AMD CDX Bus Landing For Linux 6.4 To Interface Between APUs & FPGAs

[abott]

Could be. Depends on the size of what you're trying to accelerate vs. the size of the FPGA you're using. If the FPGA isn't big enough, its performance could suck. You need a lot more die area to implement a given logic on a FPGA than ASIC.

According to this page VCN 3.1 supports various encode/decode of: MPEG-1, H.262 (MPEG-2), H.263 MPEG-4 ASP, VC-1/WMV 9, H.264 (MPEG-4 AVC), H.265 (HEVC), VP9, AV1, and JPEG.

All of which I'm sure went through tons of validation and trouble to put into silicon. An FPGA FFU is just common sense. Sure, size could be an issue. Too big? Okay, then include a coprocessor that runs some microcode instead of they don't want to put a huge FPGA inside the CPU. that's still 100x better than dedicated hardware on the chip. Media engines in hardware are a joke.

[coder]

According to this page VCN 3.1 supports various encode/decode of: MPEG-1, H.262 (MPEG-2), H.263 MPEG-4 ASP, VC-1/WMV 9, H.264 (MPEG-4 AVC), H.265 (HEVC), VP9, AV1, and JPEG.

All of which I'm sure went through tons of validation and trouble to put into silicon.

Again, you don't actually know how much of that is hard-wired vs. implemented via embedded DSP cores or the like. You presume a very clear, bright line between ASIC and programmability, but I'm telling you the line is typically quite a bit blurrier.

Media engines in hardware are a joke.

They seem to be a joke that everyone is telling.

[abott]

Yeah, and so is all proprietary software. Who cares what everyone else is doing? Fuck them. We use software encoders because of the bullshittery.

[coder]

Who cares what everyone else is doing? Fuck them. We use software encoders because of the bullshittery.

Well, that's just the point, isn't it? Hardware accel needs to be fast and efficient enough to be meaningfully better than software. Efficiency matters for laptops, since it limits the amount of battery life during video playback & encoding. Withe video conferencing now being so pervasive, that's not just about watching movies or binging TV shows on a long flight.

Also, an aspect of efficiency I think you're overlooking is that it's not enough merely to encode or decode in realtime. Many use cases involve faster-than-realtime, either to enable multi-stream or to accelerate things like video editing & transcoding. Even if all GPU users don't require faster-than realtime, each GPU maker needs to service such needs. Therefore, they can't get away from building more scalable and efficient hardware codec engines and the entry-level user is just one beneficiary.