GTX vs Quadro w/ Decklink Quad 2 Build

Hi,
I’m fairly new to Touch Designer and am planning on building out a system to support 6x 1080p60 simultaneous outputs each supporting at most 3x layers of video concurrently.

I’ve been reading through the forums and seeing a lot of Quadro cards over GTX, but they seem to all be in relation to nVidia sync and using the DP outputs on the GPUs and not using something like a decklink card for the output.

So, as the title suggests, given my circumstances would it really matter whether or not I have a Quadro GPU (M400 of M5000) vs a high end GTX (like a 1080 or Titan X Pascal) since I’m not depending on the physical outputs of the GPU?

1 Like

Using a sync card is important when you want multiple gpu’s render their frames on the same clock pulse.

If you have 2 gpu’s, each rendering their own content, and this content is used for a led wall for example, a sync card ensures that the output is from the same frame cycle. If you don’t use a sync card, it is possible that gpu 2 is a couple of frames behind gpu 1. Resulting in a visual “tearing gap” on screen.

On seamless displays you will see the difference faster than displays that are not so “close” together. Nobody will notice a frame gap when two displays have a distance in between or an edge blending.

So it is your choice if this is important to you. Besides genlock & framelock, the Quadro’s also support edid management & mosaic setups, not found on the Geforce series.

One things that Quadro do help with in your situation is their faster download speed from GPU to CPU memory, which is needed to get the frames out to the Blackmagic card. I was doing a test recently and I wasn’t able to get 8 outputs from a AJA Corvid88 on a Geforce 970, but I was using a Quadro M6000. The difference was their download speed I’m pretty sure.
So, it’s possible a Geforce will do your 6 outputs, but you may not have much headroom for much else.

Malcolm, do you think the same thing you describe would be achievable with an M5000 card? Basically I was going to do a Decklink Quad 2 for 6 1080 SDI inputs, process and add effects in Touch Designer, then send out to six monitors via SDI to another Decklink. Is this feasible or should I get the M6000?
Thanks!

Hard to say honestly. If I had to guess the M5000 could, but it’s totally a guess. It’s very dependent on what you are doing.

I haven’t tested with the P5000 or 6000, but I’ve done plenty with the K5200 and 1080 TI, and I can tell you for certain you can do it with both, depending on what you’re doing you may run into headroom problems rather quickly. With the 1080 TI, running even 6 1080p outputs out of the Blackmagic puts a pretty good hit on the GPU and depending on whether you’re pushing shaders or video, you’ll either be able to get 30-60 FPS or be struggling to get 20. It really depends on the content and the rest of what you’re trying to accomplish. Disk speed is certainly a factor if you’re trying to push large files of this type resolution (3840x3240 or 11520x1080, for instance)…SSD RAID’s are a requirement.

You’ll be much more likely to get 60fps if you’re running the outputs at 720p, but depending on your system, your mileage may vary. I’ve done this both in Windows 10 and Mac OS extensively…it can be done. I would say get the 1080 TI first and see if that fits your needs. Its a highly capable card and a quite a steal compared to the Quadro versions (the P5000 is neck and neck performance wise and I believe more than 3 times the price). That being said, in my experience, I would not expect perfect frame lock from the Decklink Quad 2. Thus far, depending on many factors, you will most likely see tearing between outputs. It will largely depend on your content and how far to the edge your GPU gets pushed, at least in my experience. Some content and configurations it has gone completely unnoticed, in others, the tearing is too distracting and takes away from the content.

Dont build SSD raids if you can avoid it, better to get NVMe PCIe based storage now, it’s blazing fast and pretty cost effective if you don’t need TB upon TB of storage.

Hi all,

I posted my flop test with the Radeon Vega FE which did half of the performance that my current GTX 108 does on my Touch Designer point-cloud shader application. Very time-consuming so I would like to ask:
in order to show a large point-cloud in VR I need to upgrade my video card (currently gtx1080 non-ti and non-Titan). I need more ram (ideally 16go) and I m afraid of buying a quadro P5000 and finding the hard way it’s not about the ram only. My current config shows 16 million points easily (thinking it can do 30 million) on the GTX 1080, as soon as I push 50 million points I am at 20fps. Should I instead take a Titan X card with 12go of ram?

What is your GPU RAM usage at? If you aren’t hitting the limit, then adding more won’t help.

Hi Malcolm,

Got a hole in my memory, but the perf manager doesn’t seem to show GPU memory usage, where do I look again? Best! P

You can see it if you middle click on any TOP. Also Windows 10’s task manager has GPU reporting tools as well.

Hi Malcolm,

OK that what I had in mind, I didn’t miss anything then, will do that for now, best, P

Hi Malcolm,

So here is a sum up view:
A - The more i reduce the number of points to show with my shader (it takes points from a few TOPs in .exr formats and displays them in 3D, position and color are store in the .exrs), the more fps I get, with 16million I am OK. In editor mode TD says 90fps.
I attached pics
B - The GPU of the card isnt that full (less than half) see pic of Win10 Task Manager, also see GPU 3D at 98%
C - The openVR TOPs it seems are taking all the resources see pic of Perf Monitor
So it’s weird, I am thinking my shader might need a faster card? It renders all points if I let it but very slowly, the ram isnt full but as i said more points make it slower.
OR I need to update / check my VIVE Open VR network.
Maybe the card needs bigger internal “buses”, or RAM transfer speeds? No idea, that’s why I wonder which:
Quadro or Titan?
Best!, P

GPU at 98% means the GPU itself is overworked. It’s not a RAM or transfer speed issue. So you need something with more cores/faster cores

Thanks Malcolm,

Let me ask this then: how is it that TD is going to offload more calculations to the CPU?
Is it automatic somehow? Does this mean that a bigger GPU is not going to help? It’s just that the less points I display the better my framerate goes, and that is pretty linearly verifiable, best! P

Hey you know what I thought you mentioned the CPU, so OK, I need a bigger GPU.
Fine, is the Titan V versus a Quadro P5000 a good choice then?
Best! P

If I can get my hands on one :slight_smile:, or a previous Titan, P

I’m running a P6000 and outputting 30 FPS to a total of (24) 1080p30 displays via 3 DeckLink Quad 2 cards. It’s definitely possible, but it does begin to bog down if you try and push too much onto it. I’m having to work quite heavily on optimizing my pipeline, including custom GLSL renderers.

I’m still not fully there yet, but it’s definitely possible. I’ve actually got two more P6000s in the rig that I’m going to attempt to use for other bits of the project and push the resultant renders to the main one to be added to the output. I have no idea how it’s going to come out.

Did you use several P6000 in one server outputting to all of 24 displays from one? How does this setup behave? Any issues?