You are here

Windows 10 Graphics Card

For anything relating to music-making on Windows computers, with lots of FAQs. Moderated by Martin Walker.

Windows 10 Graphics Card

Postby BillPhillips » Wed May 15, 2019 10:37 pm

Will installing a more capable GPU reduce DirectX Execution Times when I'm mixing? I'm using a AMD R7 250 GPU and consistently seeing DirectX Execution Times >0.5 ms in LatencyMon while mixing a song with 6 audio tracks and around 6 aux tracks. At the same time, the 12 CPU threads are idling along at around 10% Load and committed memory is less than 15 GB.

When the DirectX Execution Time reaches something less than 0.5 ms, audio playback crackles even at the maximum ASIO buffer size of 1024 samples. For a new mix, DirectX Execution Time starts at <0.1 ms and there's no crackling. One sure way to push the DirectX Execution Time over 0.5 ms is to open Izotopes Tonal Balance and Insight 2 plugins which I like to have open on my second display wen I'm making adjustments. This is what leads me to suspect that, contrary to what I believed, a more capable GPU may help.

Hopefully this link will work and provide access to LatencyMon screenshots: https://1drv.ms/f/s!Ag9MrIdQMNIZnRtfb4aE7ygizukZ

System Specs:
CPU: Intel i7-6850K @ 3.60 GHz
Motherboard: ASUS X99-DELUXE II
RAM: Corsair Vengenance LPX 32 GB (4x8GB) 288-Pin DDR4 SDRAM
SSD:
• System Drive: MyDigitalSSD 240GB (256GB) BP5e 80mm SATA III 6G M.2 2280 NGFF SSD
• Alt Boot Drive: MyDigitalSSD 240GB (256GB) BP5e 80mm SATA III 6G M.2 2280 NGFF SSD
• Projects Drive: Samsung SSD 850 EVO 1TB
• Resource Drive: Samsung SSD 850 EVO 250 GB
GPU: ASUS R7250-2GD5 (AMD Radeon R7 250) (Dual (1) HDMI and (1) VGA displays)
(GPU installed in PCIEX16-1 slot with 16 dedicated PCIe lanes)
PSU: Corsair RM1000i
Chassis: Fractal Define R5
OS: Windows 10 Pro (x64) v1809
FireWire: SIIG DP FireWire 2-Port PCIe (not in use right now)
Thunderbolt: ASUS ThunderboltEX 3 (not in use)
Audio Interface: Focusrite Scarlett 18i8 1st Gen with 2nd Gen drivers
DAW: Cakewalk by Bandlab
Main Mixing Plugin Suite: Izotope Music Production Suite 2
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby innerchord » Wed May 15, 2019 11:34 pm

This indicates some sort of system problem, surely? (I'd think that your FireWire issues are bound to be related, BTW.)

What is DirectX doing? I'm no programmer, but surely something's wrong?
innerchord
Poster
Posts: 31
Joined: Wed May 11, 2016 12:00 am
 

Re: Windows 10 Graphics Card

Postby BillPhillips » Thu May 16, 2019 12:32 am

innerchord wrote: (I'd think that your FireWire issues are bound to be related, BTW.)

I came to the same conclusion. So I removed both the firewire and Thunderbolt cards and Thunderbolt software. Crackling didn't reduce and DirectX execution time stayed above 0.5 ms.
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby Pete Kaine » Thu May 16, 2019 10:18 am

Firewire - Yes, most of those two ports are the older TI XIO2200 chips which have been made "end of life" a few years back and W10 support has dropped off a cliff for it of late. The triple port cards (where they have a TI controller) tend to be based around the newer XIO2213 chip that replaced it and is the current choice.

Direct X is the Windows native system for handling GPU requests. It's the API layer between the graphics driver and the core system and allows coders to leverage the system hardware easier.

It also shouldn't be that high in general use.

It's not part of the GPU, so I'm uncertain that replacing the card will have any impact, I'm not saying it won't as it could be a faulty card, after all, I just can't say that it will.

Could otherwise be software level although I'm not sure what you can do about it if you've got the latest version on, as they don't update it very often these days and Windows 10 should keep it up to date anyhow.

Still, it can't hurt to check: https://www.lifewire.com/how-to-downloa ... tx-2624489

However, 600us isn't a fail, not by a long shot. There is no reason there should be a correlation here, but I can't rule out what you've seen.

So, looking at this the other way are you only seeing this with Izotope plugin's running?

If so, well, you've got your pattern. I guess they are leveraging DirectX for their plugin display readouts. If this is only really happening with those plugs then it might be worth contacting Izotope, logging a support ticket and asking them to clarify their Direct x requirements or to specify what they are developing on.

If it turns out they are developing and testing only on Nvidia or Intel GPU setups, then they might need to run a check on some alternate hardware to further optimize the code.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby BillPhillips » Thu May 16, 2019 12:00 pm

Pete Kaine wrote: Firewire - Yes, most of those two ports are the older TI XIO2200 chips which have been made "end of life" a few years back and W10 support has dropped off a cliff for it of late. The triple port cards (where they have a TI controller) tend to be based around the newer XIO2213 chip that replaced it and is the current choice.

Thanks Pete. I'll get one of the newer three port cards and hopefully keep my old 828 mkII in the game.

Pete Kaine wrote: Direct X is the Windows native system for handling GPU requests. It's the API layer between the graphics driver and the core system and allows coders to leverage the system hardware easier.

It also shouldn't be that high in general use.

It's not part of the GPU, so I'm uncertain that replacing the card will have any impact, I'm not saying it won't as it could be a faulty card, after all, I just can't say that it will.

Could otherwise be software level although I'm not sure what you can do about it if you've got the latest version on, as they don't update it very often these days and Windows 10 should keep it up to date anyhow.

Still, it can't hurt to check: https://www.lifewire.com/how-to-downloa ... tx-2624489" target="phpbbpopup

However, 600us isn't a fail, not by a long shot. There is no reason there should be a correlation here, but I can't rule out what you've seen.

I downloaded and installed DirectX using the link you provided. I'll check for any improvements after I reboot.

Thank you for the DirectX role explanation. I'd understood that I should focus troubleshooting on drivers with execution times >500us. So it looks like my PC is working fine.

I've used msinfo32 to look for issues and don't find anything that jumps out at me. I see a few repeating errors but they seem to only occur once a day. So I've ignored them so far.

Because I have 32 GB memory most of which never gets used I've selected No Paging File thinking it may save some time and force Windows to use the installed memory. I still get page faults but less I think and more of the installed memory is being used.

I've also used msconfig Boot Tab/Advanced Options to enter a Maximum memory: of 32678 MB. The number is always 0 when I open the dialog so I'm not sure it does anything.

Pete Kaine wrote: So, looking at this the other way are you only seeing this with Izotope plugin's running?

If so, well, you've got your pattern. I guess they are leveraging DirectX for their plugin display readouts. If this is only really happening with those plugs then it might be worth contacting Izotope, logging a support ticket and asking them to clarify their Direct x requirements or to specify what they are developing on.

If it turns out they are developing and testing only on Nvidia or Intel GPU setups, then they might need to run a check on some alternate hardware to further optimize the code.

I came to the same conclusion after trying everything else. So last night I submitted a support request to iZotope asking if they used CUDA and if so, for any recommendations on GPU specs. Like everybody else, iZotope only list minimum system requirements and I couldn't find anything on CUDA in their knowledge base. Hopefully that will bare fruit.

Thanks again for the analysis and advice.

Bill
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby Pete Kaine » Thu May 16, 2019 3:29 pm

BillPhillips wrote:Because I have 32 GB memory most of which never gets used I've selected No Paging File thinking it may save some time and force Windows to use the installed memory. I still get page faults but less I think and more of the installed memory is being used.

I tend to do this, but I've learnt the hardware that some programs will force it to look for a swapfile and may complain if it doesn't find one. Not pointing the finger this time assuming you've seen the same issues both on and off, but should you find any oddities in the future then it might be worth considering.

In regards to the page faults, you can largely ignore them for audio systems and I tend to pay zero attention to that metric. Page faults are not actually faulting in the sense of a problem, but it's a measurement of how often it needs to recover data from the swap file in real-time use.

Historically, this was a major problem for video editing in any system that had limited mechanical HDD options. After all, if you had a swap file on the same drive you were pulling data from, then thrashing that drive with paging usage would cripple your ability to actually edit anything.

These days with each system having a bunch of SSD style options in there, it doesn't really tend to affect us.

BillPhillips wrote:I've also used msconfig Boot Tab/Advanced Options to enter a Maximum memory: of 32678 MB. The number is always 0 when I open the dialogue so I'm not sure it does anything.

Yeah, I don't think that has any impact under modern OS's.

BillPhillips wrote:I came to the same conclusion after trying everything else. So last night I submitted a support request to iZotope asking if they used CUDA and if so, for any recommendations on GPU specs. Like everybody else, iZotope only list minimum system requirements and I couldn't find anything on CUDA in their knowledge base. Hopefully, that will bear fruit.

CUDA is an application layer that allows coders to leverage the cores on the card for other uses and it's unique to Nvidia. As such I wouldn't expect it here.

Open CL which was championed by ATI is whilst another way of implementing some CUDA like features and it is supported by Intel/Nvidia and ATI so it does sometimes crop up but I'd be surprised to see it here.

I'm kinda expecting some degree of Open GL requirement might be possibly in play. Open GL is the original graphics extensions that SGI developed and then handed over to the wider community.

Microsoft released Direct3D a few years later and this was followed by a whole boatload of developers calling for Microsoft and Khronos Group (none profit group now managing Open CL licencing) to merge their efforts and help streamline development on the Windows platform.

Surprisingly this happened!

I know some DAW's will leverage Open GL for their timeline redrawing and VU meter redrawing. I'm also aware that some plugins may do this and I've seen some setups in the past where a GPU that is a few generations may start glitching on the timeline with the sequencer so I wouldn't rule it out for a given set of plug-ins coming from a single developer.

I've never seen that on Sonar, but honestly I've not been a major Sonar user since it was still Cakewalk (the package, not even just the developer!) so I wouldn't be aware of this even if it was a problem.

A newer card might have support for a newer Open GL build, so might fix the issue. The "R250" models date initially from 2013 and 2 builds of Open GL have come into existence in the years in between.

If Izotope is building their plugin's using one of the newer Open GL builds then that might explain everything... and it would also answer your initial question about a possible newer card.

There is also a punted replacement for Open GL called Vulkan that appeared a few years back, but that AFAIK so far is restricted to gaming. The fact that another Open GL build appeared in the wild about a year after Vulkan launched would suggest that they are going to run side by side for quite a while yet.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby Eddy Deegan » Thu May 16, 2019 3:52 pm

Pete Kaine wrote:
BillPhillips wrote:Because I have 32 GB memory most of which never gets used I've selected No Paging File thinking it may save some time and force Windows to use the installed memory. I still get page faults but less I think and more of the installed memory is being used.

Page faults are not actually faulting in the sense of a problem, but it's a measurement of how often it needs to recover data from the swap file in real-time use.

Oftentimes, absolutely. Just to add some context, there are other things that can cause page faults (there are so-called 'soft' page faults which have nothing to do with disk access):

According to my (admittedly a bit outdated but I think it still applies) copy of Windows Internals, page faults are caused by any of:

  • Accessing a page that isn’t resident in memory but is on disk in a page file or a mapped file. (Leads to: Allocate a physical page, and read the desired page from disk and into the relevant working set)
  • Accessing a page that is on the standby or modified list (Leads to: Transition the page to the relevant process, session, or system working set)
  • Accessing a demand-zero page (Leads to: Add a zero-filled page to the relevant working set)
  • Writing to a copy-on-write page (Leads to: Make a process-private (or session-private) copy of the page, and replace original in process or system working set)

Of the above only the first involves the swap file. Personally, I tend to leave memory management alone on Windows unless I have a good reason not to.
User avatar
Eddy Deegan
Frequent Poster (Level2)
Posts: 2319
Joined: Wed Sep 01, 2004 12:00 am
Location: Brighton & Hove, UK
Some of my musical works.
I had a weird time in Surrey once, but that was a drummer's fault.

Re: Windows 10 Graphics Card

Postby Pete Kaine » Thu May 16, 2019 5:08 pm

Good points. Ok, so from reading that list, would I be right in assuming the other three processes are regarding the CPU making calls from the RAM itself?

If so, I guess running memtest on the RAM might be a worthwhile exercise to rule out other potential page fault reasons?

Not that I'm suggesting there is, but for completion sake.

I also read that as the code may have just referenced the data internally in a location where CPU wasn't expecting it and the page fault is the system noting that it's had to go looking for it... that would be harder to test for if so.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby Eddy Deegan » Thu May 16, 2019 7:04 pm

Pete Kaine wrote:would I be right in assuming the other three processes are regarding the CPU making calls from the RAM itself?

Yes - they are all effectively associated with the CPU accessing a memory location combined with the status of the page that location falls within. Depending on that status, a little bit of housekeeping has to be done (the 'Leads to:' actions) and this manifests to userland as a page fault of some kind (although really, it's a bit of a misnomer as there are no real faults involved).

Pete Kaine wrote:If so, I guess running memtest on the RAM might be a worthwhile exercise to rule out other potential page fault reasons?

Not that I'm suggesting there is, but for completion sake.

It might be worth it if only because it's a relatively trivial thing to run, but faulty RAM would normally result in a PAGE_FAULT_IN_NONPAGED_AREA error, which in Windows manifests as a BSOD because the operating system has no reasonable means of determining what cause of action to take next (as the fault is physical and thus could affect any number of processes) so it'll bring everything to a stop.

If the faulty RAM is readable but unstable then you might just get a random number when reading it, and the CPU wouldn't notice or care in that case, though I suspect that's rare. The result could also be a crash of some kind of course, but basically consider it 'undefined behaviour', much like flipped bits from solar particles in non ECC RAM ;)

Pete Kaine wrote:I also read that as the code may have just referenced the data internally in a location where CPU wasn't expecting it and the page fault is the system noting that it's had to go looking for it... that would be harder to test for if so.

Hmm.. this sounds odd to me. Code references data at a specified logical address, and the CPU will simply go to the physical address that the logical address maps to and retrieve it, (possibly incurring page-swapping in the process). There is no 'expectation' on the part of the CPU as such, other than there is something to find there.

If the address is in a prohibited page (such as the low-level OS addresses) then you would normally get a segmentation fault. This will (should!) only affect the process that made the memory access request and usually manifests the "Such and such program has stopped working" dialogue box, but may also be a silent exit for non-GUI applications or even a BSOD in rare cases.

Historically, Windows hasn't been great at segmentation faults, they could leave the OS in an unstable state and that in no small part contributed to the perception that Windows was unstable but more recent versions of Windows are much better at this.

Paging only kicks in during the data retrieval if the data at the logical memory address associated with the request is located in the swap file as opposed to physical RAM. The way that the CPU maps memory means that you can have multiple processes sharing physical RAM such that each process 'sees' a different value at the same logical address, and the translation to the differing physical addresses is handled in hardware by the CPU (this is a process involved in the causes of some of the CPU vulnerabilities that've gotten everyone upset in the last year or so).

The swap file is really there to logically extend the capacity of physical memory, so I suspect the phrase "referenced the data internally in a location where CPU wasn't expecting it and the page fault is the system noting that it's had to go looking for it" is just a slightly roundabout way of describing the process of normal swapping, perhaps by someone who didn't fully understand it, though without more context it's hard to say for sure.

You are correct that with SSD storage, swapping is far less of an issue than it used to be. The difference in the time it takes a resource saturated server (ie: huge amounts of swapping) to recover with SSDs vs hard drives is astounding. I've seen some of my resource-intensive code (our product can be extremely memory hungry depending on the data its processing) take 10 minutes to run on a server with HD verses 15 seconds to do the same work on the same spec host with SSD.

I'm also quite eager to see where the emergence of NVMe and the like takes us ... if one day the distinction between storage and RAM is all but eliminated that's going to open up a whole world of possibilities that are unavailable to developers currently ... but I digress!
User avatar
Eddy Deegan
Frequent Poster (Level2)
Posts: 2319
Joined: Wed Sep 01, 2004 12:00 am
Location: Brighton & Hove, UK
Some of my musical works.
I had a weird time in Surrey once, but that was a drummer's fault.

Re: Windows 10 Graphics Card

Postby BillPhillips » Sun May 19, 2019 12:19 pm

Pete Kaine wrote:
BillPhillips wrote:Because I have 32 GB memory most of which never gets used I've selected No Paging File thinking it may save some time and force Windows to use the installed memory. I still get page faults but less I think and more of the installed memory is being used.

I tend to do this, but I've learnt the hardware that some programs will force it to look for a swapfile and may complain if it doesn't find one. Not pointing the finger this time assuming you've seen the same issues both on and off, but should you find any oddities in the future then it might be worth considering.

I'll try various combinations (no pagefile, minimal pagefile and recommended pagefile) to see if I notice any difference.

Pete Kaine wrote: In regards to the page faults, you can largely ignore them for audio systems and I tend to pay zero attention to that metric. Page faults are not actually faulting in the sense of a problem, but it's a measurement of how often it needs to recover data from the swap file in real-time use.

That's good news, though LatencyMon seems to give them some significance.



Pete Kaine wrote:
BillPhillips wrote:I've also used msconfig Boot Tab/Advanced Options to enter a Maximum memory: of 32678 MB. The number is always 0 when I open the dialogue so I'm not sure it does anything.

Yeah, I don't think that has any impact under modern OS's.

Agreed. I'm going to untick that Maximum Memory box and forget about it.

Pete Kaine wrote:
BillPhillips wrote:I came to the same conclusion after trying everything else. So last night I submitted a support request to iZotope asking if they used CUDA and if so, for any recommendations on GPU specs. Like everybody else, iZotope only list minimum system requirements and I couldn't find anything on CUDA in their knowledge base. Hopefully, that will bear fruit.

CUDA is an application layer that allows coders to leverage the cores on the card for other uses and it's unique to Nvidia. As such I wouldn't expect it here.

Open CL which was championed by ATI is whilst another way of implementing some CUDA like features and it is supported by Intel/Nvidia and ATI so it does sometimes crop up but I'd be surprised to see it here.

I'm kinda expecting some degree of Open GL requirement might be possibly in play. Open GL is the original graphics extensions that SGI developed and then handed over to the wider community.

You are correct. iZotope uses OpenGL. However, iZotope support was not forthcoming. Their first response seemed like a canned be quite and go away response advising me to make changes to make up for my PCs limited capabilities. None of iZotope's response indicated that the DAW PC specs I'd included had been read. My question was, apparently, to specific. I asked if iZotope used CUDA. The answer was no. After reading your initial response. I replied with "what about OpenGL and got an affirmative response.


Pete Kaine wrote: A newer card might have support for a newer Open GL build, so might fix the issue. The "R250" models date initially from 2013 and 2 builds of Open GL have come into existence in the years in between.

If Izotope is building their plugin's using one of the newer Open GL builds then that might explain everything... and it would also answer your initial question about a possible newer card.

Right again, I'm going to be looking for a new graphics card. My case is deep so I'm going to be looking at workstations cards as well, though I'm guessing workstation cards will be to expensive.

Thanks again Pete.
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby BillPhillips » Sun May 19, 2019 12:55 pm

Eddy Deegan wrote:
Pete Kaine wrote:would I be right in assuming the other three processes are regarding the CPU making calls from the RAM itself?

Yes - they are all effectively associated with the CPU accessing a memory location combined with the status of the page that location falls within. Depending on that status, a little bit of housekeeping has to be done (the 'Leads to:' actions) and this manifests to userland as a page fault of some kind (although really, it's a bit of a misnomer as there are no real faults involved).

Pete Kaine wrote:If so, I guess running memtest on the RAM might be a worthwhile exercise to rule out other potential page fault reasons?

Not that I'm suggesting there is, but for completion sake.

It might be worth it if only because it's a relatively trivial thing to run, but faulty RAM would normally result in a PAGE_FAULT_IN_NONPAGED_AREA error, which in Windows manifests as a BSOD because the operating system has no reasonable means of determining what cause of action to take next (as the fault is physical and thus could affect any number of processes) so it'll bring everything to a stop.

If the faulty RAM is readable but unstable then you might just get a random number when reading it, and the CPU wouldn't notice or care in that case, though I suspect that's rare. The result could also be a crash of some kind of course, but basically consider it 'undefined behaviour', much like flipped bits from solar particles in non ECC RAM ;)

I'm traveling for the next two weeks, but I'll do the memory test when I get home. This whole pagefault discussion has helped me better understand why they seem unavoidable. It looks like they are unavoidable! So I'll quit worrying about them.

Thanks to both of you.
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby Pete Kaine » Mon May 20, 2019 2:28 pm

Eddy Deegan wrote:Hmm.. this sounds odd to me. Code references data at a specified logical address, and the CPU will simply go to the physical address that the logical address maps to and retrieve it, (possibly incurring page-swapping in the process). There is no 'expectation' on the part of the CPU as such, other than there is something to find there.

I was thinking more along the lines of it looking for code in one of the system cache's but having to do some other kind of look-up because it wasn't being held locally for whatever reason.

Eddy Deegan wrote:The swap file is really there to logically extend the capacity of physical memory, so I suspect the phrase "referenced the data internally in a location where CPU wasn't expecting it and the page fault is the system noting that it's had to go looking for it" is just a slightly roundabout way of describing the process of normal swapping, perhaps by someone who didn't fully understand it, though without more context it's hard to say for sure.

That's slightly out of context without the first part of it.

"I also read that as the code may have just referenced the data internally... "

That was just me trying to paraphrase my take home from what you had stated and I should have been a bit more careful with phrasing it as a question. My ability to talk code stops rather short of branched execution handling, which I guess ultimately is what I was poking at here and it looks like a swing and miss! :)

(I've now looked it up a bit more and that appears to be specifically called a "cache miss", so very much so a swing and a miss... )

Eddy Deegan wrote:I'm also quite eager to see where the emergence of NVMe and the like takes us ... if one day the distinction between storage and RAM is all but eliminated that's going to open up a whole world of possibilities that are unavailable to developers currently ... but I digress!

That's where they are looking to move to with Optane in the mid-term, at least on the workstation side where I can see it being really beneficial to be fair. I guess the path for home users may prove to be a bit longer given the pricing so far and it's nowhere near memory speed yet, but you've got to start somewhere.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby Pete Kaine » Mon May 20, 2019 3:29 pm

BillPhillips wrote:You are correct. iZotope uses OpenGL. However, iZotope support was not forthcoming. Their first response seemed like a canned be quite and go away response advising me to make changes to make up for my PCs limited capabilities. None of iZotope's response indicated that the DAW PC specs I'd included had been read. My question was, apparently, to specific. I asked if iZotope used CUDA. The answer was no. After reading your initial response. I replied with "what about OpenGL and got an affirmative response.

Well, it's good they were able to confirm the OpenGL side as that at least confirms that it's worth a try with another GPU. Agreed that a little more depth to the response in regards to the system requirements would have made it easier for you though.

I've just Googled "Izotope +opengl" and the is a bit about it for the GUI in the Ozone 5 manual and vague references to it in a RX5 review but not a lot more. None of it explicitly states the advised revision number through. so that's the bit that they should maybe make a little more public if they do happen to have an advised revision over there.

BillPhillips wrote:Right again, I'm going to be looking for a new graphics card. My case is deep so I'm going to be looking at workstations cards as well, though I'm guessing workstation cards will be too expensive.

So, I've just been checking cards to see where 4.6 support officially arrives and I drew a blank on the EVGA/GIgabyte/ASUS sites, with all of the products showing 4.5.

So, one call to my video guy later and he's pointed me at the Khronos site (the guys who oversee the standard) for more information. He's pointed out that some degree of support is normally rolled into the drivers at a later date and those manufacturer websites may not be updated to reflect this.

Whilst chatting we dug up the group's website and this page is particular: https://www.khronos.org/conformance/ado ... cts/opengl

and this entry:

https://www.khronos.org/conformance/ado ... ission_210

It looks like they rolled 4.6 onto that platform in a later driver update, late last December apparently.

Now, I can understand that a given model might not fully be up to scratch to run all of the features found in a later driver set, but just to confirm before you go any further, the drivers that you're running currently are the very latest that you can obtain from www.ati.com ?

If they are and it's still not working, then I guess try another card. If buying new and via mail order then at least remember that you can look to flex your 14-day DSR rights if it ends up being a red herring.

If you do, don't worry about the need for anything crazy just for application use. Most entry-level cards these days can do a couple of screens at 2K resolutions each, if not higher. The sort of performance we need for Cubase or whatever is nothing compared with trying to run even the most basic of games these days.
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby CS70 » Mon May 20, 2019 5:30 pm

It's been a while since I coded with DirectX, but a situation that can arise from CPU hogging is that the CPU is so busy that simply cannot send commands to the GPU in time. This can show up as DirectX latency but with GPU utliization low.

So if your CPU is hogged for intensive processing the render thread will have little chance to run and the whole will result in poor performance.

My experience with izotope stuff (Ozone 5) is that it is very CPU intense and/or its graphics rendering model may cause this kind of issues when the machine is working hard. Don't know about the specific plugins tough.
User avatar
CS70
Frequent Poster (Level2)
Posts: 3886
Joined: Mon Nov 26, 2012 1:00 am
Location: Oslo, Norway
Silver Spoon - Check out our latest video  and the FB page

Re: Windows 10 Graphics Card

Postby innerchord » Mon May 20, 2019 7:19 pm

I'd like to thank Pete and Eddy for the excellent information provided in this thread.

I'm not a programmer or developer, but come at hardware issues from a user perspective.
Getting only six tracks on a modern machine screams software/hardware error, of course.
I've used much less powerful systems and graphics cards (and with iZotope plug-ins) with much better results. I think when I was running Ozone 5 many years ago I had projects up to 40 or so tracks. I remember a few screen draw glitches, and a sense of how graphically-intensive it was, but nothing that stopped me working.

I'd have reverted to a basic system configuration (which you'll have saved, naturally), test that and go from there.

I hope the OP gets the issue sorted out.
innerchord
Poster
Posts: 31
Joined: Wed May 11, 2016 12:00 am
 

Re: Windows 10 Graphics Card

Postby Eddy Deegan » Tue May 21, 2019 10:34 am

Pete Kaine wrote:(I've now looked it up a bit more and that appears to be specifically called a "cache miss", so very much so a swing and a miss... )

Ahh, light dawns, sorry Pete I get you now.

When the CPU is asked for data it will check the L1/2/3 caches for it of course but cache misses don't directly cause page faults. If they did then we'd probably see millions of them per second!
.
User avatar
Eddy Deegan
Frequent Poster (Level2)
Posts: 2319
Joined: Wed Sep 01, 2004 12:00 am
Location: Brighton & Hove, UK
Some of my musical works.
I had a weird time in Surrey once, but that was a drummer's fault.

Re: Windows 10 Graphics Card

Postby Pete Kaine » Wed May 22, 2019 2:50 pm

Yes, that makes perfect sense to be fair!

All good, if I hadn't been so cack-handed in my original post, we'd have got there sooner. :D
User avatar
Pete Kaine
Frequent Poster (Level2)
Posts: 3023
Joined: Thu Jul 10, 2003 12:00 am
Location: Manchester
Kit to fuel your G.A.S - https://www.scan.co.uk/shop/pro-audio

Re: Windows 10 Graphics Card

Postby BillPhillips » Wed Jun 05, 2019 1:14 am

Eddy Deegan wrote: Oftentimes, absolutely. Just to add some context, there are other things that can cause page faults (there are so-called 'soft' page faults which have nothing to do with disk access):

According to my (admittedly a bit outdated but I think it still applies) copy of Windows Internals, page faults are caused by any of:

  • Accessing a page that isn’t resident in memory but is on disk in a page file or a mapped file. (Leads to: Allocate a physical page, and read the desired page from disk and into the relevant working set)
  • Accessing a page that is on the standby or modified list (Leads to: Transition the page to the relevant process, session, or system working set)
  • Accessing a demand-zero page (Leads to: Add a zero-filled page to the relevant working set)
  • Writing to a copy-on-write page (Leads to: Make a process-private (or session-private) copy of the page, and replace original in process or system working set)

Of the above only the first involves the swap file. Personally, I tend to leave memory management alone on Windows unless I have a good reason not to.

Eddy, this explanation helps me better understand page faults. Once I get a new graphics card, I will go through a number of pagefile size options to determine if I notice any performance differences between the various options. If I don't see any, I'll do what you do.

Pete Kaine wrote: If so, I guess running memtest on the RAM might be a worthwhile exercise to rule out other potential page fault reasons?

Not that I'm suggesting there is, but for completion sake.

Pete, I'm back home now and ran the Windows Memory Diagnostic tool and no memory problems were found.

Pete Kaine wrote: Well, it's good they were able to confirm the OpenGL side as that at least confirms that it's worth a try with another GPU. Agreed that a little more depth to the response in regards to the system requirements would have made it easier for you though.

I've just Googled "Izotope +opengl" and the is a bit about it for the GUI in the Ozone 5 manual and vague references to it in a RX5 review but not a lot more. None of it explicitly states the advised revision number through. so that's the bit that they should maybe make a little more public if they do happen to have an advised revision over there.

I'd agree that iZotope should make it clear that having a more capable graphics card would improve iZotope product components if that proves to be true.

Pete Kaine wrote: So, I've just been checking cards to see where 4.6 support officially arrives and I drew a blank on the EVGA/GIgabyte/ASUS sites, with all of the products showing 4.5.

So, one call to my video guy later and he's pointed me at the Khronos site (the guys who oversee the standard) for more information. He's pointed out that some degree of support is normally rolled into the drivers at a later date and those manufacturer websites may not be updated to reflect this.

Whilst chatting we dug up the group's website and this page is particular: https://www.khronos.org/conformance/ado ... cts/opengl" target="phpbbpopup

and this entry:

https://www.khronos.org/conformance/ado ... ission_210" target="phpbbpopup

It looks like they rolled 4.6 onto that platform in a later driver update, late last December apparently.

Pete, this is all very useful to me. I used the links you provided to dig into OpenGL myself. I found and ran an OpenGL Extensions Viewer which concluded that my R7 250 supports OpenGL 4.6. See this screenshot. However, AMD tells me it supports version 4.5.

https://1drv.ms/u/s!Ag9MrIdQMNIZnVSqv0YTuJs81mZQ

Pete Kaine wrote: Now, I can understand that a given model might not fully be up to scratch to run all of the features found in a later driver set, but just to confirm before you go any further, the drivers that you're running currently are the very latest that you can obtain from http://www.ati.com" target="phpbbpopup ?

If they are and it's still not working, then I guess try another card. If buying new and via mail order then at least remember that you can look to flex your 14-day DSR rights if it ends up being a red herring.

I've been receiving excellent support from AMD for my R7 250 which I purchased over 2 years ago and whose GPU was released years before that. With their help, I have the next to the latest possible driver installed. There's a slightly newer version available, but I don't think it will solve my problem, which I suspect is the processing power of the R7 250 not the OpenGL version it supports. I believe a new card with more processing power should eliminate OpenGL support as a cause for the crackling that I'm experiencing.

Pete Kaine wrote: If you do, don't worry about the need for anything crazy just for application use. Most entry-level cards these days can do a couple of screens at 2K resolutions each, if not higher. The sort of performance we need for Cubase or whatever is nothing compared with trying to run even the most basic of games these days.

I'm leaning toward the AMD Radeon RX 570 8GB US$150 at newegg or the RX 580 8GB for US$190. This is a link to a newegg comparison of it to some similar and slightly better cards.

https://1drv.ms/b/s!Ag9MrIdQMNIZnVidePeBiZ19Ja1u.

I found some cards that support OpenGL 4.6 but I can't afford them and like you said, software updates should add version 4.6 support to the RX 570 sometime in the near future. AMD is still producing software updates for my R7 250. So I'm expecting updates to be coming for the RX 570 for some time.

CS70 wrote:It's been a while since I coded with DirectX, but a situation that can arise from CPU hogging is that the CPU is so busy that simply cannot send commands to the GPU in time. This can show up as DirectX latency but with GPU utliization low.

So if your CPU is hogged for intensive processing the render thread will have little chance to run and the whole will result in poor performance.

My experience with izotope stuff (Ozone 5) is that it is very CPU intense and/or its graphics rendering model may cause this kind of issues when the machine is working hard. Don't know about the specific plugins tough.

Thanks, the iZotope plugins seem to take full advantage of my CPUs 12 thereads. The CPU load usually runs at about 10% and doesn't increase when I open the visualization plugins. I'm hoping that a new graphics card with more OpenGL processing power will help.

innerchord wrote:I'd like to thank Pete and Eddy for the excellent information provided in this thread.

I agree 1000%. Without their help, I would have had no idea how to proceed with solving my problem.

innerchord wrote: I'm not a programmer or developer, but come at hardware issues from a user perspective.
Getting only six tracks on a modern machine screams software/hardware error, of course.
I've used much less powerful systems and graphics cards (and with iZotope plug-ins) with much better results. I think when I was running Ozone 5 many years ago I had projects up to 40 or so tracks. I remember a few screen draw glitches, and a sense of how graphically-intensive it was, but nothing that stopped me working.

I'd have reverted to a basic system configuration (which you'll have saved, naturally), test that and go from there.

I hope the OP gets the issue sorted out.

I used Ozone 5 as well. To me the newer versions of iZotope products have really upped their game in providing visualization displays that help me a lot. And I'm hoping that they are extensively using OpenGL and that updating to a card with more processing power will help. I'll definitely report back on what happens.
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby BillPhillips » Tue Jun 18, 2019 11:21 pm

New Graphics Card Update: This is the update I promised to provide once I'd installed a new graphics card. I replaced my AMD R7 250 4G graphics card with an AMD RX 580 8G card with the goal of increasing GPU processing power and hopefully reduce popping and crackling when mixing with iZotope Insight 2 and Tonal Balance displays open.

I finished installing the RX 580 this afternoon and opened up the song that I'd been working on to compare playback performance with that of the R7 250. There's a big difference. With the R7 250 I had to work at my Focusrite 18i8's maximum sample buffer size of 1024 samples, and even at 1024 samples popping and crackling made it difficult to mix.

With the RX 580, so far I can mix with the sample buffer at 64 samples with very infrequent popping and crackling usually after starting playback or looping or making a change. So the increased GPU processing power appears to make a big difference probably because iZotope uses OpenGL.

There are however some new issues:

(1) The card fills up two slots and is touching the cards on either side.

(2) I have one VGA display with a DVI-I graphics card connector which won't work with the RX 580's DVI-D connector. My bad. My solution is going to be buying a displayport-DVI adapter.

(3) According to AMD's Radeon settings the graphics card slot is operating at PCIe x16 2.0 not 3.0 which is where it should be. So I'm going to dig into that.

(4) CbB is freezing (display stops updating) when I move too quickly. This is new. CbB is very stable for me. So I'm thinking it has something to do with the new GPU. It could also be because I was bouncing from one sample buffer size to another quite a bit in testing performance.

I want to thank Pete Kaine and Eddy Deegan for their patience and technical expertise. I wouldn't have been able to troubleshoot my playback problems and isolate the cause to the limitations of my existing GPU in handling OpenGL code used by iZotope without them.
BillPhillips
Regular
Posts: 61
Joined: Thu Nov 04, 2004 1:00 am
Location: FL, USA
Slow but sloppy.

Re: Windows 10 Graphics Card

Postby Eddy Deegan » Wed Jun 19, 2019 2:30 am

Thank you for the nice feedback Bill. Be in no doubt that Pete is the expert on current PC tech here, I'm a developer who works 'at a lower level' so to speak so I can fill in some context, but I'm happy to hear you found the info useful.

Cheers!
User avatar
Eddy Deegan
Frequent Poster (Level2)
Posts: 2319
Joined: Wed Sep 01, 2004 12:00 am
Location: Brighton & Hove, UK
Some of my musical works.
I had a weird time in Surrey once, but that was a drummer's fault.

Next

Who is online

Users browsing this forum: No registered users