madVR - high quality video renderer (GPU assisted)

[ashlar42]

Originally Posted by SamuriHL

I can't afford an Envy at the moment, but, I am VERY curious to see what a commercial version of madvr looks like. Plus will we get the algorithms he's currently working on for the Envy? I think it would be plausible for madvr on the HTPC being used as the "test bed" for new features being brought to the Envy. THAT would be beneficial all around.

Yes, this is exactly how I'm feeling. Curious about the commercial version, curious about whether features will filter down from Envy to HTPCs or the opposite or both.
One thing is clear: there's no way I'll consider an AMD card for madVR use in the future. With Envy being Nvidia based, I think we now can make sense of madshi telling people to keep in mind RTX cards for a future-proof approach.

[huhn]

but a 2080 ti can't do 100 TFLOP at 32 bit but only at 16 bit. just saying and no HDMI 2.1.

the price was to be expected and he clarifies i don't have to "care" about this product which is transparent and fair.

madshi should really clarify if madVR will use tensor cores in the future or not that's a yes for me "- Pro Cinema: about ~10 TFLOPS, plus ~100 TFLOPS for AI / Neural Network" but not a 100 % yes from him.
which is something we should know IMHO.

[SamuriHL]

One thing is clear from the latest HDR test build release notes and bug that appeared....the Envy IS running madvr...at least, a code path within it. Because the release notes mention this and the bug he had to fix was because of some Envy code. But, remember that what they're using in Envy is not off the shelf nVidia cards. Who knows what else is under the covers of that thing. Nonetheless, I am quite happy I bought an RTX card. Even if the features don't 1:1 match the Envy in terms of new algorithms, I've got about 7.9 TFLOPS to play with on my card, so, right in the middle of the base and the pro. I'm very excited for the future of madvr!

[QBhd]

Okay this is the follow up to my scenario posted above that I was hoping to get some un-influenced opinions on before going into detail about the puzzle I can't figure out.

From above:

• CPU: FX-8350
• RAM: 8GB DDR3 1866 CL 7
• GPU: R9 270X 2GB
• No hardware decoding
• HDMI 1.3 output
  
• In your expert opinions, Is it possible to play 4K HDR Bluray remux using madVR?

Now the puzzle I can't figure out is, I can! When I got my 55" C9 OLED I tried an HDR UHD Bluray Remux and it was a messy slideshow. So I just used the TV to play back these files. There was no way my GPU could do what I want, right? However, by accident I loaded up a 4k HDR "calibration" file and to my astonishment it played perfectly! WTF?!?!?

So I did a deeper dive into some madVR settings. Turns out when the bitrate spikes to about 100Mbps queues would collapse and the slideshow would start... But increasing queues with only 2GB VRAM won't work, right? But it does! And the bigger the Present queue the better! It makes no sense. Since Present queues >4 make 1080p playback problematic. I just don't understand how these big queues are possible with 4k. Here is a screenshot of madVR OSD:



This is all with NO "trade quality" boxes checked AND tone mapping enabled (although not crazy):



The real kicker is my GPU is running out of VRAM, constantly dipping into system RAM and then clearing and repeating the cycle over and over:











So how is this possible? Could the very low latency of my RAM help? And why do bigger Present queues work with 4k but not work well with 1080p upscaled to 4k. And big CPU and GPU queues are even worse for HD content. I needed to create profiles for the queue sizes! That doesn't make sense. I wish we could create profiles for the Present queue, but I found a middle ground with 8... still okay for the 4k and the problems with 1080p are acceptable.

Anyway, sorry for the long post... What are your thoughts?

QB

[huhn]

they are using the same TU102-300X-K1-A1 like every other 2080 ti there is nothing different from it and they will most likely not even get the chip from nvidia. nvidia doesn't deliver so low volumes.

[SamuriHL]

I haven't seen that specified anywhere. Are you assuming that's the case or do you know this as a fact? I'm not questioning the validity as it seems logical but just want to know if that's been officially posted somewhere.

[Warner306]

Originally Posted by huhn

but a 2080 ti can't do 100 TFLOP at 32 bit but only at 16 bit. just saying and no HDMI 2.1.

the price was to be expected and he clarifies i don't have to "care" about this product which is transparent and fair.

madshi should really clarify if madVR will use tensor cores in the future or not that's a yes for me "- Pro Cinema: about ~10 TFLOPS, plus ~100 TFLOPS for AI / Neural Network" but not a 100 % yes from him.
which is something we should know IMHO.

I'm sure he cares about participating in his own business. Not wanting to kill off this project or limit his ability as a developer to pursue his own interests are likely what he is referring to.

The price sounds strange when things started out open-source, but I think the base model is still priced below his competition in the Lumagen Radiance Pro. I doubt the Lumagen folks are bilking everyone with its pricing.

[SamuriHL]

He made it very clear that he will retain control over madvr development. Nothing on that will happen without his input. And he said there will continue to be a free version going forward, as well. So people can relax on that issue. I'm curious to see what the commercial version brings in terms of functionality. I suspect some of his new algorithms that he's developing for the Envy will probably show up there, but, this is only a guess. Very exciting times for us!

[huhn]

Originally Posted by SamuriHL

I haven't seen that specified anywhere. Are you assuming that's the case or do you know this as a fact? I'm not questioning the validity as it seems logical but just want to know if that's been officially posted somewhere.

When I researched FPGAs, I didn't find any (that are even remotely near to what we can pay) which have the GFLOPS that I need.

what do you think an custom nvidia card cost.
is it even possible without million to mass produce a new die at processes used by nvidia.

I'm sure he cares about participating in his own business. Not wanting to kill off this project or limit his ability as a developer to pursue his own interests are likely what he is referring to.

he clearifies that if you don't want to use an external device for playback there is no need to switch to an envy multiply times now

The price sounds strange when things started out open-source, but I think the base model is still priced below his competition in the Lumagen Radiance Pro. I doubt the Lumagen folks are bilking everyone with its pricing.

did you ever read the specs of an lumagen that thing is... i have no words for it.

this is a very low volume device specifically the first ones aimed at a group that pays high dollar for there projector and it is far more capable then a lumagen and priced similar. the hardware alone will cost 1.5-3.0 k and that not what they are selling they are selling the algorithms the work that has been input into writing these algorithms. how else should they be able to even make money with such a product.

i'm pretty sure "all" algorithm will show on madVR at one time as long as he develops them.

when was madVR ever open source?

[SamuriHL]

So it's essentially a 2080 in the pro. That seems reasonable and means that in theory the HTPC's that have the RTX cards could run similar algorithms that the Envy can drive. That's quite exciting. I did ask about the automatic algorithm "tuning" that the Envy can do (if you select multiple algorithms that require a lot of processing power it can auto tune them to give you the best picture and performance) showing up in madvr and he said that it'd be too difficult given all the variations in hardware on an HTPC. I jokingly (sort of) said he could limit it to RTX cards. LMAO That would be a sweet feature, even if it's in the "pro" commercial version of madvr.

[Warner306]

Originally Posted by huhn

did you ever read the specs of an lumagen that thing is... i have no words for it.

this is a very low volume device specifically the first ones aimed at a group that pays high dollar for there projector and it is far more capable then a lumagen and priced similar. the hardware alone will cost 1.5-3.0 k and that not what they are selling they are selling the algorithms the work that has been input into writing these algorithms. how else should they be able to even make money with such a product.

i'm pretty sure "all" algorithm will show on madVR at one time as long as he develops them.

when was madVR ever open source?

The margins are probably a lot smaller than you think and you are forgetting about the countless other expenses involved in running a business. A low volume product often requires a higher price just to cover these costs, let alone make a modest profit.

I meant madVR was free, not open-source.

[chros]

Originally Posted by SamuriHL

I'm very excited for the future of madvr!

Not to mention that (I assume and don't actually know this) madvr was "only" madshi's hobby beside of his full time/contracting jobs in the last 10 years. Now this has been changed: Envy and madvr will be his full time job!

[Grimsdyke]

Originally Posted by SamuriHL

One thing is clear from the latest HDR test build release notes and bug that appeared....the Envy IS running madvr...at least, a code path within it. Because the release notes mention this and the bug he had to fix was because of some Envy code.

@ Madshi.
When I try to play a file with build 87 my TV turns black and then reports no input signal !! After turning off and on the TV MPC-BE has loaded the file but with wrong refresh rates, etc.
Build 86 works perfectly fine.

[ryrynz]

Originally Posted by Grimsdyke

@ Madshi.

Reply in the AVSforum thread.

[SamuriHL]

Yeah you won't get a response to test build issues here. You should post any issues you have with them in the proper thread on avsforum. You may want to try downloading it again and see if that helps though.

Sent from my SM-G975U using Tapatalk

[thighhighs]

QBhd
Just curious, what other settings used? What dithering, any image enhancements?
What happens if you set HDR to let madVR decide?
What happens with fast chroma upscaling algo like bicubic?
Also is there any reasons why you prefer FSE instead of FSW?
I belive 8350/270x is suitable for 23fps HDR. And for 1080@23 -> 2160 too.

[QBhd]

Just to be clear... Everything works, I just want to figure out why. It makes no sense. High bitrate 4k HDR should he a slideshow with the settings I am using. I hate not being able to figure something out.

For some reason Windowed mode does not work as well as FSE so it's what I use.

Letting madVR decide for HDR won't work since Windows and I think madVR both don't recognize the GPU as being HDR capable... But I haven't really tried that since passthrough and tonemapping both work.

Ordered dithering, no enhancements.
My settings are not super crazy, but they should be swamping my system with a 4k UHD HDR bluray remux.

Any ideas on why this IS working?

QB

[tp4tissue]

If it's decided that Envy should use an Nvidia Gpu,

They might as well release Madvr as a PC KIT, some sort of HDMI pcie input card,you can stamp these out by the thousands.

Then have a PAID Madvr software version on desktop, compatible with ALL GPUS, Why make it a standalone device, it sounds like a good idea if you got a completely different hardware branch, But if it's gonna be Nvidia GPU, Madvr is already right here working all dandy like. ..

Keep it simple..

Turn the PC into a receiver.. that's fine.. and we won't require a second $2000 for a Specialized device, which does the same thing, but is less versatile.

The majority of people don't even have a living room anymore, they just sit in their rooms with a tv hooked up to PC>

The living room is dead, <let's build more living room devices ?>

[thighhighs]

Originally Posted by QBhd

My settings are not super crazy, but they should be swamping my system with a 4k UHD HDR bluray remux.

I see 90% GPU load on your screenshots

Looks like 1080p and 720p upscaling more demanding for your GPU than native 4K resolution playback.
On my experience (fx8350 ex-user) this CPU can decode 4k remux very fine with LAV Video. But I'm not sure about your GPU. What exactly mean "bigger Present queues work with 4k but not work well with 1080p upscaled to 4k"? VRAM limitation?
For example, NGU doubling/quadrupling increases the vram usage. As i remember i used bicubic chroma and med or low NGU luma with my 2gb nvidia GPU for good performance.

[QBhd]

Yeah. I go above and beyond VRAM, but it still plays smooth without dropped frames. I want to figure out how this is possible.

And no, upscaling is not more demanding with the same settings.

I don't have a problem. I have a puzzle. Why do I have no dropped frames even though the GPU is constantly running out of VRAM.

QB