Does GPU Matter?

[Dasheesh]

^^^ russians gonna russian. microboys gonna micro. can't tell them anything.

[Dasheesh]

https://www.centerspace.net/offloading- ... o-your-gpu

[lnikj]

If you have any interest in running VCV Rack you will need a discrete GPU.

Also not true. It requires OpenGL 2 capability, which almost all integrated GPUs support nowadays. You have to go back about 10 years to find one which doesn't - or to a very obscure one.

System Requirements

Operating system: MacOS 10.7+, Windows 7+, or Linux (Ubuntu 16.04+, etc)
CPU: Intel/AMD 64-bit processor from ~2011 or later
Graphics: Dedicated Nvidia/AMD graphics card from ~2013 or later. Integrated (non-dedicated) graphics such as Intel HD/Iris are not recommended and cause significantly increased CPU usage.
RAM: 1GB
Disk space: 1GB

If you think you don't need a discrete GPU try running Rack on a laptop that hasn't got one.

[PeterP_swe]

The most an integrated GPU tends to steal is about 32mb, but they do ram sharing up to 256MB when needed.

If you're using an integrated GPU and run Window 10, do this:

1) Right click on the desktop and select "Display Settings".
2) Advanced display settings.
3) Display adapter properties for Display 1

What numbers for video memory do you see?


I'm not saying it won't work with an integrated GPU, but you're fooling yourself if you think they are as good as a dedicated one. If you're concerned about saving money it can be worth it, but only then.

Try running a heavy song session in your DAW and then open up a couple FabFilter ProQ windows. Watch the CPU meter in your DAW. Now try the same thing with a Geforce 1050 or other cheap GPU. It's going to be significantly better than even Intels latest integrated GPU.

[Obsolete317542]

xmoigtjtopx^ùpjhoittxy

[lnikj]

Apparently nVidia's OpenGL acceleration is by default multithreaded on the CPU, and if the OpenGL code is not the most robust or modern, this causes stalling of the other processes running on that CPU. This in turn creates CPU spikes, which creates audio dropouts. It took me almost two years to find out that nVidia's multithreaded OpenGL acceleration was the cause of endless CPU spikes and audio dropouts when using Fabfilter plugins! It can be disabled in their drivers but still...

That's a useful bit of information that I will mentally file away for the future. Thanks.

The complexity of modern computers ....

[metamorphosis]

If you have any interest in running VCV Rack you will need a discrete GPU.

Also not true. It requires OpenGL 2 capability, which almost all integrated GPUs support nowadays. You have to go back about 10 years to find one which doesn't - or to a very obscure one.

System Requirements

Operating system: MacOS 10.7+, Windows 7+, or Linux (Ubuntu 16.04+, etc)
CPU: Intel/AMD 64-bit processor from ~2011 or later
Graphics: Dedicated Nvidia/AMD graphics card from ~2013 or later. Integrated (non-dedicated) graphics such as Intel HD/Iris are not recommended and cause significantly increased CPU usage.
RAM: 1GB
Disk space: 1GB

If you think you don't need a discrete GPU try running Rack on a laptop that hasn't got one.

From their FAQ: "Rack requires at least OpenGL 2.0 with the GL_EXT_framebuffer_object extension. If your graphics card supports this, make sure you’ve installed the latest graphics drivers, and restart Rack."

If you're using an integrated GPU and run Window 10, do this:

1) Right click on the desktop and select "Display Settings".
2) Advanced display settings.
3) Display adapter properties for Display 1

Try looking in the UEFI instead bud.

I'm not saying it won't work with an integrated GPU, but you're fooling yourself if you think they are as good as a dedicated one. If you're concerned about saving money it can be worth it, but only then.

Try running a heavy song session in your DAW and then open up a couple FabFilter ProQ windows. Watch the CPU meter in your DAW. Now try the same thing with a Geforce 1050 or other cheap GPU. It's going to be significantly better than even Intels latest integrated GPU.

What you're saying has no basis in reality.

Quite a few plugins use OpenGL processing that is offloaded to the GPU. (Waves, Fabfilter, etc.) In my

Like I said, many applications use basic opengl/directx rendering. That's not the same as 'offloading to the GPU'. It's just using the graphics API, not using it for processing.

[metamorphosis]

https://www.centerspace.net/offloading- ... o-your-gpu

... that link has no meaning, buddy.
I've done CUDA programming. I actually know what I'm talking about...

[pekbro]

Quite a few plugins use OpenGL processing that is offloaded to the GPU. (Waves, Fabfilter, etc.) In my old PC I had an nVidia GTX 760, and it caused me a lot of headaches. Apparently nVidia's OpenGL acceleration is by default multithreaded on the CPU, and if the OpenGL code is not the most robust or modern, this causes stalling of the other processes running on that CPU. This in turn creates CPU spikes, which creates audio dropouts. It took me almost two years to find out that nVidia's multithreaded OpenGL acceleration was the cause of endless CPU spikes and audio dropouts when using Fabfilter plugins! It can be disabled in their drivers but still... If I were to build a new audio workstation these days, I would strongly consider AMD graphics, or just using iGPU.

Also, weren't there some reports that state nVidia graphics create more DPC latency? Something to look into before buying!

I have 2 GTX770s, never really had a problem and I have like 1500 plugins, 12 DAW's. Just sayin...

[PeterP_swe]

If you're using an integrated GPU and run Window 10, do this:

1) Right click on the desktop and select "Display Settings".
2) Advanced display settings.
3) Display adapter properties for Display 1

Try looking in the UEFI instead bud.

That's for the framebuffer only. Once you boot an operating system the driver will need memory to be able to run OpenGL/DirectX, so it takes and reserves that from system memory.

It'll usually reserve 1gb VRAM per 8gb system RAM. It can grow a bit dynamically but it will always start out with one fixed buffer.

What you're saying has no basis in reality.

Huh? What makes you think that?

[ThomasHelzle]

Quite a few plugins use OpenGL processing that is offloaded to the GPU. (Waves, Fabfilter, etc.) In my old PC I had an nVidia GTX 760, and it caused me a lot of headaches. Apparently nVidia's OpenGL acceleration is by default multithreaded on the CPU, and if the OpenGL code is not the most robust or modern, this causes stalling of the other processes running on that CPU. This in turn creates CPU spikes, which creates audio dropouts. It took me almost two years to find out that nVidia's multithreaded OpenGL acceleration was the cause of endless CPU spikes and audio dropouts when using Fabfilter plugins! It can be disabled in their drivers but still... If I were to build a new audio workstation these days, I would strongly consider AMD graphics, or just using iGPU.

Also, weren't there some reports that state nVidia graphics create more DPC latency? Something to look into before buying!

I have 2 GTX770s, never really had a problem and I have like 1500 plugins, 12 DAW's. Just sayin...

Using Nvidia only for ages (for 3D, AMD isn't the hottest contender) and never had such a problem either. Running a RTX 2080 TI and a GTX 1080 TI ATM in my main machine and couldn't be happier.

As for the topic: In the long run, I would assume the use of GPUs continue rising across the board. In 3D it's obviously the most important, I render all my graphics on these cards nowadays and it's amazingly fast.
In Video and Encoding/Decoding it's also extremely important, encoding a h265 video on the CPU is no fun, while on my GPUs it flies (Vegas, Resolve).

While in Audio it's still mostly limited to GUIs etc. even that can make quite a difference.
Like it was said before, a halfways decent card should do if it's only for Audio applications.
I am always more comfortable with cards over built-in graphics, since I can easily switch them out (and in the past, I was never impressed with intel graphics).

Cheers,

Tom

[Obsolete317542]

moijfghjip^ùcxpoifjgh

[pekbro]

Quite a few plugins use OpenGL processing that is offloaded to the GPU. (Waves, Fabfilter, etc.) In my old PC I had an nVidia GTX 760, and it caused me a lot of headaches. Apparently nVidia's OpenGL acceleration is by default multithreaded on the CPU, and if the OpenGL code is not the most robust or modern, this causes stalling of the other processes running on that CPU. This in turn creates CPU spikes, which creates audio dropouts. It took me almost two years to find out that nVidia's multithreaded OpenGL acceleration was the cause of endless CPU spikes and audio dropouts when using Fabfilter plugins! It can be disabled in their drivers but still... If I were to build a new audio workstation these days, I would strongly consider AMD graphics, or just using iGPU.

Also, weren't there some reports that state nVidia graphics create more DPC latency? Something to look into before buying!

I have 2 GTX770s, never really had a problem and I have like 1500 plugins, 12 DAW's. Just sayin...

What exactly are you saying? That it didn't happen to me? That I didn't solve the problem after two years of suffering horrible audio dropouts when using Fabfilter plugins by disabling Threaded Optimisation in nVidia's control panel? Are you saying that there are no problems with Threaded Optimisation in nVidia's OpenGL drivers at all? I'm confused. How could you say anything about what happened to me when you weren't there to experience it?

There are plenty of reports of people suffering from the exact same problems as I had from just after Waves switched to OpenGL accelerated graphics. The cause of these problems was never found. It's Threaded Optimization. Threaded Optimization also causes havoc for a lot of OpenGL games. They will have horrible performance until the option is turned off. After that, everything is fine. Look it up, just sayin...

Whoa, easy there bud, I said no such thing. And no I've never heard of that if you
must know. I have been using Nvidia for years, for much more than running music software.
Whatever your experience was, I'm sure thats not all of the story (whether you realize it or not), something like that would be well known round these parts certainly. Nvidia's engineers are not
stupid, I can tell you that much.

[mgw38]

Virtually no software offloads processing to GPUs for the simple fact that they're not that good for processing most types of data due to the latency of transferring data from memory to VRAM and back again. A lot of software will use the hardware acceleration of the graphics card for blitting the buffers to screen, but that's standard, and nothing to do with GPUs.

GPUs are good for AI because of the large number of cores that are optimized for doing very basic things. That is pretty much perfect for running deep neural networks. Nvidia took advantage of that with their Jetson series. I think they just released a special CUDA based SDK for AI.

[Obsolete317542]

wdsofmihjipgodf^ùxpifdjht