GPU decoder only needs to decode 2x 4K25 videos at once with the benchmark or is it more?

2 with 1 in a moving pip which transitions to a 3rd. Plus multiple sections with 2 and 3 full-screen composite envelopes. However I simplified the hevc 4k 30p Red Car with media substitution to use the same 8-bit hevc clip for all 3 which the bench as designed plays concurrently with different offsets. But easy enough for anyone to swap in their own camera clips. Have done that myself with larger 10-bit 4k camera clips with similar results.

Maybe a Vegas bug. No idea about the capabilities of hd750 decoder, but I doubt it would be the bottleneck in that benchmark project

I think the disappearing vp19 decoders are definitely a Vegas bug probably relating to their amd navi implementation... I had a vega64 in the xeon and it didn't happen there till I swapped in the 5700xt which uses the same driver. But vp18 on the same machine was fine. Vp18 has a pip bug with the 5700xt that blacks out the entire background around the pip window... this was fixed (except for a small border) in vp19. I suspect the original vp19 navi fix is the source of the occasional disappearing decoders issue.

I already know that Vegas hevc decoded on the 11900k/hd750 is substantially slower than on the 9900k/hd630... which is why I threw in the 2nd gpu, being that hevc is mostly what I shoot. When I get a chance all also test the same with ffmpeg -hwaccel on both machines to see if it's app-independent.

@Howard-Vigorita Would you know of any GPU decoder benchmark software? I've tried searching but never found anything. You would have thought they should be out there.

Btw I was looking at quality encoding tests by a user, and noticed HEVC encoding with your series of GPU's is noted as 40mbit/s bug, and it only encodes at 40mbit/s or higher. Is that an old bug that was fixed long ago?

Other thing interesting to see is how at 40mbit/s it's video quality is alleged to be the highest quality of any GPU, better than a 2 pass placebo encode. What do you make of that, It surely can't be accurate?

It also shows how truly terrible AVC GPU encoding is on both AMD 6000 series cards and 480/580, but all of these tests are for gaming, not the settings we'd use for NLE GPU encodes, and we still have the problem with GPU renders being higher quality than software renders, so not sure how credible this is

@Former user Don't know of any gpu decoder benchmark software. Ffmpeg can do a decoder-only measurement for nvidia with this powershell command but has not implemented decoding for amd:

measure-command { ffmpeg -hwaccel cuda -i CLIPNAME -f null - }

With Vegas, I find that the highest hevc bitrate I can set in the Magix Hevc template for cbr is 240mbps. But could only get 70 with various amd boards (6900, 5700, Radeon7) at 30 fps and 100 at 60 fps. Also amd boards only render 8-bit. With Nvidia I get 237at 8-bit and 240 at 10-bit. Intel: 240 all around.

 

I already know that Vegas hevc decoded on the 11900k/hd750 is substantially slower than on the 9900k/hd630... which is why I threw in the 2nd gpu, being that hevc is mostly what I shoot. When I get a chance all also test the same with ffmpeg -hwaccel on both machines to see if it's app-independent.

@Howard-Vigorita A benchmark that compares your Intel IGPU decode with others. It appear to be up to 150% faster than your 6900XT GPU, with the Intel ARC 380 decoder being much faster than everything else. I wonder if it's the same decoder in 12700K etc or of it's been improved. It might be the card to get if only interested in it's decoder and encoder, to be used as a 2nd GPU. Depends on how Vegas handles it.