 4:00 pm
Location: West Coast USA
Contact:

Re: strctr_k's troubleshooting thread

Post by valis »

I do 3D, realtime video and a bunch of other stuff that's still considerably more demanding than scope, and I use an admixture of software (NDI) and hardware capture/processing devices to achieve this. What scope faced 15+ years ago, i still deal with. It's never about "this works" and "this doesn't work". It's always about the particular hardware & software stack that accompanies it, along with the order things load and are done. This was the case in the 8086 era, before that with CP/M and such, and continues to be the case since even in a modern largely 'virtualized address space' world of computing today.

If you would like to engage on that subject further, look at the IRQ sharing, the PCI bus, ACPI/APIC addressing and your Scope & Xite cards thread I created, and responses you'll sometimes find from me like this one.

Also be aware again that system components clock DOWN in our modern era, and so running your system with almost no load isn't quite as innocuous as it seems. Much of the DPC latency concerns from GPU's for instance come from when they change P-states, and similarly an Intel CPU with only the lowest TDP and portions of the cache & CPU put to sleep by the Intel IME & PCU (Power Control Unit) portion of the CPU. Well this DOES cause problems, more especially in consumer chipsets and mobile chipsets which are tuned for different workloads than HEDT & Xeons. Meaning things like BIOS/EFI defaults and choice of components will differ based on the intended usage, and so queue figuring out the entire APIC/DST table stack and checking power states etc. Beyond this, well it's going to be up to the other devices that share, and something as simple (I mentioned this already) as using the Intel RST driver instead of the stock MS one *can* impact NVME performance more than you'd think. Throughput, sleep/wake latency hits (in order to read or write you have to transition these states) and how it handles the PCIe traffic will all change both with protocol (is it AHCI mode with an NVMe driver? AHCI mode with the stock MS driver? SATA mode with an NVME device that could perform faster? AHCI or SATA or RST/RAID mode with a samsung/etc specific driver? Just with an NVME drive you have now 9 states times the total number of devices & drivers on the market to 'figure this out', and that's before accounting for Bifurcation, PLX chips, whether you're on a direct to CPU set of lanes or lanes provided by the 'southbridge' chipset (there is only this chipset now, but we'll pretend it's still the south)).

Lastly, if you were to ever want to recover an old account I'm sure the staff here would be more than happy to help, provided you can give enough verification to prove your usage of the previous account. All you would have to do is use the admin contact forum or pm one of us from a newly created account.
User avatar
garyb
Moderator
Posts: 23248
Joined: Sun Apr 15, 2001 4:00 pm
Location: ghetto by the sea

Re: strctr_k's troubleshooting thread

Post by garyb »

ok, i get the frustration, but there are so many misunderstandings that i can't unravel them all. maybe this will clear a few, but maybe not.

there's nothing wrong with usb, in general. it only matters when the usb controller is sharing an irq with the card. i usually have a usb wireless connected in my studio at all times. there are other usb devices as well including keyboard/mouse, thumb drives, dongles, etc.

you are continuously combative for no reason. why? i'm not against you.

why wouldn't an m2 connection be fine, in GENERAL. in any specific application, there MIGHT be issues, but i wouldn't EXPECT them. if they happen, then i DO like to know.

if there haven't been posts for a while, that doesn't make a dead thread, not when it's a 20 year old product.

as far as i know(another qualifier), all the damned motherboards with PCI slots work since socket 1150. when people tell me otherwise, i'll gladly pass the info on. when people write in and ask support, they are told the same thing. nobody can guarantee your experiment for you. you are using very old parts, computer-wise.

people having issues doesn't mean that the motherboard is a no-go. most people figure it out. i know there are some who never do in spite of their genius, or lack thereof.

when have i ever refused to assist just because someone didn't buy something? what do you mean by insinuating that SC isn't helpful enough to old users? do you know how many v4.5 keys i've delivered for free, or that i archive v4.5, all with SC's blessing? what about the free ASB keys and firmware aechives? do you think these things happen without cost? what about the time and money spent keeping the old cards working in new operating systems? do you think that anyone is becoming rich that way? SC certainly doesn't have to archive or support Creamware products. it won't be a good thing if bad attitudes spoil the party to the point that the carpet is rolled up and doors close. i'd rather see revamped samplers and reworked wave drivers, something i EXPECT. too bad that people have to sell their homes to make things like that happen, well, as they say, "no good deed goes unpunished".

again, i just don't understand the hostility. people are only trying to help you and they are getting no advantage from it. this isn't kvm and it surely better not devolve into that. just so you know, SC is not involved with this site. i am only involved for the same reason that i am involved with SC, because i love Scope. that's it. i could give a fuck about computers. they suck! what is nice about computers is only what can be done with them. i love audio and audio tools, not computers. i love music, not computers. i'm perfectly fine with those that love computers the most, however. it takes all kinds.

please don't respond to this post. it's not necessary. it's not that important.

i really am sorry that i wasn't more helpful.

finally, i don't care what m$ says about their concern for your "privacy". if they cared, Windows would be different. what i DO know, is that daws running win10 do best when Windows can "call home" like it wants to, at least if Scope cards are running. YOU are free to do as you please. i don't make laws, and i am no God, except when using the ban hammer, because fights are not allowed here. please lets get back to the current topic. i don't think there's more that anyone can tell you at this point, but i might be wrong. maybe the solution is near. maybe it has already been suggested, as i said, i don't know. i'm sorry.

ps- if anything, you are MY son. have respect for your elders, even if they are idiots.
User avatar
Bud Weiser
Posts: 2684
Joined: Tue Sep 14, 2010 5:29 am
Location: nowhere land

Re: strctr_k's troubleshooting thread

Post by Bud Weiser »

garyb wrote: Thu Nov 05, 2020 10:49 pm ... i love Scope. that's it. i could give a fuck about computers. they suck! what is nice about computers is only what can be done with them. i love audio and audio tools, not computers. i love music, not computers.
We could (or should ... LOL) be twins ! :D
garyb wrote: Thu Nov 05, 2020 10:49 pm please don't respond to this post. it's not necessary. it's not that important.
plz forgive I did ...

:)

Bud
User avatar
garyb
Moderator
Posts: 23248
Joined: Sun Apr 15, 2001 4:00 pm
Location: ghetto by the sea

Re: strctr_k's troubleshooting thread

Post by garyb »

:D
User avatar
valis
Posts: 7306
Joined: Sun Sep 23, 2001 4:00 pm
Location: West Coast USA
Contact:

Re: strctr_k's troubleshooting thread

Post by valis »

At the risk of seeming somewhat pedantic, let's close the loop of just the NVME side of this discussion to prove a larger point for our general audience (we have about 1200-1400 *actual* visitors a month--compare that to the number who post and interact!). It's nice to have simple answers to questions....So the correct answer to the "do NVMe drives cause problems" is No. The correct answer is also Sometimes, and Yes.

Case in point: Intel Kaby Point Z370, Intel Coffee Lake-S system with an i7 8700, GTX 1080, Samsung 970 EVO Plus, everything works great. Add an RME PCIe interface into the 4x slot, and things work fine. But guess what, the NVMe slot already causes the GTX 1080 to run in 8x mode on this particular board, as the NVME shares bifurcation with the second 16x slot, and using anything connected to the second slot drops the chipset into 8x + 8x + 4x instead of 16x + 4x. Now, let's just keep the system configuration here for a moment and look at what the changing drivers and interface modes with the NVME drive alone does to things, ie cover some of those combinations I referred to earlier:

With the stock MicroSoft NVMe driver (presuming the motherboard is in AHCI mode not RAID, which uses Intel's RST or MS's RAID driver) you'll find that the Random IOPS is around 30,000 give or take, and Random Write is roughly 120,000 IOPS. Setting SATA to RAID mode with MS driver those go up about 3%, and then SATA RAID mode with Intel RST the IOPS double for Random Reads to 60K & go up a factor of 1.5X for Random Writes to 180,000 IOPS.

Now, the question is how do these gains come about? If we analyze the bus with performance monitor & other dev tools (install MS VisC or use SysInternals & Nirsoft tools etc), what we will find is that the drivers are putting themselves at a higher priority, and/or settings bits to dominate the rate at which hardware is polled etc. And certainly between the controller modes your entire DSDT table changes and thus the ACPI/APIC connection points (which at this point in the PCIe realm is all virtualized logically, though still faces limits electrically). And so the IOPS gains come at the expense of 30% or more of the PCie *TIME* slice allocated to the PCIe link using the NVME drive.... but you don't see this in most benchmarks as the biggest gains are in Random read/write IOPS simply because there's more time to finish handling them (they are *random*) in each given timeslice.

Now, if I watch the impact on realtime render benchmarks, what you can see is a 7-14% reduction in the GTX 1080's performance, and keep in mind this card has ALREADY been in 8x rather than 16x, just with the NVME card added. This isn't just a gaming issue, for my needs this is quite noticeable in something like say TouchDesigner 099 or Notch/Smode with complex projects (these are things I *do*, and they are all realtime and mix multiple streams of video, audio and 2d/3d datasets on the GPU in a way that our DSP cards do for audio). The difference is we are more forgiving to visual artifacts, and our gear is too. Audio buffer underruns will cause speakers to trash and bleed, as we all know, while a burst of digital artifacts to the eye might even be pleasing if it's on cue (ok maybe glitched audio will too in this era).

Now, let's try SATA AHCI mode for the SATA controller, which will let us use the Samsung NVME driver (third party drivers won't load in RAID/RST modes unless they are RAID/RST compatible drivers--and I've only seen that on the server/workstation SSD side of things). Ok, so I hadn't mentioned sequential Reads & Writes before, because these only went up from 30K with the MS AHCI driver to 32K with the MS or Intel RAID drivers, and then to 3538 (MB/s) with the Samsung driver. That's not much of a gain, right? If we look at Random IOPS however, you'll be quite surprised. First, let's look at straight reads/writes with the SATA mode & the Samsung NVME driver: we've now gone up to 3540 or so for Reads (MB/s) and up to 3350 for Writes which is a moderate gain of about 15% over the stock MS driver in SATA mode. However, our Random IOPS have gone up to 328,125 for Reads and 258300 for Writes using the Samsung NVME driver. If you compare above that's a factor of 10x for Read IOPS and more than 2x for Write IOPS.

Where does this additional performance magically come from? Well, we might presume the driver is just 'more compatible' somehow. And digging into the details you'll find this is somewhat true, it can help the SSD's controller and the host negotiate features that the MS driver may drop for the sake of safety & more robust compatibility/reliability.

In any case, if we watch again the timeslices given to the NVME data transfers, what we find with the Samsung driver is when we saturate the NVME link with heavy reads & writes, the GPU suffers tremendously *IF it is also trying to place enough of a strain on the system bus to saturate its PCIe link and is 'bottlenecked' starving for bandwidth AND time slices* for data shuffling around. GPU compute tasks won't care as much, but TouchDesigner or even Modern Warfare 2019 (which uses about 99% of a modern GPU and streams more textures than any of my VJ or 3D projects) certainly will, and the impact on performance for Modern Warefare is enough to drop us from 1440p to 1080p to keep the same fluidity of gameplay. Meaning our DIPS in framerates are the biggest impact--which is actually similar to Scope's needs in terms of 'bus time' to complete ASIO buffer transfers! And outside of gaming alone, I can show that Touchdesigner 099 goes from being able to handle 4 4K screens in a horizontal array (that's a lot of pixels) down to about half that, or you see framerate drops/duplications (just like our Scope cards and the ASIO buffer).

Mind you, everything *seems* to be working and by relying on benchmarking tools in isolation, we would never know the cause of the issues. By testing and evaluating components against each other, and understanding a bit of the way they share resources...well we know this story for years here in Scope land.

Now, what happens if you try to then use a card in the second 16x slot, say for another SSD or PCIe adapter for an NVME drive? In other words, we've had a system long enough to have outgrown it, as we're now going to force the second 16x slot and the NVMe m.2 slot to 'share' the 8x lane that we already know is bifurcated to feed the m.2 slot. EVEN IF that card is just 4x electrical, they are contending for bandwidth with each other. A PLX chip is another way around this (rather than bifurcation) but that swaps the lane sharing for a latency penalty as it basically 'switches' back & forth between the two devices (thus sharing the lanes like a switch, rather than separating the electrical lanes per device as with bifurcation).

Well, what happens is if we saturate that second NVMe drive, we at the very least start to see framerates dip below 50% of what they were in when TouchDesigner only had the GPU and no NVMe drive at all to contend with (we were only using SATA drives). And guess what happens if we just move the NVME drive onto the NVME PCIe card and put it in the 4x electrical slot? Well, we now have no room for our soundcard, and if we were pushing FOUR 4K screens with (even compressed) video data, we're going to find that the choice between a pro level soundcard & enough buy bandwidth to keep the HAP video streams playing without frame drops is an easy choice. There are USB soundcards that will solve this issue after all...

The real issue here is that the consumer chipsets don't have enough PCIe lanes for this sort of data transfer, so you either have to keep your system configurations within the bounds of what they are intended for (limited peripheral usage--aside from maybe USB3.x & possibly TB) *as well as* age & resource appropriate software. OR you need to step up to HEDT & Workstation (Xeon/Threadripper in modern parlance) hardware. Both of these classes of machine will have PCIe bandwidth aplenty...

Now this example is taken from a client of mine who was trying to run an Apple style presentation on a $1700 gaming PC. The convention's budget was about $230K for their staging elements alone, so it might boggle my mind that they were trying to shoehorn an off the shelf unit into such a position, but this is where I get my income from so I just pointed out the limitations of the system they had, and suggested we either put a system in place that could do all the things they wanted (HEDT build, which would have taken too much time to deploy even if one was readily available) OR drop some of what they were trying to do and reconfigure (USB soundcard, and ultimately we dropped 1 4K output and ran with 3 outputs on 1 GPU instead of using the Blackmagic ATEM unit to split out 4 screens from an 8K bonded pair--which was causing a different issue and was done to account for the lack of a second GPU which is what they ORIGINALLY spec'd, all on a consumer class gaming chipset!)
User avatar
valis
Posts: 7306
Joined: Sun Sep 23, 2001 4:00 pm
Location: West Coast USA
Contact:

Re: strctr_k's troubleshooting thread

Post by valis »

Note that this is JUST the NVME drive/driver side of the discussion, there are a lot more things that you can change in a system and therefore more variables! And as such, we have no idea what's really going on with a system we can't access and can only respond to the information provided. Also, this stands as a good example of how there's not just a 'yes' or 'no' answer when we deal with our tech toys. And it points out why MS likely coded such a 'conservative' driver for NVME cards, it keeps them from having other issues.

More to the point, gary's assertion that these are simply tools is correct. We play with our systems for our own delight and at our own peril, and if you're experiencing something with Scope that others haven't experienced it might be a good idea to 'walk the chain' of your unique setup with us in a manner that's coherent and useful to all of us.
User avatar
Peter Drake
Posts: 214
Joined: Thu Nov 08, 2018 12:27 pm
Location: Outer Milky Way

Re: strctr_k's troubleshooting thread

Post by Peter Drake »

strctr_k wrote: Thu Nov 05, 2020 4:35 pm should i post the wavelength plugins not in ah peter drake's archive for completeness, and/or dig out more stuff? should i
Please don't involve me with your drama.
Thank you,
Peter Drake
User avatar
Bud Weiser
Posts: 2684
Joined: Tue Sep 14, 2010 5:29 am
Location: nowhere land

Re: strctr_k's troubleshooting thread

Post by Bud Weiser »

valis wrote: Sat Nov 07, 2020 3:19 pm At the risk of seeming somewhat pedantic, ...
Interesting read,- thank you !

But some computer (consumer) mainboards work better than others and come w/ different feature sets.
For audio/MIDI DAW ... who needs hi end graphics really ?

It´s true M.2 NVMe PCIe x4 slots share w/ graphic card slots,- but you´re not urgently forced to use both.
As a musician, I better sacrifice some graphics/video stuff for click- and pop free audio.
I´m fine w/ iGPU now on 2 machines,-
ASROCK Z97 DAW w/ SCOPE/XITE (PCIe card in 3rd PCIe-X16 slot w/ x4 speed still) and Samsung 970EVO Plus M.2 in fastest M.2 slot (sharing w/ 1st and fastest PCIe-X16) side-by-side.

ASROCK Z390m also works pretty well w/ iGPU and Win10 office usage.
Just only by interest I installed ASIO4ALL using onboard Realtek HIgh Definition Audio and surprise,- no clicks and pops.
Can only be better w/ a pro audio/MIDI interface,- but it´s Intel core i3 8100 only,- so I´ll leave it for office and entertainment usage.

The RME card you mentioned above,-

IIRC, RME cards behave different than SCOPE cards.

RME doesn´t refuse to be configurated by OS and DAW apps and are less picky on IRQ sharing.
All internal RME PCIe cards are "1-Lane PCI Express Endpoint device (no PCI Express to PCI Bridge) revision 1.1" - a per-lane data rate of 250 MB/s and a transfer rate of 2.5 gigatransfers per second (GT/s).
Well, that´s not much and already the limitation of the card itself,- and even that sounds poor now, THESE are most excellent low latency audio cards w/ excellent driver support.

In opposite to the XITE-1 PCIe card, where you cannot change buffer size on-the-fly, I don´t see any reason to put the RME card into a PCIe x4 slot.
RME works pretty well in PCIe x1 slots.

SCOPE is different because it uploads projects/devices into XITE-1,- and THAT is faster when the PCIe card sits in one of the (blue) PCIe X16 slots WHEN these are connected directly to CPU,- and it might guarantee an exclusive IRQ,- the card just only sharing memory w/ Intel graphics.
32GB RAM in my DAW machine should be enough though.

The number of lanes per sé doesn´t seem to be the problem.
The cards,- RME and SCOPE/XITE,- are ALL single-lane cards.
So, you say,- WHEN using M.2 NVMe PCIe x4 storage,- there´s no way using single-lane PCIe cards w/o problems like bandwidth reduction in "consumer" machines and even when using iGPU only ?

I really love AsRock because they precisely explain hardware sharing details in their mainboard-manuals.

:)

Bud
User avatar
valis
Posts: 7306
Joined: Sun Sep 23, 2001 4:00 pm
Location: West Coast USA
Contact:

Re: strctr_k's troubleshooting thread

Post by valis »

Bud Weiser wrote: Sat Nov 07, 2020 8:30 pm
valis wrote: Sat Nov 07, 2020 3:19 pm At the risk of seeming somewhat pedantic, ...
Interesting read,- thank you !

But some computer (consumer) mainboards work better than others and come w/ different feature sets.
For audio/MIDI DAW ... who needs hi end graphics really ?
(Valis edit).......(/Valis edit, profit!)
And so you agree with the thesis overall, that we cannot simply state "this works' and 'this doesn't work' categorically as in 'NVME SSD's ALL WORK WITH SCOPE" or etc. It's overly simplified and must be considered in the context of the specific chipset, peripherals, motherboard implementation (you are correct not all boards implement the M.2 slot and PCIe lane sharing the same way, nor do they use the same BIOS/EFI, etc...though ultimately they all still have the same chipset limitations overall).
User avatar
Bud Weiser
Posts: 2684
Joined: Tue Sep 14, 2010 5:29 am
Location: nowhere land

Re: strctr_k's troubleshooting thread

Post by Bud Weiser »

valis wrote: Sat Nov 07, 2020 11:10 pm
And so you agree with the thesis overall, that we cannot simply state "this works' and 'this doesn't work' categorically ...
Yes,- halfway ...
valis wrote: Sat Nov 07, 2020 11:10 pm ... as in 'NVME SSD's ALL WORK WITH SCOPE" or etc. ...
I´d say, in general, they all do,- just because NVMe is a (Intel) software protocol to connect Non Volatile Memory via PCIe w/o the need of proprietary drivers.
What has that to do w/ the drive itself except the controller supports that protocol ?

Win 7 needed 2 MS files/updates (or hotfixes ?) adding NVMe support,- but Win10 should work out of the box.
How good it works and if there´s necessity to sacrifice other devices for performance optimization is matter of mainboard design IMO.

On my AsRock Z97, there´s the limitation of only ONE Ultra-M.2 slot (x4 and 32GB/sec) existing and the 2nd M.2 slot is x2 only,- but that´s history since some time (perhaps since Z170 already ?).
I recognized we get more than only 16 PCIe lanes now,- and (please correct me if I´m wrong) I believe AHCI is SATA related and not for NVMe/PCIe.
I mention because you talk about AHCI in former long post above.

I´m pretty sure, on my AsRock Z390 I´d be able to use both M.2 slots w/ NVMe SSDs,- this also for audio work,- as long as I don´t boot and run OS off of one of these drives.
valis wrote: Sat Nov 07, 2020 11:10 pm ...though ultimately they all still have the same chipset limitations overall.
True,- but also here, manufacturers do one´s own thing ...
Often, the Intel chipset offers more than manufacturers implement/realize w/ their mobo designs.

My experience is, finding the right mainboard is much more difficult than finding "the right drive".

:)

Bud
User avatar
garyb
Moderator
Posts: 23248
Joined: Sun Apr 15, 2001 4:00 pm
Location: ghetto by the sea

Re: strctr_k's troubleshooting thread

Post by garyb »

Bud Weiser wrote: Sun Nov 08, 2020 8:49 pm
My experience is, finding the right mainboard is much more difficult than finding "the right drive".
kinda true, but it depends on how you use it. i'd say that physical layout is more likely to be a problem than operation would be, especially with 14/15 dsp cards.
Post Reply