The Tower of Prisma
Solved a lot cooling issues this way and freed up some rack real estate in the process.
Mounted this way the boards are running about 60F cooler.
On to OC'ing them.
Sorry about the blur, was standing on a chair and I'm an old fart.
You said in this configuration they are running 60F cooler? What were your numbers before and after? My setup looks very similar with the box fan under it. Plus I have two fans on each prisma, the one stock drawing air in and the one philipma1957 suggested to pull air http://www.amazon.com/gp/product/B00A460TK6/ref=oh_aui_detailpage_o01_s00?ie=UTF8&psc=1 Those Silverstone Tek 140mm x 38mm Fans are really quite even in power mode. I have them all hooked up to a fan controller that GrapeApe recommended http://www.amazon.com/gp/product/B00DN3IT7M/ref=oh_aui_detailpage_o02_s01?ie=UTF8&psc=1 I have them all running at full power, the stock fans are loud, like a hairdryer as some people say. But if that setup you have going is actually running 60F cooler and the stock fans are cycling you just maybe on to something. With my setup minera is reporting 85-86F running @ 240mhz. Im just curious what your numbers were before and after. Also im not sure where minera is getting the temp numbers from, does anyone know? I saw a post on this thread where someone said it could be the RPI temp, so I don't have a clue. Now I have to go give the leaning tower of prismas a try, will report back. Also im a little worried of them tipping/falling over, so maybe secure them somehow from the top. Without getting all esoteric, horizontal the back bracket was typically 176F @ stock clock (240) and unit was sporadically rebooting. Mounted vertically as in the photo with the box fan below on high (~2000-2500 CFM) the back bracket is is a chilly 110F-112F. There has always been a 50F delta T between the front chips and the back chips. And that Delta remains in this config. Ambient air temp oscillates between 70F and 80F due to the powered roof vents cycling. And air is drawn in from the outside (4F-20F air) through a screen door. I'm almost 100% sure that the temp reported on the dashboard is the temp near one of the vertical connectors on the Pi. We used some beefy zip ties to hold it to the wire mesh. The first time the fan cycled it scared the shit outta' me cuz' it got real quiet and I was sure I blew sumpin' up. Yes they are louder than hair dryers and they are the loudest miner in the DC. A little birdy whispered in my ear and told me that 70% of the heat dissipation in a Prisma is through the heat sinks the other 30% is out the component side of the board. Be back l8r with pics. |
|
|
|
|
Since going vertical, up-time on both units is > 11 hours with no spurious reboots.
The RPi controlled Prisma is OC'ed to 250, consistently > 1400 GH/s.
The BE controlled Prisma is OC'ed to 260, consistently > 1500 GH/s.
We have a BE controlled 24x Tube verticalized that is OC'ed to 280, consistently > 860 GH/s.
We have covered/blocked the square hole on the exhaust side of the Prisma's and 24x Tube.
Because we noticed a lot of cold air escaping through the hole.
Blocking the hole forces that cold air to flow across the internal heat sinks providing an additional measure of cooling.
Initially, the blockage material was a bunched up paper towels.
Now the blockage material is aluminum foil tape.
The type of tape one uses to seal wood stove flu's.
Will take some pic's later today (non-blurry ones) and post links.
|
|
|
|
|
FYI, if you get a Prisma cold enough the fan will cycle.
|
|
|
|
The Tower of Prisma Solved a lot cooling issues this way and freed up some rack real estate in the process. Mounted this way the boards are running about 60F cooler. On to OC'ing them. Sorry about the blur, was standing on a chair and I'm an old fart.  |
|
|
|
I'm gonna be setting up some Prismas to host in the next day or two, probably on DPS2K PSUs. Definitely not in load-balanced, which shouldn't be necessary since these boards all have isolated 12V lines. A pair of Prismas shouldn't quite max out a DPS2K, maybe 110% rated. It wouldn't take much of a short to run up to OCP (about 120% rated). Hopefully we don't test that hypothesis with a "live-fire" experiment.
Thanks for the info on what's been going on.
Your calcs are correct if the Prisma's draw as advertised. They don't. @ 240 clock rate ours (2x) are drawing ~1250W each. When OC'd to 250 they draw 1300W each. When OC'd to 270 they draw ~1400W each. We are powering one with a DPS1600BB (using your most excellent adapter board) and the other with 2x HP 1000's. Both are being feed by 8 16ga PCie cables because in our initial test using 4 cables they got more than warm to the touch. |
|
|
|
Thumbs up to that Phil.  We need a steady growth of BTC value finally. mine should come today. will photo setup .
very happy with the btc price trend we are near 400usd fingers crossed we stay around 400-450 usd a coin for the next few weeks then jump to 450 -500 usd near thanksgiving.
With a continued decrease in % of network difficulty increase. |
|
|
|
I've received a photo of the adapter that has been created for connecting to a RPi. The interface is USB, so that's good news.  I will ship overseas, but I can not estimate a price as I don't have package dimensions. The weight will be somewhere around 8kg. Hey CrazyGuy (or anyone else with a clue), I assume (we both know what that means) the signals used are RX, TX, GND. Looking at the picture, what are the left most and right most pins on the little white connector? In a moment of weakness, I bought a CP2102 USB-UART adapter off of Amazon for ~$9 ( http://www.amazon.com/gp/product/B00CD264HG/ref=oh_aui_detailpage_o00_s00?ie=UTF8&psc=1 ). Hang is it has 6 pins (fortunately they're labeled on the PCB) if I knew the pin out on that board I can jumper the correct pins on the one in hand. This would give all concerned a second source for the USB-UART boards. Thanks bro' |
|
|
|
Cover off of the BE with 40mm fan blowing 70 degree air on it, declocked to 230. This time it made it about 1 hour and 10 minutes. This is beginning to piss me off.  |
|
|
|
@PlanetCrypto
Everything you are describing is what I was experiencing even the fake hashing. I find that the BEController is very sensitive and it is the culprit here. I moved the controller to a COLD SPOT and problems disappeared for me. Put it somewhere cool. The EVGA 1300 G2 is just fine with the cable configuration talked about several pages back it's how I'm powering all 3 of my Prismas.
I've been hashing now at 98% of expected for over 36 hours no problems now whatsoever. HWe is 0.33%
So the hypothesis is that with the extra board count attached to the BE (versus a single 24x Tube) causes it overheat and freak out. I just took it out of case and put a 40mm fan over it. Am and have been a big fan of the 1300's, they're just a little pricey compared to a 1800W-2500W server supply. We have 5 or 6 1300's here in our environment, they whirr quietly and do their job. 98% of expected @ what clock rate? |
|
|
|
Have been arguing with 2x Prisma's connected to a BE.
Can't wait for those USB-UART adapter(s) to show up and get rid of this rinky-dink BE controller.
Usually 2 reboots (hard or soft) are required for the BE to recognize all the boards.
Changing ASIC Section WakeUp period from the stock 3600 to 7200 helped immensely.
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s
All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out.
Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each.
The PCIe cables were slightly more than warm to the touch, regardless.
Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma.
Data Center Temp is 70F-80F.
The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing.
Hash rate indicated on the machine is normal, but pool reports 0.
Worker diff currently locked to 512. Testing.
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out. Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each. The PCIe cables were slightly more than warm to the touch, regardless. Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma. Data Center Temp is 70F-80F. The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing. Hash rate indicated on the machine is normal, but pool reports 0. Worker diff currently locked to 512. Testing. drop the clock. 250 is too high try 230 . I found the short tubes were heat sensitive and preferred underclocking handles out and 2 fans not one. also try hashing them at mmpool.org my short tubes ran really well at that pool. Roger that. Declocked to 230, now let's see if it runs more than an hour. Do you have a spec/model # of a suitable/matching fan?Have measured the chip temps with IR and while there is a difference between the front chips and the back chips the delta T isn't that great. Tj is probably within operating specs. |
|
|
|
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s
All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out.
Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each.
The PCIe cables were slightly more than warm to the touch, regardless.
Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma.
Data Center Temp is 70F-80F.
The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing.
Hash rate indicated on the machine is normal, but pool reports 0.
Worker diff currently locked to 512. Testing.
Thanks for the Kill-A-Watt data, I appreciate it! In our experience the EVGA 1300's pickup a few percent improvement in efficiency when run off of 240V. Running off of 120V they are ~90%-92%, and typically ~92%-95% off of 240V. Was trying to make the point that OC'ing a Prisma with an EVGA 1300 feed by 120V is a recipe for burning up a $200 power supply. They are a beefy PC supply, but would imagine they will be short lived if used to power an OC'ed Prisma for any length of time (i.e. days). A warm environment would exasperate this situation. |
|
|
|
 When this happens the hash rate listed in the BE is normal (It thinks it's hashing). A reset or 2 causes it to start hashing @ the pool again. All of our other of our miners act normally when this happens (So it's NOT a networking issue). The is a "known good" BE as it was used flawlessly on a 24x tube for months. I'm scratching my head on this one. But re-booting every hour is flat unacceptable. |
|
|
|
Have been arguing with 2x Prisma's connected to a BE.
Can't wait for those USB-UART adapter(s) to show up and get rid of this rinky-dink BE controller.
Usually 2 reboots (hard or soft) are required for the BE to recognize all the boards.
Changing ASIC Section WakeUp period from the stock 3600 to 7200 helped immensely.
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s
All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out.
Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each.
The PCIe cables were slightly more than warm to the touch, regardless.
Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma.
Data Center Temp is 70F-80F.
The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing.
Hash rate indicated on the machine is normal, but pool reports 0.
Worker diff currently locked to 512. Testing.
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out. Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each. The PCIe cables were slightly more than warm to the touch, regardless. Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma. Data Center Temp is 70F-80F. The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing. Hash rate indicated on the machine is normal, but pool reports 0. Worker diff currently locked to 512. Testing. |
|
|
|
|
Have been arguing with 2x Prisma's connected to a BE.
Can't wait for those USB-UART adapter(s) to show up and get rid of this rinky-dink BE controller.
Usually 2 reboots (hard or soft) are required for the BE to recognize all the boards.
Changing ASIC Section WakeUp period from the stock 3600 to 7200 helped immensely.
Clocking @ 250, handles in, powered by DPS-1600BB's, 8 PCIe plugs per Prisma, Real Perf Hash Rate = 2885 GH/s, Hashing to BAN-Hash rate is ~2750 GH/s
All the screws on the hash boards were loose, some more than others, some so loose they were about to fall out.
Initially powered them with 2x EVGA 1300's feed with 120V and 4 PCIe cables per Prisma, Kill-A-Watt indicated @ 240 they were drawing 1250W each, @ 270 they were drawing 1350W each.
The PCIe cables were slightly more than warm to the touch, regardless.
Hence the shift to DPS-1600BB's which can supply 1800W each and 8 PCIe's per Prisma.
Data Center Temp is 70F-80F.
The BE also seems to get confused after an hour or so if worker difficulty is greater than 1024 and stops hashing.
Hash rate indicated on the machine is normal, but pool reports 0.
Worker diff currently locked to 512. Testing.
|
|
|
|
We looked at using those Hydronic radiator cores, and opted instead for mono extruded finned tube ( https://www.finnedtube.com/extruded-fin-tubes) with "Turbolators" from MDE. For a plethora of reasons. The tube from MDE can be ordered to spec. As a hint, temperature cycling of the cores causes the aluminum fins to loosen causing poor thermal conductivity to/from the copper tube (as AL expands and contracts at a different rate than copper). Corrosion due to galvanic action between 2 dissimilar metals also becomes an issue over time. Standard Hydronic Circulators (which by the way are NOT Pumps even though they are referred to as such) can be ordered from Supplyhouse.com (formerely known as PEX Supply). We prefer the Taco brand circulators with the built in check valves (best bang for buck) and can be found here http://www.supplyhouse.com/Taco-Cast-Iron-Pumps-289000 . Flow rate(s) of the pump(s) is dependent on a great many things from a system perspective and is way beyond detailing in this venue. Suffice to say a solid competency in thermo-dynamics and hydro-dynamics is a minimum requirement. Using BTC Miners as a water heater/boiler is a 3 loop system. Loop 1 circulates fluid through the condenser coil in the sealed tank, Loop 2 circulates fluid, on thermostatic demand, to the PEX imbedded in the floor, and Loop 3 is a heat dump loop that circulates fluid to something to dump excess/waste heat when tank temperature exceeds design specs (i.e. this loop maintains a delta-T sufficient to maintain Novec vapor condensation and is dependent on the particular fluid 7000, 7100, 7200, 7300 or a blend thereof's boiling point that maintains the desired chip junction temperature Tj). All loops are feed from and discharge to a heat reservoir tank of sufficient size to buffer sudden changes in fluid temperature and account for fluid expansion/contraction. We use a 50/50 mixture of ethylene glycol and distilled/de-ionized water as the heat transfer fluid. Our systems run @ atmospheric pressure. If the system is designed correctly one can, lower chip Tj (thereby facilitating severe OC'ing), optimize transfer fluid temperature for comfort heating, and create an extremely reliable system that can operate continuously for years unattended. I was thinking about making something to transfer heat from miners, it may not be as complex, but by using scrap hydronic heating tubes (copper pipes with aluminum fins) and a pump I could make a system to transfer heat to another area. The aluminum fins are supposed to disperse heat from water that is inside of the copper pipe. Basically, instead of using a boiler, you use bitcoin miners. I would imagine you would need a pump that could withstand the heat, and also you would not want to build up too much pressure. Thoughts?  bobsaq3 and sidehack are spot on at stock/factory clock rates. We prefer to allow a little more head room for OC'ing the Prisma's and hence have migrated to one DPS-1600BB per Prisma. OC'ing a Prisma only makes financial sense if the cost per Kwh is low ($0.08/KWh or less). Then there are the heat dissipation issues that come with filling a room/Data Center with OC'ed Prisma's. Pursuant to that we are prototyping Open Bath Immersion (OBI) cooling using the 3M Novec Engineered Fluids in custom built polycarbonate immersion tanks. With the ultimate goal of immersing 32 OC'ed Prisma hash boards, controlled by an RPi w/wireless, in an OBI tank powered by 5 DPS-2000's. Power consumption should be ~10Kw per tank and dissipate ~32K BTU continuously, enough to heat the average American sized home in a Minnesota winter AND GET PAID TO DO IT. The residual/excess heat could be used to heat a garage, heat a driveway, heat sidewalks, melt snow off of the roof, heat a pool/hot tub/fish pond, etc. . . . . . . Our Youtube channel (PlanetCrypto) has some videos of our initial testing. Initial testing of 24x Tube Erupter boards in a Walmart fishtank: http://www.youtube.com/watch?v=meOf0FJBkGAhttp://www.youtube.com/watch?v=y_yfHJ8yoh8A polycarbonate immersion tank with 2x Bitmain S3+ hash boards: http://www.youtube.com/watch?v=gcb9TyQP5ZAhttp://www.youtube.com/watch?v=fStHV_Q-x7sWe no longer power the fan in the S3+ tank as it is not required for condensation. The heat generated from the S3+ tank supplements the in-floor heat in our little DC. We have laid PEX in the ground outside the DC as a place to dump excess heat in upcoming expanded systems. When we get a 32 Prisma board system up and running we'll post some vid's on the channel. Awesome videos. If you flipped the 24x boards over it might look cooler since they were designed to push heat through the pcb. You would get lots of tiny bubbles instead of the big eruptions through the mounting holes. It might not be more efficient, but I think it would be cooler to look at. I want to build an immersion tank as well, but lack of funds is preventing me from doing so. I do have several design ideas though. I know it was a test tank, but if you added a second radiator to the fish tank in a V formation you would get better condensation. You don't want to use fans in the tank because the name of the game is less energy used to cool the miners. Wow, sorry for sounding like a prick. I've spent a lot of time researching and designing a few different versions. It's nice to see someone putting a system together. Yes, I am jealous. Heating a house with miners can definitely happen. It got to a frigid 40 degrees here last night in north Florida. I have 6 S1's and 2 S3's, with a box fan and relocating 2 miners the coldest part of the house was 68. The room where most of the miners were was 82. My central heat isn't working right now, so that was the miners keeping it warm in the house. The tube Erupter tank is disassembled, we were going to jump on those boards but the Prisma's came out (shout out to Crazy Guy for giving us the heads up). Had lots of ideas too when we got into immersion cooling. Thank the Bitcoin Gods we spent a day with Phil Tuma @ 3M. Phil very kindly told us/me why a lot of our/my ideas wouldn't work (mostly cuz' he had already tried them @ 3M's expense). Saved us a BUNCH of time and money not exploring crappy ideas. Not saying any of your ideas are crappy, just my hair brained ideas. V'ed coolers take up too much room in the tank versus mono extruded fin high thermal conductivity copper tube with Turbolators ( https://www.finnedtube.com/extruded-fin-tubes & https://www.finnedtube.com/turbulators). Fans don't survive cuz' the Novec fluid dries out the bearings and they stop spinning anyway. Fans also dilute the vapor layer and consequently require larger condensation surface area/lower cooling fluid temperatures to condense the less dense vapor. PVC coated wire leeches DOP (dioctyl phthalate) into the fluid which deposits itself on hot components (chips and copper back planes) and reduces thermal conductivity on the parts that really need it ( http://en.wikipedia.org/wiki/Bis(2-ethylhexyl)_phthalate) PTFE coated wire doesn't exhibit this problem but it's expensive as hell. Unless the tank is purged of air, eventually, the water in the air will condense out, corrode metal components/circuit traces/solder joints and cause "ghosts in the machine" (bags of silica gel also solve this issue). The 24x Tubes have a higher board chip density than the S3's. Bitmain wouldn't sell us chips for our own high density board design (guess they were jealous). Evidently somebody (friedcat) was listening cuz' the Prisma's have a much higher density than anything else out there on the market (big shout out to friedcat, thanks o brilliant one). And in the immersion cooling world, board chip density is where it's at. Suffice to say there are lots of hidden non-intuitive gotcha's in OBI cooling. But once over the learning curve it works like a dream and makes managing heat dissipation a breeze (pun intended). Besides it looks cool as hell in a clear polycarbonate tank. LOL. We've had visitors pull up a chair and sit for hours watching the bubbles, kinda' like washing machine TV. |
|
|
|
bobsaq3 and sidehack are spot on at stock/factory clock rates. We prefer to allow a little more head room for OC'ing the Prisma's and hence have migrated to one DPS-1600BB per Prisma. OC'ing a Prisma only makes financial sense if the cost per Kwh is low ($0.08/KWh or less). Then there are the heat dissipation issues that come with filling a room/Data Center with OC'ed Prisma's. Pursuant to that we are prototyping Open Bath Immersion (OBI) cooling using the 3M Novec Engineered Fluids in custom built polycarbonate immersion tanks. With the ultimate goal of immersing 32 OC'ed Prisma hash boards, controlled by an RPi w/wireless, in an OBI tank powered by 5 DPS-2000's. Power consumption should be ~10Kw per tank and dissipate ~32K BTU continuously, enough to heat the average American sized home in a Minnesota winter AND GET PAID TO DO IT. The residual/excess heat could be used to heat a garage, heat a driveway, heat sidewalks, melt snow off of the roof, heat a pool/hot tub/fish pond, etc. . . . . . . Our Youtube channel (PlanetCrypto) has some videos of our initial testing. Initial testing of 24x Tube Erupter boards in a Walmart fishtank: http://www.youtube.com/watch?v=meOf0FJBkGAhttp://www.youtube.com/watch?v=y_yfHJ8yoh8A polycarbonate immersion tank with 2x Bitmain S3+ hash boards: http://www.youtube.com/watch?v=gcb9TyQP5ZAhttp://www.youtube.com/watch?v=fStHV_Q-x7sWe no longer power the fan in the S3+ tank as it is not required for condensation. The heat generated from the S3+ tank supplements the in-floor heat in our little DC. We have laid PEX in the ground outside the DC as a place to dump excess heat in upcoming expanded systems. When we get a 32 Prisma board system up and running we'll post some vid's on the channel. Awesome videos. If you flipped the 24x boards over it might look cooler since they were designed to push heat through the pcb. You would get lots of tiny bubbles instead of the big eruptions through the mounting holes. It might not be more efficient, but I think it would be cooler to look at. I want to build an immersion tank as well, but lack of funds is preventing me from doing so. I do have several design ideas though. I know it was a test tank, but if you added a second radiator to the fish tank in a V formation you would get better condensation. You don't want to use fans in the tank because the name of the game is less energy used to cool the miners. Wow, sorry for sounding like a prick. I've spent a lot of time researching and designing a few different versions. It's nice to see someone putting a system together. Yes, I am jealous. Heating a house with miners can definitely happen. It got to a frigid 40 degrees here last night in north Florida. I have 6 S1's and 2 S3's, with a box fan and relocating 2 miners the coldest part of the house was 68. The room where most of the miners were was 82. My central heat isn't working right now, so that was the miners keeping it warm in the house. The tube Erupter tank is disassembled, we were going to jump on those boards but the Prisma's came out (shout out to Crazy Guy for giving us the heads up). Had lots of ideas too when we got into immersion cooling. Thank the Bitcoin Gods we spent a day with Phil Tuma @ 3M. Phil very kindly told us/me why a lot of our/my ideas wouldn't work (mostly cuz' he had already tried them @ 3M's expense). Saved us a BUNCH of time and money not exploring crappy ideas. Not saying any of your ideas are crappy, just my hair brained ideas. V'ed coolers take up too much room in the tank versus mono extruded fin high thermal conductivity copper tube with Turbolators ( https://www.finnedtube.com/extruded-fin-tubes & https://www.finnedtube.com/turbulators). Fans don't survive cuz' the Novec fluid dries out the bearings and they stop spinning anyway. Fans also dilute the vapor layer and consequently require larger condensation surface area/lower cooling fluid temperatures to condense the less dense vapor. PVC coated wire leeches DOP (dioctyl phthalate) into the fluid which deposits itself on hot components (chips and copper back planes) and reduces thermal conductivity on the parts that really need it ( http://en.wikipedia.org/wiki/Bis(2-ethylhexyl)_phthalate) PTFE coated wire doesn't exhibit this problem but it's expensive as hell. Unless the tank is purged of air, eventually, the water in the air will condense out, corrode metal components/circuit traces/solder joints and cause "ghosts in the machine" (bags of silica gel also solve this issue). The 24x Tubes have a higher board chip density than the S3's. Bitmain wouldn't sell us chips for our own high density board design (guess they were jealous). Evidently somebody (friedcat) was listening cuz' the Prisma's have a much higher density than anything else out there on the market (big shout out to friedcat, thanks o brilliant one). And in the immersion cooling world, board chip density is where it's at. Suffice to say there are lots of hidden non-intuitive gotcha's in OBI cooling. But once over the learning curve it works like a dream and makes managing heat dissipation a breeze (pun intended). Besides it looks cool as hell in a clear polycarbonate tank. LOL. We've had visitors pull up a chair and sit for hours watching the bubbles, kinda' like washing machine TV. |
|
|
|
would it be safe to do that mod on the dps2000 and run two prismas?
You shouldn't have a problem. The DPS1200s power 1 Prisma each without a problem. as well as 2x Dell 750 bobsaq3 and sidehack are spot on at stock/factory clock rates. We prefer to allow a little more head room for OC'ing the Prisma's and hence have migrated to one DPS-1600BB per Prisma. OC'ing a Prisma only makes financial sense if the cost per Kwh is low ($0.08/KWh or less). Then there are the heat dissipation issues that come with filling a room/Data Center with OC'ed Prisma's. Pursuant to that we are prototyping Open Bath Immersion (OBI) cooling using the 3M Novec Engineered Fluids in custom built polycarbonate immersion tanks. With the ultimate goal of immersing 32 OC'ed Prisma hash boards, controlled by an RPi w/wireless, in an OBI tank powered by 5 DPS-2000's. Power consumption should be ~10Kw per tank and dissipate ~32K BTU continuously, enough to heat the average American sized home in a Minnesota winter AND GET PAID TO DO IT. The residual/excess heat could be used to heat a garage, heat a driveway, heat sidewalks, melt snow off of the roof, heat a pool/hot tub/fish pond, etc. . . . . . . Our Youtube channel (PlanetCrypto) has some videos of our initial testing. Initial testing of 24x Tube Erupter boards in a Walmart fishtank: http://www.youtube.com/watch?v=meOf0FJBkGAhttp://www.youtube.com/watch?v=y_yfHJ8yoh8A polycarbonate immersion tank with 2x Bitmain S3+ hash boards: http://www.youtube.com/watch?v=gcb9TyQP5ZAhttp://www.youtube.com/watch?v=fStHV_Q-x7sWe no longer power the fan in the S3+ tank as it is not required for condensation. The heat generated from the S3+ tank supplements the in-floor heat in our little DC. We have laid PEX in the ground outside the DC as a place to dump excess heat in upcoming expanded systems. When we get a 32 Prisma board system up and running we'll post some vid's on the channel. Great videos, thanks for sharing. Thats a great setup you got going there. Awsome way to heat your house in Minnesota. Just out of curiosity what was your source of heat before? Im sure its cold there now and just wondering how the heat from your mining setup is going and if you had to turn the old way to heat the house on yet? Also I would like to know your energy savings compared to last year, or whenever before you started that setup. Really looking forward to the Prisma videos, thanks for sharing. Have heated with everything under the sun (fuel oil, propane, electric base board, wood, wood pellets, Bitmain S1's (currently), immersion cooled Prisma's (future). January heating bills: Electric baseboard 3 years ago = $550/month (electrical cost per KWh has risen since then). Wood pellets = $110-$120/month S1's = positive cash flow (i.e they generate more revenue than they use in electricity, breakeven on the S1's is USD/BTC ~$300 @ current difficulty @ $0.08/KWh). As a heating device I'm allowed to run these off of electricity billed at the "Dual Fuel" rate (>$0.08/KWh) versus the "Regular Service" rate ($0.1168/KWh). Have friends serviced by a different electric company that run their S1's at the "Off Peak" rate ($0.04/KWh). During the day they open windows to exhaust the excess heat and keep the house at a comfy 80F. At night they close most of the windows to maintain the desired temp. 7 S1's are currently providing a 50F+ delta T from outside temp (~30F) with one 20" box fan exhausting heat from a second floor window. And the pellet stove/furnace hasn't kicked on yet. As the temps continue to drop (it gets to -30F+ here in the winter), I'll turn off the fan and close the window, then move an appropriate number of S1's into the house from the DC to compensate (i.e create a larger Delta T from outside temp). Going forward, the Prisma's are the heater of choice as their $/GH/s is far better than the S1's/S3's. |
|
|
|
would it be safe to do that mod on the dps2000 and run two prismas?
You shouldn't have a problem. The DPS1200s power 1 Prisma each without a problem. as well as 2x Dell 750 bobsaq3 and sidehack are spot on at stock/factory clock rates. We prefer to allow a little more head room for OC'ing the Prisma's and hence have migrated to one DPS-1600BB per Prisma. OC'ing a Prisma only makes financial sense if the cost per Kwh is low ($0.08/KWh or less). Then there are the heat dissipation issues that come with filling a room/Data Center with OC'ed Prisma's. Pursuant to that we are prototyping Open Bath Immersion (OBI) cooling using the 3M Novec Engineered Fluids in custom built polycarbonate immersion tanks. With the ultimate goal of immersing 32 OC'ed Prisma hash boards, controlled by an RPi w/wireless, in an OBI tank powered by 5 DPS-2000's. Power consumption should be ~10Kw per tank and dissipate ~32K BTU continuously, enough to heat the average American sized home in a Minnesota winter AND GET PAID TO DO IT. The residual/excess heat could be used to heat a garage, heat a driveway, heat sidewalks, melt snow off of the roof, heat a pool/hot tub/fish pond, etc. . . . . . . Our Youtube channel (PlanetCrypto) has some videos of our initial testing. Initial testing of 24x Tube Erupter boards in a Walmart fishtank: http://www.youtube.com/watch?v=meOf0FJBkGAhttp://www.youtube.com/watch?v=y_yfHJ8yoh8A polycarbonate immersion tank with 2x Bitmain S3+ hash boards: http://www.youtube.com/watch?v=gcb9TyQP5ZAhttp://www.youtube.com/watch?v=fStHV_Q-x7sWe no longer power the fan in the S3+ tank as it is not required for condensation. The heat generated from the S3+ tank supplements the in-floor heat in our little DC. We have laid PEX in the ground outside the DC as a place to dump excess heat in upcoming expanded systems. When we get a 32 Prisma board system up and running we'll post some vid's on the channel. |
|
|
|
|
Why Waltz when you can Rock 'N' Roll.
Saw a bunch of folks complaining about PSU costs and thought I'd stick my $.02 worth in.
From a post I made recently in the Re: [PRICE REDUCED] DPS-2000BB 2000W Server PSU Interface Board Thread here in Bitcointalk:
"FYI, as far as as I can tell from looking at the pin outs, the DPS-1600BB (147.5A @ 12.2V 1800W), DPS-2000BB (164A @ 12.2V 2000W), and the DPS-2500BB (205A @ 12.2V 2500W) are pin compatible and the DPS2K adapter board and should be a "Plug 'N' Play" on these supply's.
Just ordered qty 16 DPS-1600BB's and when they come in will test and report back.
We scarfed these in the salvage market for $8.74 each + $10 shipping.
Total cost each was $18.74. Helluva' supply for $20.
These supply's have NO internal fans and in a stock configuration (i.e. installed in an IBM BladeCenter) are cooled by case air.
Adding 40mm fans to the side of the power supply case for cooling is MANDATORY.
We anticipate using these in our mining farm to power OC'ed AM Prisma's.
One supply per Prisma.
Of other interest is, the efficiency of these supplies under light loading (250W draw) is 87%, at or near rated output the efficiency rises to 92%.
So from a Bitcoin mining perspective it's better to run them WOT (Wide Open Throttle) than to be conservative.
Individuals in the RC community have reported about the DPS-2000BB's that when tweaking the output voltage to 14.6 and tweaking the Over Power Cutoff the DPS-2000's can supply 2800W (~200A @ 14.6V), Booyah!"
We own and use a variety of PC power supplies in our environment. We favor the EVGA 1300's (3x OC'd S3+'s per unit) efficiency is ~92% and they are silent.
Problem is they are pricey as hell.
So a while back we started rotating to Server based power supplies.
Most notably the HP 1200's, IBM DPS-1600's, and the IBM DPS-2000's.
And as a footnote the IBM DPS series of supplies will do current sharing/load balancing across multiple supplies.
So if 1800W/2000W isn't enough, wire them up in parallel and connect the current sharing pins.
|
|
|
|
This particular PSU was used in IBM BladeCenters. I think each 6U machine had 4 PSU in it. Not sure if they had any other purposes from the start.
FYI, as far as as I can tell from looking at the pin outs, the DPS-1600BB (147.5A @ 12.2V 1800W), DPS-2000BB (164A @ 12.2V 2000W), and the DPS-2500BB (205A @ 12.2V 2500W) are pin compatible and the DPS2K adapter board and should be a "Plug 'N' Play" on these supply's. Just ordered qty 16 DPS-1600BB's and when they come in will test and report back. We scarfed these in the salvage market for $8.74 each + $10 shipping. Total cost each was $18.74. Helluva' supply for $20. These supply's have NO internal fans and in a stock configuration (i.e. installed in an IBM BladeCenter) are cooled by case air. Adding 40mm fans to the side of the power supply case for cooling is MANDATORY. We anticipate using these in our mining farm to power OC'ed AM Prisma's. One supply per Prisma. Of other interest is, the efficiency of these supplies under light loading (250W draw) is 87%, at or near rated output the efficiency rises to 92%. So from a Bitcoin mining perspective it's better to run them WOT (Wide Open Throttle) than to be conservative. Individuals in the RC community have reported about the DPS-2000BB's that when tweaking the output voltage to 14.6 and tweaking the Over Power Cutoff the DPS-2000's can supply 2800W (~200A @ 14.6V), Booyah! |
|
|
|
|
Show Posts
