Is this Heartbeat usage Normal while rendering?

Eni wrote on 4/8/2022, 5:19 PM
https://i.imgur.com/6jkcIe5.png


Vegas Pro 19 - IS this normal behavior? as I haven't had this happen with Vegas Pro 14 and 16 that I've used.

Some common questions answered:
-No nothing is bottlenecking.
-No the Temps are very low
-Yes I've changed ram usage to 16GB.
-Using MAGIX Nvidia decoder for render
-4K rendering with color correction

Comments

Eni wrote on 4/8/2022, 5:20 PM

RogerS wrote on 4/8/2022, 5:31 PM

Yes, the NVENC encoder is usually up and down in Vegas. You can try Voukoder to load it more evenly though don't expect a massive time difference (10% faster?)

Of course depending on what Fx, etc. are applied the GPU could also be waiting on the CPU.

I would dynamic ram preview back to a normal value like 5 or 10% as there are diminishing returns and your system is only 32GB ram total.

todd-b wrote on 4/8/2022, 6:07 PM

@RogerS It's over 20% slower slower using Nvenc compared to Intel and AMD hardware encoding due to the poor implementation of Nvenc in the MagixAVC Nvenc encoder. The current Benchmark that you curate doesn't show that, that is why with @jetdv assistance a new automated benchmark can better highlight the problem.

A benchmark to show this problem is a simple 4K AVC to 4K AVC transcode, your current GPU heavy benchmark hides the problem. You need to come to the table on this RogerS, you are delaying a very important community project

RogerS wrote on 4/8/2022, 7:52 PM

I haven't seen that big an increase with Voukoder vs MagixAVC with NVENC (can't very well swap GPUs on a laptop and QSV doesn't work for me anymore) but maybe on some projects you would see a 20 percent increase.

I support what you are trying to do with more diverse benchmarks and automating the process. My only suggestion is to keep it simple for users to both do and interpret as it's easy to overwhelm with data. We need lots of participants to make it meaningful. Maybe TechGage would consider a partnership?

Musicvid wrote on 4/8/2022, 7:56 PM

Normal buffers.

todd-b wrote on 4/8/2022, 8:02 PM

@Musicvid not normal, only happens with NVENC. @RogerS I we definitely need a transcode, and I think GPU heavy (entire project) but if you think otherwise maybe just transcode and your current benchmark. Jetdv script may be able to upload results directly to your google docs after people complete the benchmark if you think that's a good idea. That's how puget benchmark work. It assumes users are unreliable at best, dishonest at worst. I will try voukoder vs MagicAVC 4K AVC and check results

RogerS wrote on 4/8/2022, 8:14 PM

FWIW I just did a simple transcode (UHD X-AVCS to UHD) and MagixAVC with NVENC was 1:28 and with Voukoder NVENC (recommended settings + QCP 19) got 1:18, so close to 10%.

TechGage uses MedianFx to test GPU and then AI-assisted Fx. I think if you did the legacy one, a transcode and for a third one, an AI Fx, that would be meaningful as Vegas is Intel CPU-oriented and that is something to consider for someone interested in using these Fx. Love the idea of automatically uploading to Google Doc (either mine or migrating mine elsewhere) to make this as easy a possible for users.

todd-b wrote on 4/8/2022, 8:28 PM

@RogerS Confirm your approximate 10% difference with 4K AVC. I"ll try other transcode resolutions to see if difference is higher.

4KAVC to AVC MagixAVC NVENC 1m41s (13% difference)

4KAVC to AVC Voukoder NVENC 1m29s

1080P AVC to AVC MagixAVC NVENC 30s (Same)

1080P AVC to AVC Voukoder NVENC 30s

720P AVC to AVC MagixAVC NVENC 35s (40% difference)

720P AVC to AVC Voukoder NVENC 25s

The 1080P figure is weird, will have to test again later

RedRob-CandlelightProdctns wrote on 5/17/2022, 5:21 PM

Although not contributing to the benchmarking conversation I wanted to chime in that on my new laptop (32GB GDDR5, i7-12700H, RTX 3070 Ti (8 GB RAM), I'm getting the same "heartbeat pattern" on my render using Magix AVC (NVENC) codec. It's very fast to me (2 hr show using 1080p59.97 footage renders in about 30 minutes) but super stuttery (heartbeat) as it renders.

I'm guessing Vegas transfers a bunch of frames to the NVIDIA card, it recodes them then sends them back to Vegas which writes it to the output file [repeat], and that handshaking/transfer is causing the behavior? Wonder if there's a way to speed that up?

CPU running about 40-60%, Intel GPU 35% (video decode), NVIDIA GPU 35% (video encode)

Although I'm watching that pulsing fast-pause-fast-pause-fast-pause [repeat] while it renders, my CPU meter in PerfMon isn't quite as evident as the OP's image. (pardon the watermark in the center.. just installed Hypersnap [love that tool!] on the laptop and haven't registered yet)

Musicvid wrote on 5/17/2022, 7:42 PM

Repeat: Normal Buffers

3POINT wrote on 5/18/2022, 12:03 AM

Repeat: Normal Buffers

I don't see this behaviour with other NLE's when using NVENC encoding, (also not with QSV encoding in Vegas). I doubt this is "normal".

todd-b wrote on 5/18/2022, 12:14 AM

@RedRob-CandlelightProdctns The pulses are seen when encoding via NVENC a basic transcode. The anomaly and slow down is most apparent because no other processing is hiding the problem. Infact with enough processing this latency is not of much consequence.

If you do the 4K benchmark project that has a thread dedicated to it in this forum, the difference between Magix NVENC (pulsing latency) and VOukoder NVENC(no pauses) the difference in render speed is tiny, however if the processing load is much less Magix NVENC may be up to 13% slower, 720P 40% slower. Many people know and understand the phenomenon including Techgage.com 's Rob Williams but everytime this is brought up @Musicvid will say it's entirely normal.

Musicvid wrote on 5/18/2022, 6:52 AM

So Vegas' software encoders don't work exactly the same way as machine encoders or other editors? And that makes Vegas "not normal?" I'm flabbergasted.

3POINT wrote on 5/18/2022, 7:51 AM

I would say the cooperation between Vegas and hardware encoders (especially NVidia) is not optimal as with other editors. To call it normal or not normal is how you look at.

Musicvid wrote on 5/18/2022, 8:46 AM

I actually prefer the term "different."

3POINT wrote on 5/18/2022, 10:38 AM

No matter how you call it, they all produce the same (less) quality despite "different" encoding speeds. That's the main reason that I don't use NVENC or QSV encoding for final production.

lan-mLMC wrote on 5/18/2022, 11:34 AM

No matter how you call it, they all produce the same (less) quality despite "different" encoding speeds. That's the main reason that I don't use NVENC or QSV encoding for final production.

If I have enough time, the x264's CRF rendering mode is always my first choice.  It gives H.264 the best quality.

3POINT wrote on 5/18/2022, 1:00 PM

@lan-mLMC Same for me, I use NVENC or QSV only for quick test renders.

todd-b wrote on 5/18/2022, 8:07 PM

So Vegas' software encoders don't work exactly the same way as machine encoders or other editors? And that makes Vegas "not normal?" I'm flabbergasted.

@Musicvid It's Nvenc that has the extra latency and the extreme pulsing of the hardware encoder . If there is a lot of fx processing overhead causing a slowdown it less of a problem. I don't have QSV or VCE but when i've got people to try this before they didn't have the pulsing latency and don't have the same speed difference between voukoder and MagixAVC QSV/VCE

In Resolve Studio it's said that from GPU decode, to GPU processing to encoding data never leaves the GPU, which removes latency dramatically BUT it requires video cards with high VRAM, recommend minimum card for 6.5K+ is Nvidia Rtx3090 because of it's 24GB Vram. I have not ran out of Vram with a 3080 in 8k resolutions but I have had the GPU decoder turn off because it couldn't fit the decoded frames within the Vram.

Vegas and Resolve NVENC are extreme opposites, there's problems with both and I'd never want to have the same GPU VRAM requirements in Vegas, but MEP and Premiere do an amazing job without the same requirement for high GPU vram. The problem spoken about in this thread is to do with MagixAVC Nvenc latency though.