cheers
Topic: X6500 Custom FPGA Miner - page 19. (Read 220050 times)
wow, didn't realize there was a fizzisist fork of the modular miner....runs much more smoothly, way fewer stales on gpumax.
cheers
cheers
In the last couple weeks, we've been experimenting with different FPGA core voltages (VCCINT). Our measurements have shown that on most rev3 boards, performance can be increased significantly by slightly increasing this voltage. Because of this, all rev3 boards starting today will be shipped with VCCINT set to 1.23V (instead of the original 1.20V). This applies to all boards shipped on or after 3/29/2012.
While this increases the board's power consumption slightly (data here), it also increases the potential limit on clock rate, which is especially useful with the new "overclocker" bitstreams.
If anyone would like to increase the core voltages on an older board, it simply means swapping two resistors (R2 and R8). You can either attempt this yourself or send it back to us and we'll do the work free of charge. We only ask that you pay the return shipping charge.
While this increases the board's power consumption slightly (data here), it also increases the potential limit on clock rate, which is especially useful with the new "overclocker" bitstreams.
If anyone would like to increase the core voltages on an older board, it simply means swapping two resistors (R2 and R8). You can either attempt this yourself or send it back to us and we'll do the work free of charge. We only ask that you pay the return shipping charge.
Looking at the thermal data of the FPGA, I'd say that a heatsink temperature of 50°C and a board temperature of 60°C should be perfectly fine. 10°C more might cause invalids but shouldn't really hurt it. So as long as the oil stays around 50°C it should be fine.
Great, good to know. I now tried it without a fan for a day and the temperature has stabilized at around 51 °C. In that time it has produced 2 invalid shares (showing as 0.0%).
You beat me to it, I just received my tech grade mineral oil yesterday and the rest of my parts should arrive by the end of the week.
Cool to hear that you are giving it a shot as well. Looking forward to see your finished setup!
Awww you know that you will end up with oil on you desks? And u lost your warranty btw. 
Good thing the X6500 doesn't have a warranty to begin with. =)
It's called cappilar effect .... i did that once with my computer and the oil crawled up the mouse cable and after a short while my mouse began to bleed oil ...
Yes, I was wondering if capillary effect would be a problem. That's why I run the cables up the wall for a bit before going down again. I would think that the effect isn't strong enough to overcome that kind of distance. But only time will tell.
I would think that in the case of a mouse cable the risk is higher, because by using the mouse, the cable is moved back and forth sometimes and helping in transporting oil where it shouldn't be. I think the situation here is much better, as everything is very static.
Are you getting any long poll errors.
Long poll errors with MPBM on GPUMAX are a known issue. However I can't test things myself because I don't have a GPUMAX account, and there seems to be a months-long backlog on invites.
However, judging from other people's results, it looks like that long polling error is not critical, and does not mean that long polling doesn't work at all. There are only very brief time frames, during which it fails (when GPUMAX is trying to re-route your LP request internally).
Are you getting any long poll errors.
Yes, with a proper PC it appears it should run great.
Does the x6500 work on gpumax?
This element should be avalable i guess some weeks after the 3. gen arrives
Hi Icoin, wow that looks fantastic. I suppose it'll come with spring loaded screws or something to mount onto the board. I just wished the Spartan-6 had a metal lid instead of it being encapsulated in epoxy like at the moment, that seriously hampers our cooling efforts. Ah well...
Hi everyone
I realized that there is a cooling problem for the fpga, thats why i developed a watercooling element specialy for the 6500 boards. The element is developed in a way that it should fit to all generations developed by now. Till today i could only use the measurements of the 1. gen but a 3. gen Board is on its way to me.
This element should be avalable i guess some weeks after the 3. gen arrives
This is easy integratable in a existing CPU/GPU watercooling solution
I realized that there is a cooling problem for the fpga, thats why i developed a watercooling element specialy for the 6500 boards. The element is developed in a way that it should fit to all generations developed by now. Till today i could only use the measurements of the 1. gen but a 3. gen Board is on its way to me.
This element should be avalable i guess some weeks after the 3. gen arrives
This is easy integratable in a existing CPU/GPU watercooling solution
Why somebody should invent a suboptimal solution while there is already the better solution watercooling? Water can carry a lot of energy per volume and the temperature range is perfect. Oil cooling would only make sense if the temperature would be higher or there were high voltages involved.
Water cooling is WAY too expensive.I bet with proper heat sinks, dense packing and a clean air flow through the heatsinks air cooling should be sufficient.
For simple oil immersion you just need a bucket and the oil :)
Chipset waterblock is about $40 to $50 for low quantity orders, so it's at least $80 for each board. Notice that I'm not talking about pipes, pumps, water-to-air exchangers, fans and other accessories. FPGA miner costs about $420 to $570, so just two waterblocks will cost you around 16% of your miner and that's too much. Mining gear should be as cheap as possible.
2. Spartan package is not so good for heat transfer, so you can't cool it efficiently from just the top. PCB backside will definitely need some airflow.
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
Why somebody should invent a suboptimal solution while there is already the better solution watercooling? Water can carry a lot of energy per volume and the temperature range is perfect. Oil cooling would only make sense if the temperature would be higher or there were high voltages involved.I doubt if cryogenic or oil would be less expensive, since you need customized heat exchanger and pipes to.
I bet with proper heat sinks, dense packing and a clean air flow through the heatsinks air cooling should be sufficient.
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
Why somebody should invent a suboptimal solution while there is already the better solution watercooling? Water can carry a lot of energy per volume and the temperature range is perfect. Oil cooling would only make sense if the temperature would be higher or there were high voltages involved.I am a bit surprised that there are no finished solutions for this problem. I can't imagine that there are no companies (scientific research?) that wouldn't need to submerge some equipment in mineral oil for cooling once in a while.
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
Why somebody should invent a suboptimal solution while there is already the better solution watercooling? Water can carry a lot of energy per volume and the temperature range is perfect. Oil cooling would only make sense if the temperature would be higher or there were high voltages involved.
And there are companies that produce Water cooling equipment, and yes, you have to adapt it to your special needs, its like the water pipe system in a house that has to be assembled by a plumber as avery house is more or less different.
Thanks for the links kakobrekla, I was more talking about a solution that a mere mortal like me or a small business can buy off the shelf (or through some specialized company). A case and some equipment just for cooling small stuff or small computers like we use them for mining... so far all I could find was home-made stuff.
Basically something like this, but built by a company so people/organizations in need could just place an order:
http://www.g4tv.com/articles/37474/build-a-submersion-cooling-case/
Basically something like this, but built by a company so people/organizations in need could just place an order:
http://www.g4tv.com/articles/37474/build-a-submersion-cooling-case/
I am a bit surprised that there are no finished solutions for this problem. I can't imagine that there are no companies (scientific research?) that wouldn't need to submerge some equipment in mineral oil for cooling once in a while.
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
Inert fluids are better than mineral oil. Like 3M Novec...
Iceotope Liquid Cooled Cabinets (with Novec)
http://www.youtube.com/watch?v=d2TS_qQ3UNQ
3M™ Novec™ 1230 Fire Protection Fluid Electronics
http://www.youtube.com/watch?v=1E7Nr72rxB8
You can put a piece of paper in this and it will be dry when you pull it out.
I am a bit surprised that there are no finished solutions for this problem. I can't imagine that there are no companies (scientific research?) that wouldn't need to submerge some equipment in mineral oil for cooling once in a while.
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
A (small) tank with some sort of rail/frame to attach equipment/PCBs and then established oil flow with external heat exchanger would be a neat product. But my five minutes of research didn't end up on any web site/manufacturer with a product like this. The closest thing I found was cryogenic equipment...
I can imagine it would only be interesting if it would work for a very high density, say minimum of 125 X6500 or Icarus boards (since this is the hashing power that the rig "box" is supposed to deliver, if it ever shows up).
These guys did a lot of experimenting across several revisions to a fully submerged aquarium gaming PC. Might be some tips in their articles about how they did cooling, etc:
http://www.pugetsystems.com/aquarium-computer.php
http://www.pugetsystems.com/mineral-oil-pc.php
http://www.pugetsystems.com/submerged.php
Yup, I got the oil from their partnered vendor. Mineral oil has the general opinion that it is not worth it in the PC arena and water cooling is better. I think mainly because people would like to use it to cool high end gaming rigs. For FPGA's we have significantly less wattage to worry about.
Some mining rig photos: I went through with my plans for a mineral oil cooled setup. :-)
What you can't see here: The USB and power cable is rising up to the top, as I placed a nail there above the whole setup to run the cables along that. This is an attempt to prevent oil from trickling through the cables. So far the oil stays where it should, hopefully that will last.
All in all it seems to work fairly well. I would like your input though on what is an acceptable temperature to run the FPGA at. First I tried running without a fan and the oil would slowly start heating up and after about 30 minutes the thermometer showed 46 deg C (you can see in the picture that the thermometer is measuring in the middle of the heat sink). At that point I stopped and added a fan. It's not really helping all that much, but it is keeping the setup at about 43 deg C right now.
On the other hand I have yet to see a single invalid share. The system has been mining for over a day at 200 MHz (per FPGA) with 0 invalids. So would that indicate that I can risk a slightly higher temperature? I would really like to run this completely passive.
I was also thinking that this might be a good excuse to play around with 3d printing a little bit and maybe build something that would increase the surface area for the oil to cool down and maybe even have some kind of circular setup driven by convection. Although if I do that, which seems like a fun project, it would probably be just me guessing how to do that rather than being based on actual physics. =)
What you can't see here: The USB and power cable is rising up to the top, as I placed a nail there above the whole setup to run the cables along that. This is an attempt to prevent oil from trickling through the cables. So far the oil stays where it should, hopefully that will last.
All in all it seems to work fairly well. I would like your input though on what is an acceptable temperature to run the FPGA at. First I tried running without a fan and the oil would slowly start heating up and after about 30 minutes the thermometer showed 46 deg C (you can see in the picture that the thermometer is measuring in the middle of the heat sink). At that point I stopped and added a fan. It's not really helping all that much, but it is keeping the setup at about 43 deg C right now.
On the other hand I have yet to see a single invalid share. The system has been mining for over a day at 200 MHz (per FPGA) with 0 invalids. So would that indicate that I can risk a slightly higher temperature? I would really like to run this completely passive.
I was also thinking that this might be a good excuse to play around with 3d printing a little bit and maybe build something that would increase the surface area for the oil to cool down and maybe even have some kind of circular setup driven by convection. Although if I do that, which seems like a fun project, it would probably be just me guessing how to do that rather than being based on actual physics. =)
You beat me to it, I just received my tech grade mineral oil yesterday and the rest of my parts should arrive by the end of the week. I think I have a pretty good design, building it completely out of acrylic glued like a fish tank. I plan to include the cooling into the lid. One lid will be two heat sinks connected through the lid. The other plan is the nuclear option with a peltier cooler sandwiched between them. The enclosure itself is designed for convention so the cooling in the lid may be completely unneeded.. we will see.
These guys did a lot of experimenting across several revisions to a fully submerged aquarium gaming PC. Might be some tips in their articles about how they did cooling, etc:
http://www.pugetsystems.com/aquarium-computer.php
http://www.pugetsystems.com/mineral-oil-pc.php
http://www.pugetsystems.com/submerged.php
It is also capaple of fully running a bitcoin-qt client( i recompiled one for ARM boards) and also act as a p2pool server at marginal cpu usage levels.
But p2pool is not recommended for the current mpbm version as this will cause high stale levels due to timing constraints.
While it is used as a board running the miner connected to a pool its mostly ideling. The cpu is really bored.
But p2pool is not recommended for the current mpbm version as this will cause high stale levels due to timing constraints.
While it is used as a board running the miner connected to a pool its mostly ideling. The cpu is really bored.
P2Pool should work fine with current master / testing tree. However it looks like p2pool is running into performance issues on the pandaboard as well, causing ~6% DOA for me. If I host it somewhere else it's running fine.
Might be somehow related to the I/O perfromance of your SD card. Mine is very fast and it just unns fine. ( well i also did some modifications to ubuntu to optimise).
As far as TheSeven told me mpbm has some latency issues wich are the price of the modular approach.
Yes, it might well be related to SD card performance, that's what I was suspecting as well. (Getting a very high response latency variance close to found shares.)
MPBM surely isn't the miner with the lowest latencies, but since I've fixed a new block detection bug and decoupled the cancellation of jobs into another thread, it seems to work well enough. I'm getting around 1.7% DOA stales with a remote P2Pool instance, running MPBM on a pandaboard.
It is also capaple of fully running a bitcoin-qt client( i recompiled one for ARM boards) and also act as a p2pool server at marginal cpu usage levels.
But p2pool is not recommended for the current mpbm version as this will cause high stale levels due to timing constraints.
While it is used as a board running the miner connected to a pool its mostly ideling. The cpu is really bored.
But p2pool is not recommended for the current mpbm version as this will cause high stale levels due to timing constraints.
While it is used as a board running the miner connected to a pool its mostly ideling. The cpu is really bored.
P2Pool should work fine with current master / testing tree. However it looks like p2pool is running into performance issues on the pandaboard as well, causing ~6% DOA for me. If I host it somewhere else it's running fine.
Might be somehow related to the I/O perfromance of your SD card. Mine is very fast and it just unns fine. ( well i also did some modifications to ubuntu to optimise).
As far as TheSeven told me mpbm has some latency issues wich are the price of the modular approach.
Jump to: