Support forum for DG Tools

NvME?
If you watch the location/placement (heat from video card) they are amazingly fast

Nah, just a basic SATA one. Heck, I never even heard of NvME until you mentioned it. I'll have to check it out...to keep TheBeast happy.

https://www.newegg.com/Product/ProductL ... isNodeId=1
https://www.newegg.com/Product/Product. ... -_-Product

OK, in 1000 words or less, why would I want one of those? Are they ridiculously faster?

OK, in 1000 words or less, why would I want one of those? Are they ridiculously faster?

yes

Max Sequential Read
Up to 2500 MBps

Max Sequential Write
Up to 1500 MBps

If your motherboard supports them, or an add-on card is needed
Just watch the install location, The key (native install slot) is right beside the graphics card a lot of times, heat can be an issue

Heh, my mobo supports them.

Hey gonca. I can't make MPC-HC open AVS scripts that deliver YV12 (works fine with ConvertToRGB24 at the end). I have tried installing helix and xvid codecs but nothing works. Any ideas? It works fine on Win8.1 with helix.

New Win10 install?
32 bit or 64 bit MPC-HC?
I hate to say this but
If you are using DGDecodeNV as a source filter make sure it is working properly on your system

If using Windows Defender exclude the DGDecNV and Avisynth filders

No, DG everything works fine. All my scripts open fine in VirtualDub delivering YV12, and x64 likes them just fine too. And they all open fine in MPC-HC with ConvertToRGB24. It's Win10 64-bit and MPC-HC 64-bit. It's some kind of codec issue for YV12.

I'll place my 950 Pro order tomorrow. TheBeast is growling.

Try a portable version of 32 bit MPC-HC, works fine on my system
Also, try excluding the present MPC-HC folder from Defender

Do you have a link? I found two. One installed crap on my system and one did not work.

MPC-HC runs fine, it's not Defender. It just can't open YV12 stuff.

No dice, no joy.

Let it go, it's not critical for me.

One more thing
MPC-HC uses LAV by default now
Don't know if LAV Video Decoder supports Helix or Xvid
Might have to set up external filters for those formats

PS
The NvME ssd are great at I/O intensive apps but they might only shave a couple of seconds off your boot time

gonca wrote: Don't know if LAV Video Decoder supports Helix or Xvid

What i can say : With LAV filters, i can't open Xvid avi file with VirtualDub, it says there is no codec. With ffdshow (i know dev is stoped since a few years) i can open Xvid video file with VirtualDub.
But, with LAV filter i can play Xvid avi file with MPC-HC if i remember properly...
As i need the possibility to open files with VirtualDub, i'm still using ffdshow for now. I'm not sure if i can install both ffdshow and LAV filters, to have LAV filter handle things in priority, and leave the others unsupported to ffdshow.

You can install LAV after ffdshow and use this tool to set your preferred filters
https://www.videohelp.com/software/Pref ... er-Tweaker

I got MPC-HC working with my YV12 scripts. The solution was to downgrade versions from 1.7.11 to 1.7.10. I'll check the LAV development thread to see if the author knows anything about this.

How did I discover this? Simple, the version working fine on my Win8.1 machine was 1.7.10, so it was a no-brainer to try downgrading.

Hehe, I just got a 10% performance improvement for all the CUDA filters. The point is that you do not need to synchronize after a kernel launch if it is followed by a CUDA memcopy, because stream 0 (which I use) implicitly serializes operations. So instead of:

cuLaunchKernel();
cuCtxSynchronize();
cuMemcpyDtoH();

we can just do:

cuLaunchKernel();
cuMemcpyDtoH();

eliminating the overhead of cuCtxSynchronize(). If you had a little CPU work to do before you need the results (such as setting up for the next copy-up/launch/copy-down cycle), you can run it concurrently with the kernel this way, because launches are asynchronous:

cuLaunchKernel();
SomeCPUWork();
cuMemcpyDtoH();

Now I am looking into running the kernels on the Y, U, and V planes in parallel (https://devblogs.nvidia.com/parallelfor ... ncurrency/). Also, using a kernel to do my NV12-to-YV12 conversion instead of CPU.

Sounds like you are having fun.

Indeed. But I am also discovering some horrifying things. For example, the WDDM driver model (Vista and Win7+) reduces performance of CUDA by a lot, and I mean a LOT. The WinXP diehards really have a strong argument here, at least if they use CUDA apps extensively. An alternative is to use a Tesla card as a secondary adapter and run it in TCC mode. And there is always Linux if you don't need frame serving or Linux + Vapoursynth if you need frame serving.

http://stackoverflow.com/questions/1994 ... in-windows

Maybe this is an option:

http://www.elitewarehouse.com/supermicr ... aQod37QEhg

After investigating my last two optimization ideas (concurrent kernels and kernel for YV12 conversion, I think I will look into NVENC stuff.

Maybe this is an option:

Oh yeah, I ordered two

On another note, (if you happen to know) how does your software run under Linux with WINE or other emulator type packages?

gonca wrote:

Maybe this is an option:

Oh yeah, I ordered two

At $109,852.50 ea ...please adopt me

Guys, you should have ordered the 40TFlops version!

I don't know about Wine, but I do know that you can run it in a virtual machine if you know how to get around the licensing. By that I mean you have to force the VM to supply a motherboard serial number other than 0 and have a corresponding key. No, I will not tell you how, unless you are my friend and buy me a drink (it's merlot these days, laying off the beer and whiskey).

You'd have to virtualize WinXP to avoid WDDM.

At $109,852.50 ea ...please adopt me

After paying for two of those I can't afford to adopt a gerbil

No, I will not tell you how, unless you are my friend and buy me a drink

The way Windows is going I might just buy you a case of merlot

Here's hoping for native Linux versions once Windows 11 proves to be so full of spyware that nobody will use it, thus forcing us all to look for a different OS.

@hydra3333

If you can make my cat shut up, I will adopt you. No, you are not allowed to kill her.

@gonca

Do you mean "hamster"?

@Aleron

After I found out about WDDM limitations, native Linux support is on the table. What is it, VDPAU for CUVID? I assume CUDA is straightforward. I have two extra machines in the shed, so I could bring up Linux. I used to work on Unix at Bell Labs back in the day, and I even wrote a few Linux drivers at STMicroelectronics. Linux would imply Vapoursynth for frame serving, I suppose. I could dust off my native Vapoursynth support.

What Linux distribution do y'all recommend?