All the AMD pages with the detailed specifications of Rembrandt products show only 1080p decode capabilities, for instance:
Even for the new rebranded Rembrandt, e.g. Ryzen 7 7735HS, the same information is shown.
If in fact Rembrandt is better than that, then this is another failure of AMD to adequately document their products.
Even if Rembrandt is better than documented, Phoenix still adds at least AV1 encode/decode, 10-bit encode for all 3 codecs, 4320p encode for all 3 codecs and up to 5 times higher encode/decode frame rates.
Oh. The hardware encode/decode max bandwidth thing? THAT? I don't think it means what you think it means...
Let's see. Here's what it says on my
6650U:
And what it says for the
7840HS:
So if you take it at face value and that this is the color depth/format/bitrate it can do, VCN4 on 7840HS can't do 10 bit H264 encodes or decodes, can only encode at 8 bit for H265 and decode at 8 or 10 bit H265, and can only decode 8/10 bit VP9, and it can only encode/decode 8/10 bit AV1.
So if the assumption is that they can only do 1080p, why include a framerate that's non-standard? Like, who puts out 1080p videos at 786 frames per second? Most published content out there are 30 or 60 fps.
They also make the phrasing as awkward as possible. Why go out of your way to include "(SDR)" on the encoding categories...when 10 bit AV1 is clearly NOT SDR? The question is...is that what its capabilities really are? Or what the theoretical bandwidth limitations for the video decode hardware or the memory subsystem feeding the GPU would be?
For example, looking at the numbers quoted for the 6650U, if you tally up what is needed for 8 bit 1920x1080 @205fps, that's around 13Gbps of signal bandwidth. 8 Bit 2560x1440@115fps is also around 13Gbps of signal bandwidth. How much is 8 bit 3840x2160@51 fps? Also around 13Gbps.
If you go through the all of the bandwidth requirements for decoding/encoding the bitstream, the numbers all converge around 13GBps - Rembrandt can sustain a 13 Gbps video stream through VCN3.1.1, and that's consistent across all formats it can work with. The encoder/decoder can process a 13Gbps video data at whatever format it supports at real time. If your video codec's bandwidth is whatever fraction of that 13Gbps, it's maximum encode/decode can happen at the inverse of that fraction. For example, if I am encoding 10 Bit H265 at 3840x2160 @30fps on my 6650U, it'll need 8.96Gbps of bandwidth, and in the perfect world, it should happen at 1.4x of real time. Of course, that's theoretical - actual perfromance depends on the code quality, driver quality, hardware state, and etc. The Rembrandt using vceencc 7.14 on VCN3 actually ran it slightly below realtime (26fps instead of 30), but VCN can and will do it. How do I know? I asked the hardware. There's a hardware capability query function on vceencc:
C:\Users\WANg> C:\Users\WANg\Downloads\VCEEncC_7.18_x64\VCEEncC64.exe --check-features
device #0: AMD Radeon
H.264/AVC encode features
10bit depth: no
acceleration: Hardware-accelerated
max profile: High
max level: unknown
max bitrate: 100000 kbps
ref frames: 1-16
Bframe support: yes
HW instances: 1
pre analysis: yes
max streams: 16
timeout support: yes
H.264/AVC input:
Width: 128 - 4096
Height: 128 - 4096
alignment: 32
Interlace: no
pix format: YUV420P[7], YV12[2], BGRA[3], RGBA[5], ARGB[4], NV12[1](native)
memory type: DX11(native), OPENCL, OPENGL, HOST
H.264/AVC output:
Width: 128 - 4096
Height: 128 - 4096
alignment: 32
Interlace: no
pix format: NV12[1](native)
memory type: DX11(native), OPENCL, OPENGL, HOST
H.265/HEVC encode features
10bit depth: yes
acceleration: Hardware-accelerated
max profile: main
max level: unknown
max bitrate: 100000 kbps
ref frames: 1-16
pre analysis: yes
max streams: 16
timeout support: yes
H.265/HEVC input:
Width: 128 - 8192
Height: 128 - 4352
alignment: 32
Interlace: no
pix format: YUV420P[7], YV12[2], BGRA[3], RGBA[5], ARGB[4], NV12[1](native), P010[10](native)
memory type: DX11(native), OPENCL, OPENGL, HOST
H.265/HEVC output:
Width: 128 - 8192
Height: 128 - 4352
alignment: 32
Interlace: no
pix format: NV12[1](native), P010[10](native)
memory type: DX11(native), OPENCL, OPENGL, HOST
device #0: AMD Radeon
H.264/AVC decode features
10bit depth: no
acceleration: Hardware-accelerated
max streams: 16
H.264/AVC output:
Width: 32 - 4096
Height: 32 - 2160
alignment: 32
Interlace: yes
pix format: NV12[1](native), BGRA[3], RGBA[5]
memory type: DX11(native)
H.265/HEVC decode features
10bit depth: yes
acceleration: Hardware-accelerated
max streams: 0
H.265/HEVC output:
Width: 32 - 7680
Height: 32 - 4320
alignment: 32
Interlace: yes
pix format: NV12[1](native), BGRA[3], RGBA[5]
memory type: DX11(native)
VP9 decode features
10bit depth: yes
acceleration: Hardware-accelerated
max streams: 16
VP9 output:
Width: 32 - 7680
Height: 32 - 4320
alignment: 32
Interlace: yes
pix format: NV12[1](native), BGRA[3], RGBA[5]
memory type: DX11(native)
AV1 decode features
10bit depth: no
acceleration: Hardware-accelerated
max streams: 0
AV1 output:
Width: 32 - 7680
Height: 32 - 4320
alignment: 32
Interlace: yes
pix format: NV12[1](native), BGRA[3], RGBA[5]
memory type: DX11(native)
If it fits the criteria outlined, VCN will accept the codec and generate a video bitstream based on whether it adheres to width/height/color depth restrictions - VCN is actually quite flexible at what it can do, but it comes at the cost of not being all that fast (at least not on APUs where they are limited by main system RAM bandwidth and CU count).
Even then, I am not sure if this is entirely correct. How's that? Well, let's see - we can fire up Google Chrome, load up Youtube, pop up a 2160p60 HDR video (that's 10 bit and H264) and see how the video codepath handles it. Huh. Video Codec 0 is on nearly 90%, and CPU is only at ~13%. There's some framedrops here and there but it looks good, and the machine is on Win10's balanced power plan.
What about 1440p?
Yeah, it works. The codepath is sending playback to the VCN and not the CPU. So eh, VCN 3.1 can actually decode 10 bit H264. But then you can probably do that with Vega on a Renoir or Cezanne as well. There's really not that much functionality difference between VCN3 on RDNA2, and VCN2 on Vega for Renoir/Cezanne.
What that page tells you is not an indication of whether the VCN can encode/decode at a specific format at a specific bitrate, it's more like what your theoretical limitations should look like.
The Phoenix? If you feed the claimed bandwidth numbers and crunch them, they all converge around 40Gbps. Is it an improvement? Compared to Rembrandt? Yeah. Will it transcode videos at 2-3x multiples of Rembrandt? Probably - but that's something that Alder Lake can do already using Quicksync - the question is whether QSV have the same flexibility as VCN - my past experience with QSV is that it works on certain popular color/resolution/fps gradients only - feed it 8 bit H264 1080p30 and it'll work. Feed it 8 bit H264 1920x1200p24? It might fail with a "resolution not supported, this is not on the menu, guy" error.
On this HP Elitebook 845G9 with Rembrandt, I have a good quiet general purpose Windows laptop with decent casual gaming capabilities (RDNA2 is found in the steamDeck, XBox S/X and the
PS5 so the drivers should be fairly solid), a pair of DDR5
SODIMM slots, decent screen (1920x1200 400 nit IPS, though), good runtime on the battery and full USB4 support which acts as a form of future proofing...as in, I can just plug in something that'll do AV1 encoding in the future if I am so inclined (like an Intel A750 Arc or a later Radeon desktop GPU off a TB3 GPU dock), which this Beelink mini-PC can't, and I didn’t buy it primarily to transcode videos.
So is this Beelink good for media center duty? Sure...if your content is up to 1440p60, and fairly standard - why not, but the better choice for that use-case would be an Alder or Raptor Lake mini-box - they tend to run hotter but solid XeLP GPUs (a tad slower than their Vega/RDNA2 APU counterparts but much better than their previous iGPUs) and very good Quicksync functionality. But if it's better casual gaming and good general purpose CPU, the Rembrandt is good and available now. Or if you are willing to take a hit on GPU gaming performance but keep the Zen3 cores? A Desktop Cezanne mini-PC like a Lenovo m75q-2 with something like an R7 Pro-5750GE will actually be competitive, be cheaper and ships today. AMD was also rather quiet about the mainstream versions of Raphael (the AM5/Zen4 APUs that came out in Q3 '22, but they are mostly meant for high end machine with expensive board options). This is why there are no Zen4 TMMs (like Lenovo ThinkCentre m75q-3 or HP EliteDesk 805G9s) out yet.
If it's the Phoenix/Zen4 and you absolutely want RDNA3 graphics? Eeeeeeeeh, you might have to wait until Q4 for them to start showing up on mini-desktops...by that time it's just another 1-3 months until CES2024 when Meteor Lake (with Xe2 LPG on Intel 4) and 8x5x/Strix Point (Zen5 and RDNA3+ on TSMC N3) announcements happen.
