I don't see many people pointing their nvidia's at tensory algorithm based coins anytime soon. 1070/1080 Tis still makes the most mining Zec. May be when Z9s are out, BTM will be worth mining with GPUs, or if btm price doubles before that.
If a batch 3 comes out, it becomes a D3. The Z9 will remove ZEC mining. If the GPU go to bytom, we will earn less then. We dont know if the coin price will double or halve. This is a niche coin and it is already 600m cap, so, I would think it is not so easy to double. The A3 was very different. It showed 20k per month profit before it launched, users broke even in 1-2 months and there is no gpu threat. The B3 had 2 batches with no gap between batch releases and has gpu threatening it. If gpu mining profitability for other coins fall, they will come to bytom to make more. If bytom price doubles for whatever reason, the GPUs will come here to push the profit down to the point mining bytom on gpu equals mining other coins on GPU. This thing cant make too much profit. The GPU miners put a cap to it.
Power consumption will come to play at that point, 300 watts vs 6x200/150 watts for a 1080 ti rig for identical hash rate. B3 will still come out as a winner but definitely income will be much less than what it is now.
Yes, there is a power consumption advantage (as all asics have over gpu) but this means that even if the coin price increases, the GPU miners will come in and mine the coin, pushing profitability back down. If the GPU miners earn little from all other coins out there, they will come to bytom to mine because it is their most profitable coin. This pushes B3 profits down. If the bytom price drops by half and becomes unprofitable, the gpu miners will leave and B3 owners hold a toaster. The risks are stacked against B3 owners. Other asics such as A3, L3, didnt have this additional issue. This is an asic coin but the profitability is capped and somewhat pegged to GPU profitability.
Anyways, lets hope this works out and we break even in 5-6 months from now guys. Cross fingers*
I mean I hope everyone looked into BYTOM, Tensority, Sophon, and the general ecosystem around this launch before making their purchase.
This was never going to be a quick win. BYTOM has a complex roadmap with a LOT of work needed to get there.
The bottom line is that the SOPHON TPU didn't sell well at all.
They were far too early to market with a GP TPU. Its why Google doesn't sell its TPU or TPUv2 to integrators.
They had a lot of "low corner" SOPHON ASICs that they couldn't use to manufacture SC1/2 cards.
They put them to work on a blockchain that uses proof-of-tensor as a PoW mechanism, in a design that doesn't require the expensive DDR that large ML pipelines use for DAGs.
At this point I'm doubtful we'll see a Batch 3 or 4 unless they're sitting on a few themselves.
Though, this would not be the most efficient product for them to run in their internal datacenters.
Much more likely to see a B5 based off of the BM1682/1684.
Understanding for how AI/ML is used domestically, and considering Bitmain's willingness to work with Beijing, I'm betting that the 1680 won't be their last TPU ASIC.
PS: Worth noting that Bitmain had already contacted the few existing Sophon SC1 customers with promises of coupons for their next generation TPUs.
