I wonder if this could be used as an add-on for NVENC encoding for Intel and AMD rigs/GPUs. Would be super nifty if so.

Seems like a neat concept, even if it's a mobile 3050 with a halved memory bus, which doesn't spell well for the 4 lanes of PCIE 3.0 that Thunderbolt 3.0 provides.

Even then, it can probably find it's niche.

And even if home AI acceleration doesn't take off, we'll have a wide availability of hardware, perfectly used for the thing that Doesn't Add to Society (but sells tons and tons of very expensive cards).

Cryptomining.

There's a lot of crypto money still being invested in marketing this bullshit. WCCFTech basically has weekly reports on how crypto has entered bearish market. Elon Musk has started hyping Doge coin again. Your local crypto enthusiast is writing on social media how fiat money is even worse than crypto, banks are collapsing, buy crypto now, buy low, sell high!

It's inevitable.

ToothlessI wonder if this could be used as an add-on for NVENC encoding for Intel and AMD rigs/GPUs. Would be super nifty if so.

You certainly could, it should behave just as if connect via PCIe, the latency penalty of thunderbolt should probably pose no issue.

But makes no sense as despite the RTX A naming, this is just a jacketed RTX3050 in almost everyway. For your idea specifically, it faces the same 3 stream limit as GeForce parts previously (now lifted to 5 for GeForce yet remain 3 for the A500!) - developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new/
This is an ampere part so no AV1 yet.


This is also a thunderbolt part w/ RTX A GPU, with AI gimmick marketing. is absolutely doomed as a product.
If intel decided to do this - USB-C powered TB A380, it would probably outsell this as that product would serve an actual use case.
In the mean time, even using spare PCIe 1x for an currently overpriced A380 would make much more sense if intention is AV1 encoding.
Actually that makes no sense either - at that pricing, one should consider selling current GPU and upgrade to ADA or NAVI3X for main rig, unless GPU is for external streaming server rig.
To throw more spanners into the mix, Intel is release meteor lake soon, which likely brings iGPU/QSV AV1 support, but for existing builds is not an option.

I've put more thoughts into this than initially intended, or asked for. There are also niche situation I did not mention such as encoding server using something like older GeForce with software circumvention. (NV probably patched this for AV1 cards because corporate fucks why wouldn't they)

And this time it took 5 years for NV to get from 1030 to 1630, all using TU117 - a silicon using the Volta NVENC rather than Ampere NVENC. Depending on how common AV1 encoding is deployed on corporate uses such as zoom meetings, NV may choose to use a silicon that are AV1enc-incapable, because leatherman do not care about DIY at all. With advent of RX6500, the future of low end Radeon is uncertain, and nobody seems interested in serving the (proper)low-end market. Intel might provide an odd path via its future low-cost CPU w/ QSV, or maybe Ryzen too. But in 3 years time (this tech cycle), I don't see a particularly cheap (sub $100) way to do some AV1.