Safe power draw

[elidell]

Hello world.
I recently built an 8 gpu hashing rig (7x gtx 1070 and 1 gtx 1080), on an Octominer motherboard.  It has only 8gb of ram and 1.6gh intel mobil cpu.  Should be fine for hashing..

Having a closest full of power supplies ranging from 750w - 1400w I was fairly certain i could do this with one psu.  I didn't go with the 1400w only because it is a server psu that is super loud.  Instead i went with a 1200w thermaltake that is super quiet.  

My initial test on wpa hashes just running a wordlist with no rules, i was only drawing about 750 watts, so pretty good, but when i applied a rule set, I saw it jump up to 1200watts draw from the wall. (i have a watt meter between the psu and wall outlet)

I was able to put power limits on my Nvidia cards to a max 125 watts with little to no performance difference (meaning it took about the same amount of time to exhaust the list) and this brought my power consumption down to between 1000 and 900 watts.

My question is what would be ideal as far as the limits pf the psu and my total draw?  Is it safe to run the psu at 1200 watts draw?  What is the safe percentage of total capability to actual use of the PSU in question?  How much of an effect does power draw have on performance?

[soxrok2212]

GTX 1070 is rated at 150W while the 1080 is 180W. That’s a total of 1230W JUST for GPUs. Typically, I shoot to have a few hundred extra watts of headroom.

Further, noise should be the smallest concern when building a rig for hashcat. Top priority should be cooling and making sure you have appropriate parts. You’d *probably* be safe with a 1600W server grade PSU, but I wouldn’t overclock and even this is really pushing the limits.

Having a weak PSU can not only degrade performance, but it may eventually damage your hardware too.

[undeath]

It has only 8gb of ram

I'm surprised hashcat even worked with that. Usually you should have at least ram=vram.

[elidell]

It has only 8gb of ram

I'm surprised hashcat even worked with that. Usually you should have at least ram=vram.

why would i need anymore than 8gb of ram on the motherboard?  its literally just running hashcat on ubuntu in headless mode?

[elidell]

GTX 1070 is rated at 150W while the 1080 is 180W. That’s a total of 1230W JUST for GPUs. Typically, I shoot to have a few hundred extra watts of headroom.

Further, noise should be the smallest concern when building a rig for hashcat. Top priority should be cooling and making sure you have appropriate parts. You’d *probably* be safe with a 1600W server grade PSU, but I wouldn’t overclock and even this is really pushing the limits.

Having a weak PSU can not only degrade performance, but it may eventually damage your hardware too.

right, i agree, but i can easly set the power limit to 125 or even 130 on the cards and i stay about 250 watts below the capacity of the PSU.  I ran a few bench marks and noticed very little difference difference with the power limit in place..

WPA/WPA2 benchmark:
No Power limit: 2230.2 kH/s
Power limit set to 125 on all cards: 2063 kH/s
Power limit set to 135 on all cards: 2107 kH/s



so that should suffice no?

Regarding noise, i don't want the wife to complain when im hashing for any considerable amount of time.

[undeath]

why would i need anymore than 8gb of ram on the motherboard?  its literally just running hashcat on ubuntu in headless mode?

afaik the nvidia drivers have some weird behaviour where they map all the vram used by hashcat to your system ram.

btw, the true power draw test is probably running -m1000 -a3 ?a?a?a?a?a?a?a?a?a?a -w4 -O

[epixoip]

It's not weird behavior and it's not specific to Nvidia by any means. You need host buffers for the device buffers. When you allocate a buffer on the device, you have to allocate a buffer on the host as well. Otherwise how else would you get the results? If you have 8GB of VRAM per device, you could conceivably allocate 100% of that (yes, a misinterpretation of the OpenCL spec results in most vendors limiting each allocation to 25% of VRAM, but you can in fact allocate multiple buffers.) If you're allocating 8GB of buffers on each device, and you have eight devices, you could conceivably need to allocate 64GB of buffers on the host as well. If the host cannot allocate 64GB of RAM for Hashcat, it will crash with CL_OUT_OF_HOST_MEMORY. At an absolute minimum available host memory should never be less than VRAM * 0.25, but the rule of thumb is RAM >= VRAM.

Keep in mind that password cracking is not cryptocurrency mining, and most mining rig setups suck for password cracking.

[epixoip]

Oh sorry, I didn't actually address your question regarding the power supply output. So you have a 1200W PSU and are concerned about it drawing 1200W from the wall. Ok, let's talk this through.

First, the power rating on a power supply is the output power, not the input power. If a power supply is rated at 1200W, then it can output at least 1200W reliably. In other words, you can safely push a quality power supply at 100% of its rated output. Problems generally occur when you demand more of the power supply than it is capable of. For example, your 1200W PSU may be able to push 1350W for a few hours before exploding.

Second, power supplies cannot transform and output 100% of their input power. This is why we have the "80plus" efficiency rating system - an 80plus certified PSU can convert a minimum of 80% of input power to output power. And of course some power supplies are more efficient than others; this is why there are different tiers of the 80plus certification. You can find more about that here: https://en.wikipedia.org/wiki/80_Plus#Ef...ifications

You didn't specify which Thermaltake 1200W PSU you have, and Thermaltake of course has many different 1200W PSUs of various efficiency: https://www.plugloadsolutions.com/80Plus...=71&type=2



From this chart it should be immediately obvious that power supplies are most efficient at 50% load, and efficiency rolls off sharply at 100% load. So let's assume you have one of the 80Plus Gold units that are around 88% efficient at 100% load. And you stated that you measured 1200W on the input side, which means you're probably around 90% load so we can use the 100% load value of 88% efficient. If you're only able to transform 88% of your 1200W input power, then you're only pushing ~1056W on the output side - well below the PSU's 1200W rating.

Is it safe? Yes - but your electric bill will feel the pinch. You would certainly benefit from a larger, more efficient power supply.

[elidell]

Thanks you very much.  That is super helpful.  I am actually using the TOUGHPOWER iRGB 80 Plus Platinum 1200w.

So what you are getting at, is that drawing 1200w from the wall, i am getting about 86% of that, and outputting about 1032 w.  Idle in headless mode i am hovering around 80w so we can conservatively guess my GPUs are consuming about 900 watts when Im pulling 1200w in? 

That's probably around 120w / gpu (average) when im not placing a power limit on them, im guessing.  Were you saying earlier that this is bad for the cards? can you explain that part?

[elidell]

I guess my biggest problem is ram right now and i actually double checked. I actually only have 4g of physical ram and an 8g swap file. Though it seems to be fine for my general use case right now, I have noticed that it takes a while to initialize even a small word list, i assume that's because of ram? my board only has a single slot for ddr laptop ram. I can up it to 16 while would be better, but i could never get it to 64 unless i increase my swap drastically, but i don't really think that will help