At that point, the fixed load of fans and controller probably start to become a more significant part of the overall power draw. It's kinda like how at low speeds in a car, rolling resistance will dictate your energy use more than wind resistance though wind resistance dominates at high speed.
|
|
|
Power use for increased hashrate isn't linear. The marginal increase in power use grows as hashrate increases. To run a higher clock speed, you need a higher voltage to have more juice to make all the transistors switch faster, but a higher voltage also results in a higher current draw, which means that since power is current times voltage, and both are increasing, you get an exponential increase. Like if it runs 1V/1A at 1GH the power is 1x1=1W, but if to get 1.1GH you increase to 1.1V, the current increases to 1.1A so power is 1.1x1.1=1.21W so you see an 21% increase in power use for a 10% increase in work.
Same the other way. If you decrease the voltage from 1 to 0.9, the current drops from 1 to 0.9, and now power use is 0.9x0.9=0.81W so you see a 19% drop in power use for only a 10% drop in hashrate.
That's not real numbers, by the way. The actual interactions are a lot more complex, but it's a conceptual illustration. When you undervolt and underclock something, the power use per hash will decrease more than the linear decrease in hashrate. So yes, two undervolted S9 at half power should have a higher hashrate than one S9 at full power, because the chips are running more efficiently.
There's also additional benefits from running colder reducing the current load, but that's a less obvious effect.
|
|
|
I'm not sure how much worse it is now but when I finished my EE/CpE over a decade ago, half my senior design class had never used a soldering iron, and that's at one of the best engineering schools in the midwest (usually ranked #1 when comparing quality versus price). It's definitely disappointing how much hands-on stuff isn't being taught.
|
|
|
Yeah we had to swap packages on the USB, and do a whole redesign of the main regulator. There've actually been three different versions of the Compac F so far, depending on what parts we could get ahold of. That also pushed the hub redesign, which I had been needing to do anyway.
|
|
|
If you're running a GS hub, just make sure you're feeding it actual 12V (the input voltage tolerance depends on the version of hub you're using) and you won't need the inverter and power supply.
|
|
|
Merch, 419 and ASICPuppy hsve received stock of the new hubs.
|
|
|
That's what the USB port is supposed to read, yes. But what about the core voltage, measured as described?
|
|
|
Power is volts times amps. 130A 240V makes 31.2KW and at 208V that means 150 amps. From 600V it's 52 amps.
Across one phase. Across three phase 130A at 240V 3phase 130 x 240 x 1.73 = 53,976 W Okay but he specified 130A on 240V single phase, which means he was drawing 31.2KW total. Switching to 3ph, let's say it steps down to 208V l-l, the equipment will still be asking for 31.2KW which means 150 amps total, 50 amps per leg. At the 600V side the total draw is around 52 amps (plus transformer losses), around 17 amps per leg. The 1.73 rule doesn't make power magically appear, just changes how it gets there.
|
|
|
Everything's tested for hitting at least 220GH at 400MHz across 24 hours before it ships. Peformance beyond that is expected but not guaranteed.
Weird they're underperforming at 400. Have you any kind of meter with which you can read the core voltage? Factory is 1.48V from the USB jack to the marked side of the block in the top half center with the stick plugged in but not hashing.
|
|
|
How well does it run at 400MHz?
|
|
|
NotFuzzyWarm and some others who know power could probably chip in with better specifics. My knowledge is largely academic.
|
|
|
Power is volts times amps. 130A 240V makes 31.2KW and at 208V that means 150 amps. From 600V it's 52 amps.
Any step-down transformer stage will lose a bit of power. If your 600V three-phase is significantly cheaper than your single phase (probably split phase 240?) you could probably come out ahead. When I ran a small DC, we switched into a "small industrial" rate schedule when we crossed 100KW draw on a 3-phase 400V feed and it dropped the average rate from 8.9c to 6.5c per KWh.
Since power is billed by KWh, raising your voltage to drop the current won't necessarily save on your bill. It'll depend on if they bill 600V 3ph at a lower rate than residential single phase.
|
|
|
I don't know why he doesn't update the listing photo. That's from the very first batch of hubs four years ago with the plastic case. Now they're a very snazzy aluminum with a three-color clear vinyl decal label. Way sexier.
But yes ASICPuppy should have the new hub in stock now. Pay attention to the new underside label, because there's new information for the new design. Most importantly how the self-resetting fuses on the ports behave and that if you're pushing it with more than about four sticks you should give it a bit of airflow through the case.
|
|
|
Yeah if you want to run more than a few sticks, you'll want to use the 6-pin for long term reliability.
|
|
|
It's physically identical to the existing hub, but uses a high-efficiency central regulator instead of several lower-power ones. That means there's no more "6A per port pair" limit. We also put 3.5A PTC resetting fuses on each port for safety.
|
|
|
First batch should start shipping by the end of next week.
|
|
|
Fix your quotes, dude.
1, yes ticket mask is internal to the ASIC. Like he said, only nonces above the ticket mask are put into the stream of data coming back from the chip. It's similar to how cgminer will get a pool diff, so only nonces above that value will be forwarded to the pool server.
2. When you testd nonces, did you give it all of the appropriate data? ASICBoost uses an additional data field not previously implemented (left as zeroes I believe) in order to generate multiple midstates simultaneously from the same data. If you're not integrating that into the hash, most of your nonces would test bad because you're testing with incomplete data. This explanation is probably insufficient and Kano can tell you more.
3. Sounds like a CRC issue
4. If SHA256 had universal nonces, wouldn't that invalidate the algorithm as cryptographically secure and make exploitation trivial?
|
|
|
You can find sockets on aliexpress for around $300 but they won't handle heat very well without modification.
|
|
|
No it can't fall out, it's a fixed rotating knob. If you spin it too far, it just starts back over at the beginning. It's basically a tiny volume knob but without stop-blocks at each end.
|
|
|
If it hurts, more cooling.
And yes, Windows generally sucks at USB IO and realtime tasks. The factory tests exclusively on linux machines. Kano's code is written for linux and ported to Windows to make people happy, but it's not ideal.
|
|
|
|