<e​longated:matrix.org> Why you want to brick something? It’s cpu mining

<e​longated:matrix.org> Unless you want only intel and and mining

<e​longated:matrix.org> Unless you want only intel and amd mining

<s​yntheticbird:monero.social> Formality vs Objectivity

<s​yntheticbird:monero.social> Formality: RandomX is an algorithm made to be inefficient on ASIC. X9 is a very efficient CPU, there is no problem

<e​longated:matrix.org> Botnet-dominant-algo ?

<s​yntheticbird:monero.social> Objectivity: RandomX is an algorithm made to fight ASIC centralization. X9 is 77% more efficient than the competition, drawing people around it. There is a problem.

<s​yntheticbird:monero.social> and I can even argue that RandomX is made so that "everyone" can access mining, so in that sense intel and amd wining is not an issue

<s​yntheticbird:monero.social> you just have to adapt it when ARM will conquer the HPC world

<e​longated:matrix.org> Botnet operators rejoice

<e​longated:matrix.org> Someone made an efficient cpu ? Fork it

<e​longated:matrix.org> What was qubic using to mine ?

<e​longated:matrix.org> Not bitmain but same cpus you want to have better hashrate

<s​yntheticbird:monero.social> actually iirc, they forced their nodes to have threadrippers

<s​yntheticbird:monero.social> and you want EPYC cpus

<s​yntheticbird:monero.social> but anyway, Qubit sponsoring is still \* ahem \* a mystery \* AHEM AHEM \*

<e​longated:matrix.org> “Mining”

<s​yntheticbird:monero.social> so I would argue that doesn't count because they really could have just bought the X9 and attack monero too

<e​longated:matrix.org> ztrash ecosystem, right before their pump

<e​longated:matrix.org> Reorg made sure we have slow money flow due to increased conf by exchanges

<e​longated:matrix.org> Nobody in right Mind is going to buy x9 to kill a coin they are mining

<e​longated:matrix.org> But you could do that by renting servers

<e​longated:matrix.org> Same thing randomxv2 wants efficiency for

<e​longated:matrix.org> Anyways you guys killed xmr in 2018 , feel free to do it again

<s​yntheticbird:monero.social> what happened in 2018?

<s​yntheticbird:monero.social> I wasn't born yet

<g​ingeropolous:monero.social> yeah, just upping the dataset seems not right. though, at some point, we may get 2gb cache .. granted, that could be a while.....

I didn't say that I want to brick X9. Because even if it's bricked by some specific change, next version will return with the same efficiency gap. It's more sustainable to close the efficiency gap by making it more fit for regular CPUs, so this is the main plan.

what's the word on the x9

haven't done any nonce analysis lately but i'd wager it's no different from the x5 pattern

anybody done the heavy lifting yet?

also jumping straight to x9 seems to infer there was an x7 that didn't hit the market, curious

been keeping eyes on sophon and no real sign of new cores on that front, possible that bitmain put together a rx-specific core for the x9 instead of reusing their existing core

did we ever figure out if sg2042r was particularly different from stock sg2042?

No, sg2042r never had public datasheets, so no idea

nonce patterns seem to be the same

hm, figures probably roughly the same host firmware, job distribution, program generation etc

just a more efficient core

i never had luck getting a decently priced x5, best offer i got was $1500 lol L

i did have some potential leads on loading custom code on x5, it really looks like they run the sg2042r more or less bare metal, small bootstrap program to bring up the other cores and load jobs/send responses

found some snippets of rv64 code inside the arm-compiled xmrig binary on the host firmware

eureka: I posted some nonce analysis from before focusing on their stripes as well, if anything they have stayed stable or gone down over time

11:52:25 <DataHoarder> irc.gammaspectra.live/00bff44cf801ed35/out.png

11:52:46 <DataHoarder> remade nonce pattern, from randomx fork or so to last block today

12:14:13 <DataHoarder> zoom into their patterns irc.gammaspectra.live/73e2bfbf1b91b112/out.png

12:15:44 <DataHoarder> nonce % 2^28, remove groups nonce / 2^28 that are 0 or > 10 (0 has a lot of contamination, and higher ones don't appear in nonces)

12:16:12 <DataHoarder> then their pattern is on the bottom 1/16th of this. That is the range of the plot

amazing bitmain chose this path again.

The zoomed-in nonce suggests 32 mining threads per CPU (there are more than 32 stripes, but they fade out quickly after 32nd).

There are 11 stripes (0-10) and I have stacked them all together via modulo. 0th is removed due to the amount of other stuff there. I haven't calculated difficulty / hashrate for these bands yet, I estimate I'll be done sometime next year

On the zoom-in stack they appear slightly offset from why would be powers of two, so I'll check that a bit closer as well

Do we have insight into the binary code what was making these splits?

the SG2042 is really a pretty weak chip compared to AMD Epyc arxiv.org/html/2406.12394v1

but they would have released the SG2044 last year github.com/RISCVtestbed/riscvtestbe…s/files/hpcasia24/hpc_asia_wang.pdf

SG2042 had 4 channels DDR4-3200. SG2044 has 8 channels LPDDR5-9000

SG2044 would be the most likely chip in the X9 arxiv.org/abs/2508.13840

same cores as SG2042, but upgraded to vector extension 1.0, and upgraded memroy controllers

2x24GB is 1687euro ldlc.com/en/product/PB00668682.html

DataHoarder: interesting, also that strange gap in the middle there

odd pattern changes when the nonce pattern stabilised (after 2500000) too

perhaps the X9 has already been running for some time, and the weakening is them preparing to ship it?

07:28 < DataHoarder> Do we have insight into the binary code what was making these splits?

i haven't dug super deep into their xmrig binary, just a cursory glance, but i would assume it's per-chip job distribution as that typically creates the band effect on most ASICs

i don't have a working copy of IDA atm so can't dig too far

<t​om229:matrix.org> m-relay:

Sorry for being out of the loop. Is HashVault completely dead right now or something?

They have been in and out DDoS to several degrees

Usually they still produce blocks

Their last block was recent blocks.p2pool.observer/block/2c369b…7f742b18b63b5e8ad627ef003c26c938a9c

And it was reported by their API

Thanks.

It does look like it was 30 blocks or so ago

Probably not good for SupportXMR to be having like 40% hashrate :|

Yeah, but HV had issues already a while ago

When the growth went even faster

The growth of SupportXMR?

why pay to mine on supportxmr when you can mine to moneroocean for free? I dont get it.

Why not use P2Pool? It's free and more decentralized.

Yep, jpk68 . TBH. New people join in, and they go straight for biggest and forget about it

besides the overhead of running your own node, but also good point

They don't even think about it

tom229: You can mine on P2Pool with a remote node

I'm working on getting p2pool on browsers as an experiment

And somehow bridge that to xmrig

Interesting. So P2Pool is in the browser, but the miner is native?

It would be cool to see if P2Pool could be integrated into that guy's WASM webminer software

Yeah wasm, and they can peer on each other via webrtc. And connect to remote or local monero

Local hashing ofc, even in full mode it's not performant on WASM

not to change topics, but it looks like these custom risc chips are becomming a problem. First the custom sg2042 and now whats ever in this bitmain x9

does anyone know what rxv2 does to mitigate?

It's not 100x improvement over CPUs

V2 is about CFROUND as rounding mode switch on AMD is slow, to bring them back to par

And AES mixing of registers instead of XOR

It could include more now :)

There is backlog to read as well where it was discussed

I could be wrong, but it seems to me that RandomX v2 also significantly improves the energy efficiency of x86_64 processors.

Meaning that optimized RISC-V 'ASICs' have less of an efficiency advantage there

awesome. Thanks for the link and explanation. Looks promising

interesting speculation that the only reason bitmain is selling these is because of rxv2. So maybe they're worried

You mean that they're trying to get rid of them before the hardfork?

Either that, or they're already deploying X9 successor in July

Not "yet" jpk68 but it could do that

they almost certainly have a successor for july. Thats their whole business model. Sell the old stuff with "pre order" to finance the new line of production

mine with the new stuff until profitability wanes to ~2 years, pre-order that 6 months out, repeat