08:58:15 tevador hyc If you're not in Ravencoin algo channel yet, join and read last few messages https://discord.gg/MsrzH6 08:58:45 GPUHoarder says that this board does 6.5 KH/s @ 75W https://www.xilinx.com/products/boards-and-kits/alveo/u50.html#specifications 09:45:43 how much does it cost? 09:54:07 answer: $2000+ 09:55:00 hash/$ is 4 times worse, but hash/watt is the same as Ryzen? 09:57:49 In terms of prices, ASIC+HBM will be roughly the same and will be probably limited by HBM too, so same 6.5 KH/s, maybe a bit more 09:57:54 roughly the same perf/watt if the board really pulls 75 W 09:58:00 but efficiency for ASIC vs FPGA is much better 09:59:19 DSP processors in FPGA doing 64-bit FP math is 2x worse than proper ASIC circuits, HBM is the same power, other logic in FPGA is 10x worse than ASIC 09:59:21 ASIC with HBM is a supply chain nightmare 09:59:29 now I just need to know proportion between these 10:00:33 would be interesting to know of the FPGA board is compute bound or memory bound 10:01:04 judging by hashrate and HBM bandwidth there, it's memory bound 10:01:30 Vega 64 did only 4 KH/s without RandomX instructions at all, and it has higher bandwidth 10:02:41 4 KH/s doing 55 random scratchpad accesses per iteration. If they save on L1, but have lower bandwidth than Vega, 6.5 KH/s seems about right 10:04:17 if there is ever a RandomX ASIC, it won't have HBM 10:04:48 it's not economically viable either way 10:06:50 yes, https://whattomine.com/coins/101-xmr-randomx?utf8=%E2%9C%93&hr=6500&p=75&fee=0.0&cost=0.1&hcost=0.0&commit=Calculate 10:06:54 20 years to ROI 10:07:26 even if ASIC saves on power and has the same price: https://whattomine.com/coins/101-xmr-randomx?utf8=%E2%9C%93&hr=6500.0&p=30&fee=0.0&cost=0.1&hcost=0.0&commit=Calculate 10:07:36 14 years ROI 10:08:05 in reality, CPUs will catch up and overtake in 14 years 10:11:46 even with free power it's just too slow return compared to cost of development and cost per device 10:12:29 But I think ASIC based on that FPGA design could easily do 200-300 h/s per watt 11:41:45 that channel is ... pathetic 11:42:12 people complaining that mining isn't profitable for them - the equilibrium state will always be near zero profit. 11:42:12 X30R anyone? lol 11:42:25 lol 11:58:56 hm, seems I can't talk in that channel 12:09:10 yes, me neither 12:15:50 oh well, no loss 12:19:45 You need to ask admins to get write permissions there 12:39:35 it doesn't seem like the channel will be productive enough to be worth the trouble 12:40:01 if that last post is seriously looking for 30 PoW algos. someone clearly hasn't learned their lesson yet. 12:55:06 "The only real way to stop a machine, is to disallow it to submit shares. The only way to know if a machine needs to be stopped from summitting shares, is to have a way to accurately ID each and every machine. The best way to do that is with hardware ID identification ability and a database of all hardware IDs that is allowed to mine RVN." 12:55:11 are they for real? 15:23:08 LOL 15:24:18 somebody unclear on the concept of "decentralization" 16:19:36 these guys still haven't figured out that static algorithms are a dead end 16:19:44 Alterhash / ethash, whatever... 16:21:29 sech1: Do you think it's possible to come up with randomx modification for cpu + gpu? 16:21:53 anything you develop to cater to GPUs will open the door for FPGAs and ASICs 16:22:14 leveraging a GPU means aiming for massive parallelism 16:22:35 it means small compute cores, replicated many times 16:41:15 ID every machine. yes. much decentralized. total wow 16:44:50 Lots of advantages though. You can know the network hash rate without having to estimate. You can ban those who try to cheat. Those who have too much hash rate. Or those you just don't like. Keep track of who does what. Even modulate the hashing power of people to match what you think the network hash rate should be. And to protect the environment. 16:45:11 hardware IDs are easily spoofed 16:45:23 Eventually you just keep your own machine mining, because that's the best end game. 16:45:33 you could also remove mining altogether, but that's bad for PR I guess? 16:45:41 You've solved global warming, 51% attacks, and all miner disputes. 16:45:58 how do you emit new blocks without mining? PoS? 16:47:45 you don't need blocks anymore at that point, just a database, and Coin Ltd. decides which transactions are valid 16:52:32 sech1: what is the story of RainForest? I haven't heard of that algorithm 16:53:25 it was some weird algo that turned out to be simple CRC32 in the end, GPUs did 100 TH/s :D 16:53:38 GPU miners got 1000s times faster when this flaw was discovered 16:53:45 lol 16:54:22 moar hashes = moar security 17:15:24 How long time ago have you received msg from any AV? 17:15:39 the last was in 2019? 17:26:04 yes, there hasn't been any follow-up after that 17:34:23 I still didn't get write access in RVN discord, so I'll post my comments about Alterhash here: 17:34:45 it's very likely that the addition of 2 X16R Single hashes into ethash will make the algorithm compute bound (based on the available GPU hashrates: RX580 does ~30 MH/s in ethash, ~8 MH/s in X16R) 17:34:58 secondly: there will be very large hashrate differences between blocks depending on the algorithm chosen for X16R Single (Hamsi is ~10x slower than Skein, for example) 18:36:00 "There is 64x memory accesses in between of x16r compute algos" - that's meaningless, GPUs have thousands of memory accesses in flight, but don't have enough compute to calculate X16R as quickly 18:37:18 Only 2 functions out of 16 are computed per hash, it can be possible for GPU 18:39:48 it's still 16 chained hashes 18:39:56 that's a lot of compute 18:40:16 32 additional hashes 18:41:06 let them show some performance numbers first 20:04:56 tevador no, they use only 2 functions to calculate the hash and they change every block 20:26:37 OK, so I misunderstood 20:26:55 but still that means RX580 hashrate will drop from 30 to 20 MH/s 20:27:09 on average