Topic: Swedish ASIC miner company kncminer.com - page 1106. (Read 3050068 times)

And I'd be willing to bet a few satoshi's that the unit you took to Atlanta did not have the same firmware as those units that were subsequently shipped out to Day 1 customers onwards...and that the voltage was tweaked upwards in the interim thus increasing the power requirements (in the name of increasing performance/stability, I'm sure).  Couple the fact that the actual physical design of the product was changed quite substantially (8 VRMs to 4 VRMs) and it's not exactly apples-to-apples

Even in the "launch video" you have footage of the kill-a-watt pulling ~850W at the wall at (I assume) 240V with (I assume) 8 VRMs per board.

Not insinuating anything malicious here..just saying that what was put out there ended up being inadequate given the issues that cropped up with some of the 850 PSU's (HX850 that you had so much fun with, for instance).

Umm, that's cool, but I can assure you not that's clearly not he case on all units. I have pictures of the one I took to Atlanta using a Kill-A-Watt, and there were enough witnesses that were present, including members of this forum (Bargraphics, Phin Gage), and representatives from competing companies that saw this for their own eyes. Also there's plenty in this thread amongst the first recipients to verify otherwise.

True I didn't say it was zero just minimal.  Compare the annual savings to the higher PSU cost and you see the break even time is measured in years (possibly a decade at lower wattage and lower power rates).  The flatness of the curve really kills any savings in power costs.  Now prior to 80-Plus program a lot of PSU were junk and the curve was very steep.  If you have a PSU line that has 85% efficiency at 50% load and it collapses to something horribly like 68% efficiency at 90% load the 17% delta in efficiency by is huge.  This is where the "aim for 50% load" rule of thumb came from and while it was good advice at one time, its day has past.

If one wanted to spend more upfront to cut their power bill upgrading to a Platinum PSU and switching to 240V provides a bigger gain.  Platinum PSU is going to be ~3% higher efficiency and 240V is going to improve efficiency by about another 1% to 2%.  For powering a 600W load I would take a highly rated 850W Platinum PSU over a 1250W (or 1500W, 1600W as advocated by some) Gold or lower rated PSU.

That's misleading


Between 50% and 80% maybe but look at the chart references on this thread a few posts back (not by me so I'm not cherry picking). http://www.techpowerup.com/reviews/CoolerMaster/V850/5.html

If you go up to 100% you are down 2% from 40% or 50%, not 1%. But this is again somewhat misleading, see below.


It will save about 2% from running at full load or 1% from running at 80%. Even running at 80% is better than 100% though. If you are pulling 1000W 24/7 and paying 5c/kWh then the difference in power alone is $8/year, ignoring the power supply itself running cooler. If you are paying 15c this is $24/year.

Now when it comes to heat produced within the power supply, that ALL comes from the losses. If a power supply is 90% efficient, that is 10% losses. At 88% efficiency that is 12% losses, so 20% more heat, not just 2%. That's definitely significant to the temperature and cooling of the power supply.

I wouldn't pay a lot more for a bigger power supply but I would definitely pay somewhat more.

I understand what you are saying, but when CYPER starts with a wall wattage and says it is "in within specs" you use the efficiency to work backwards from that to find the DC output wattage and therefore if it is within the 840 listed.  The entire exercise is a "paper" one, but that's what he presented so that's what I gave him back.

And I "grasp" all of this fine, thanks.  You don't seem to grasp what I was trying to convey to CYPER.  Read the prior paragraph a few times and let it sink in.  You are picking an argument where there is none.  I will admit that I did not "show my work" and list in exhaustive detail all the "givens" that went into my statement to your satisfaction since I was just trying to fire off a quick blurb to him, and not you.  But if you want it:  here is more for you.  CYPER listed 900W and said the V850 was in spec.  Because of the higher efficiency, it only showed 900W at the wall whereas a HX850 would have shown slightly more owing to its lower efficiency.  At 900W, at the listed 90% typical and 93% efficiency at fifty percent load, I used the 90% figure to arrive at a value of 810W DC and thus it was in spec at that wall wattage (though barely..I consider a headroom of 30W to be riding the razor's edge at that level).  Obviously, I should have put in a caveat that I would expect the HX to show a higher wall wattage at a similar DC load, and also go into gruesome detail (as I did in the exchange with you later on) about my own observations of *actual* steady sustained and spike power that I observed (although I thought I was alluding to that fact, but perhaps I was too vague).  I did not do that, and I apologize that it confused you.  Better?

The reality is that those machines pulled more at the wall than what he even cited at times, and thus were out of spec on the spikes rendering that hypothetical exchange moot.  Thankfully (for KnC's hosted miners and others) CoolerMaster "V" series units didn't shit themselves and handled the load.  The HX850 units (or at least a number of them) did not conduct themselves in the same stalwart nature.  At the end of the day, as is my only argument all along,  a blanket "850W PSU" recommendation was insufficient and it wound up causing issues.  KnC later fixed it with firmware..yay for them..but the initial information missed the mark.  

Which wasn't the point.  Saying "Of CoolerMaster V850, barely @120V by virtue of higher efficiency" is nonsense.  The efficiency of the power supply has absolutely nothing to do with what DC load it can handle.  Nothing.  The only difference between a 850W 90% efficient unit and a 850W 20% efficient unit is how much additional AC power is used to generate the same amount of DC power.

The DC load isn't going to change, the ability of the PSU to handle that DC load isn't going to change.

You also seem to not grasp that LOWER efficiency means for a given observed AC load the DC load is even lower.   You corrected my use of a (intentionally) high 90% efficiency to point out it was "only 87%" when the 3% lower efficiency only means the DC load was EVEN LOWER for a given observed AC load.  If a unit isn't over spec at 90% efficiency it isn't going to be over spec at 87% efficiency or 1% efficiency.

Awesome  Coolermaster V1000 working great for me as well.  Lots of sad people who took Day 1 and Day 2 deliveries with corsair HX850's were not as pleased with the results.  These days with the new firmware, non-issue thankfully.

BUT WHEN YOU CALCULATE BACKWARDS FROM A WALL WATTAGE TO A DC WATTAGE, YOU USE THE EFFICIENCY RATING TO DO SO AND SINCE CYPER QUOTED A WALL WATTAGE IN HIS STATEMENT OF BEING IN SPEC, IT IS APROPOS TO USE THE EFFICIENCY AS PART OF THE SUPPORTING DATA.  IN ADDITION, IT IS PROBABLE THAT THE COOLERMASTER UNITS HAVE A MARGIN BUILT INTO THEIR 840 RATING WHEREAS THE CORSAIR UNITS OBVIOUSLY DID NOT.

All caps is fun, does that make me as cool as you think you are?

Actually, I have a degree in Engineering and I do technology work for a living...math is very much my strong suit.  I'm telling you that those PSU's (HX850) were shutting off.  I just reported to you that mine ran 950W constantly (bounced between 945 and 954) as I watched it in real time.  I have seen momentary spikes up to about 975W (read 972 briefly when it happened). Since my UPS only seems to report wattage in 9W increments, it very well may have been higher.  I'm pretty sure I don't have a "special unit".  It's obvious (to me, at least, who seems to actually understand these things) that the failure mode related to the HX850 was caused by overcurrent protection kicking in.  This failure mode could have been prevented by buying a 1000W PSU, as many people did (and those who had HX1000 reported no such problems to my knowledge).  Now, you can play armchair engineer all day (as you are apt to do on these threads, and often talk out of your ass though few call you on it) but that is a fact.  That is something that GE themselves verified with a hands-on test.

As for "strange" blah blah blah...um, you don't happen to remember the OTHER datacenter that had numerous boards blow up because they were running HX850's?  Seriously dude, did you hit your head or are you just trying to be a revisionist historian here to serve some sort of agenda?  Look back at the thread...all the info is there in gory detail.  The fact that 1) KnC lucked out by using Coolermaster which seemed to have a bit of headroom in the specs and more tolerant overall and 2) they quickly released firmware which dramatically reduced power requirements has been the reason(s) that they haven't had catastrophic issues.  

And to specifically address the "nonsense" thing...you sure talk with a lot of authority on how these things operate for someone who doesn't even have one.  But I guess it's easier to just theorize rather than deal with actual facts gained from hands-on observations, right? 

EFFICIENCY DOESN'T CHANGE THE DC LOAD.  IT ONLY CHANGES HOW MUCH MORE POWER IS USED ON AC SIDE.

AC LOAD = DC LOAD / EFFICIENCY. 
SO LOWER EFFCIENCY = HIGHER AC LOAD FOR THE SAME DC LOAD.
SO AN 800W DC LOAD is ALWAYS GOING TO BE 800W DC. POWER SUPPLIES ARE RATED BASED ON THE DC LOAD.  

Typo.  I was saying at most the PSU will be 90% efficient.   Still the efficiency curve on a modern "80-Plus" PSU is very very flat, almost a horizontal line.   The difference between 50% load and 80% load is usually 1%.  It isn't going to run significantly cooler, last significantly longer, or save any meaningful amount of power.  Key word meaningful be sure to compare any savings to the cost of oversizing the PSU.  It isn't the 1980s anymore.  PSU have gotten a lot better in the past 3 decades.

RM850 and HX1050 and Seasonic 1050 are all working well for me.

No the best efficiency is near the middle load. Get a bigger power supply than you need. It will save money on energy, run cooler and quieter, and probably last longer. Also if you can run on 240, that gives another few percent usually.

Great, what about all the other brands and models @850W?  KnC never specified V850's..they just told people to get an 850W power supply.  Some of them (HX850) were obviously not up to the task.

Simplistic or not it was accurate.  Most modern PSU provide 90% to 100% on the 12VDC rail because they do conversion of all AC power to 12VDC and then power the other rails off the 12V rail.  

Sorry typo ment to say at worst efficiency was .....  Still if the efficiency was WORSE than 90%, say 88% then that only reduces the relative DC load as it compares to the observed AC load.  850W/0.88 = 956W.  850W/0.85 = 1000W.  You telling me people have reported >950W at the wall?
Nonsense.

So once again 850W is fine.  A single person reported a single problem with a single model of a single brand of power supply and you want to say 850W is insufficient as a blanket statement.  Strange KNC's entire datacenter is powered by nothing but 850W PSUs.  KNC still recommend 850W on the site.  Many people are using 850W PSUs today without issue.  The idea that one needs 1250W or 1600W to power a 600W DC load is well just silly.


Math isn't your strong suit is it?   840W / 0.8773 = 957W.   The LOWER THE EFFICIENCY the higher the observed AC load compared to the actual DC load and power supplies are rated by their OUTPUT not INPUT.  So you are actively disproving your point.  Now if the unit was 99.9% efficient then yes >850W at the wall would indicate the unit was running beyond peak load.  I used 90% because most 80-Plus PSU are within a few % of 90% and certainly don't go much above it on the high end.  Still we can use the lower 87.7% that only means the DC load was EVEN LOWER!   

Bitcoinorama replied to a PM with the following over the weekend re delivery of a Nov order placed now (and whether they were already manufacturing and stockpiling for immediate delivery)

Um, yep....why do you think those PSU's were shutting off?  And your "spec calculation" is over simplistic, btw...just because a PSU is rated for an overall wattage doesn't mean that an individual rail is.  In some cases, the total +12V wattage is less than the overall PSU wattage rating because they were coming in just under the wire, so to speak, to meet a "common" PSU rating cut-off and the lower voltages are necessary to add in to reach the overall number.  For instance, the Corsair HX850 that was the center of the controversy is an 850-rated PSU but only rated a maximum of 840W +12V with apparently little to no safety margin built-in given the observed results.
http://www.corsair.com/en/media/cms/manual/corsair-psu-spec-table-091813.pdf

As for the "at best efficiency"...well, at near 100% load you aren't going to get that, now are you?  Again, according to manufacturer's specs you could expect 87.73% @115V input.  So put the lower actual rated rail wattage coupled even with lower real-world efficiency when running with a wall power reading of over 970W (and yes, those levels were typical prior to version .95 of the firmware for jupiters) and guess what?  You are running that thing over-spec at time of KnC initial release and that's why those things were shutting themselves off and then coming back with a bang in some cases upon recovery.


Others PSU models/brands are actually under-rated in that the combination of the +12V rails actually oversubscribes the overall rating, and the rating itself has a safety margin built-in to it so that the PSU is rated (according to manufacturer documentation) to be able to deliver over the rating (which would be the one you would probably want to buy).

And yes, once again, machines with fully functional chips with initial firmware could very much pull that much wattage..hence the issue.  With .94 firmware mine would peak at 945W long-term sustained at the wall according to my UPS readout with higher spikes depending on what cores were enabled/disabled at the time.  I never ran .90 but I know that rev pulled at least as much. Was I the only one paying attention when all that was going down?  When you actually have one to play with, try putting a pre-.95 firmware on it and let us know how much power you see it pull, k? 

edit:  clarified verbiage to better express thought chain.

Of CoolerMaster V850, barely @120V....and by virtue of higher efficiency than the corsairs would only show that hypothetical 900W (but still only rated to 840W on +12V) versus a higher value that would be shown by the corsairs at same DC output wattage.  But the Jupiter's were pulling 950-ish and more at the wall at release up through .95 firmware being released.  And the 240V value would actually go the other way with the higher efficiency gained by the voltage jump...the wall wattage would show lower at 240V versus 120 at same DC output.  The Corsair HX850's that some people were buying were definitely out-of-spec at the wattage being pulled during initial product release due to a smaller actual +12V rail wattage than the overall PSU (again 840W) with seemingly no safety margin and slightly lower efficiency (indicating lower quality parts as well as more internal heat generation which can have a cascade effect when pushing the components to absolute limits).  Since KnC didn't recommend a specific brand/model...they just said "850"...I have to say that their overall recommendation was faulty and was the cause of several failures.

edited to better show train of logic in statements

That's what all of us with orders are all hoping for.

By rights they ought to be, as it specifies "November shipping" right on the product page.

Of course it's possible for them to ship in November, but not be delivered until December.  Especially if you don't select the express shipping option.  (does anyone actually not select express shipping?)

no one knows, but supposedly november batch shipping will commence nov. 15.