Pages:
Author

Topic: X6500 Custom FPGA Miner - page 48. (Read 220050 times)

staff
Activity: 4284
Merit: 8808
December 05, 2011, 11:03:46 PM
A warning to everyone: Cooling these boards properly is crucial. Improper cooling will increase your power consumption, increase error rates, and shorten the life of the FPGAs. If the board or heatsinks feel hot at all, you are not cooling them properly. We are not responsible for any damages that may arise from insufficient cooling.

To emphasize here:   Semiconductor devices can be prone to thermal runaway: Higher temperatures make them less efficient which leads to higher temperatures until its dead.

The GPUs most people use to mine today have automatic thermal throttling.   Even if the fan fails the GPU will slow itself down before it self destructs.  The FPGA mining devices do not.   Cheap fans which are acceptable in devices which can self-throttle when the fan fails are not acceptable for devices which do not throttle.

It might be the case that miner will crash and fail into a low power state if it overheats, thus saving you from lasting damage— or it might irreparably harm itself.  You don't want to be the person who finds out which of these failure modes is more likely.


hero member
Activity: 720
Merit: 525
December 05, 2011, 09:48:30 PM
I have 12 boards running right now, but only for testing. Every board that ships out will be tested for at least an hour to make sure that everything is ok.

When I finally set aside some boards for the FPGA.contract and build some type of enclosure for it, I'll post a pic in the "pictures of your rigs" thread. Smiley

Funny thing about running a bunch of FPGA boards. I'm sitting next to over 3 GH/s and shivering! The fans I have going are overkill, but it doesn't seem safe to run these completely passively. You would probably need to underclock them quite a bit or put on a bigger heatsink. With a little bit of airflow though, they stay cool to the touch. The ideal cooling scheme will need to be worked out, and I'm excited to see how you guys will handle it.

A warning to everyone: Cooling these boards properly is crucial. Improper cooling will increase your power consumption, increase error rates, and shorten the life of the FPGAs. If the board or heatsinks feel hot at all, you are not cooling them properly. We are not responsible for any damages that may arise from insufficient cooling.
legendary
Activity: 1022
Merit: 1000
BitMinter
December 05, 2011, 02:15:28 PM
10 of your boards, 3 be quiet 140mm fans... a miners dream !
donator
Activity: 2772
Merit: 1019
December 05, 2011, 08:27:49 AM
Thanks sadpandatech and Turbor! I have another pic for you, although the ratsnest of cables might not be as sexy...



nice stuff. you should put this in the "pictures of your mining rigs" thread.
sr. member
Activity: 410
Merit: 252
Watercooling the world of mining
December 05, 2011, 02:26:38 AM
Nom nom nom,can't wait to get my toys Cheesy

Then i at least abled to test my code.
I hope the handrouting will get us some new performance numbers. Wink
hero member
Activity: 720
Merit: 525
December 04, 2011, 07:40:03 PM
Thanks sadpandatech and Turbor! I have another pic for you, although the ratsnest of cables might not be as sexy...

legendary
Activity: 1022
Merit: 1000
BitMinter
December 04, 2011, 08:40:13 AM
Looking good !  Wink
hero member
Activity: 504
Merit: 500
December 03, 2011, 09:57:56 PM
  That's hotter than a table loaded with playboys!   Shocked
hero member
Activity: 720
Merit: 525
December 03, 2011, 09:51:58 PM
The boards are here!



hero member
Activity: 504
Merit: 500
December 03, 2011, 08:41:05 PM
  Looking good. Kept wondering why I was not getting updates on this thread and finally realised I had never posted or flagged it to watch. ;p

  Cheers
hero member
Activity: 720
Merit: 525
December 03, 2011, 04:37:05 AM
Bad news on the production progress, everyone. I didn't receive the boards today and when I called manufacturer to find out why, they are telling me that they won't ship until tomorrow. I was given no warning from them that there might be a delay, so I'm very disappointed in the manufacturer right now.

I'll keep you all updated with any news I hear from them. If they end up offering us a big refund because they didn't meet the agreed upon ship date, we'll find a way to pass that savings on to the customers.

I have a tracking number and it will be arriving in less than 12 hours! Overnight shipping with Saturday delivery is expensive, but it's worth it to give us the weekend to work on testing and packing these. Unfortunately, they're refusing to give any discount because of the delay, because a few times they had to pause production and call me with questions. Even though I answered right away, they somehow say that explains the 2 day delay...
hero member
Activity: 720
Merit: 525
December 02, 2011, 10:43:03 AM
By the way, I did some calculations of the cost/performance comparison to the ztex boards a while back in this thread: https://bitcointalk.org/index.php?topic=7216.920

But you forgot to mention that there are volume discounts: https://bitcointalksearch.org/topic/--7216

You're right. I compared the single quantity price for each of ours, which I think is fair because we haven't published any bulk prices yet. We have given quite a few bulk quotes out privately, though, and plan to publish a full breakdown like yours as soon as we start shipping boards to customers.

Here's a thorough comparison, although still unfair because ours is at the single quantity price. By the way, no one has actually paid $580, they have all paid $560. Also, as far as I can tell, the ztex board doesn't come with a heatsink. Cablesaurus recently added the option to supply your own heatsink, knocking $10 off the price of the board. At the bottom, I show a comparison to the price if you ordered right now, with no heatsinks ($550).



Note how, if you're optimistic about the future performance of this board, we still beat the ztex quantity 100 price versus our quantity 1.
donator
Activity: 367
Merit: 250
ZTEX FPGA Boards
December 02, 2011, 08:57:20 AM
By the way, I did some calculations of the cost/performance comparison to the ztex boards a while back in this thread: https://bitcointalk.org/index.php?topic=7216.920

But you forgot to mention that there are volume discounts: https://bitcointalksearch.org/topic/--7216
hero member
Activity: 720
Merit: 525
December 02, 2011, 08:29:24 AM
I'm a bit curious about the advertised hashrates on these boards compared to other FPGA devices. Specifically, it looks like the ZTEX board is using the same FPGA as the X6000 (in fact, possibly a slightly inferior one), but they claim they're getting an average 190MHash/s rate, which is nearly what your X6500 board claims to get. Is this just a matter of incompatible metrics (minimum MHash versus average MHash), or is there something in their FPGA logic that you could harness to get better hashrates on these boards?

Edit: I reread the thread, and noticed that this is a conversation you had yourself - where it turned out there was, in fact, some logic optimizations they did that aren't incorporated into these boards yet. Well, that's fine, so long as you'll be revisiting it sometime after the first batch have shipped Wink

Yep, our code is a little bit lagging behind the other guys right now, but I fully expect that to improve soon. Until we've proven that, I certainly won't make any guarantees about higher performance. As of yet, we have only 2 prototype boards running, both at 125 MHz (250 MH/s), but we are still publishing the guaranteed hashrate at 200 MH/s until we've confirmed that it works perfectly on all of the production batch of boards. I would much prefer to set the bar low and exceed customers' expectations later, then to do the opposite.

Also, the actual real world performance is a bit lower than that because our mining software isn't perfect yet. For example, we don't have long polling yet. This appears to reduce our effective hash rate by about 10%. Still, that is something that can be fixed in software, not a limitation of the hardware.

By the way, I did some calculations of the cost/performance comparison to the ztex boards a while back in this thread: https://bitcointalk.org/index.php?topic=7216.920
legendary
Activity: 960
Merit: 1028
Spurn wild goose chases. Seek that which endures.
December 01, 2011, 08:56:17 PM
I'm a bit curious about the advertised hashrates on these boards compared to other FPGA devices. Specifically, it looks like the ZTEX board is using the same FPGA as the X6000 (in fact, possibly a slightly inferior one), but they claim they're getting an average 190MHash/s rate, which is nearly what your X6500 board claims to get. Is this just a matter of incompatible metrics (minimum MHash versus average MHash), or is there something in their FPGA logic that you could harness to get better hashrates on these boards?

Edit: I reread the thread, and noticed that this is a conversation you had yourself - where it turned out there was, in fact, some logic optimizations they did that aren't incorporated into these boards yet. Well, that's fine, so long as you'll be revisiting it sometime after the first batch have shipped Wink
hero member
Activity: 720
Merit: 525
December 01, 2011, 06:04:25 PM
Bad news on the production progress, everyone. I didn't receive the boards today and when I called manufacturer to find out why, they are telling me that they won't ship until tomorrow. I was given no warning from them that there might be a delay, so I'm very disappointed in the manufacturer right now.

I'll keep you all updated with any news I hear from them. If they end up offering us a big refund because they didn't meet the agreed upon ship date, we'll find a way to pass that savings on to the customers.
hero member
Activity: 720
Merit: 525
December 01, 2011, 08:02:07 AM
fpgaminer actually has an FT232R breakout board so he will test it with his X5000 to confirm that it works. If there are no surprises there, you could either use a board like that with a few wires to jump to the JTAG header on the X5000, or use the cable I linked to, which has the FT232R chip inside it. I think that solution will be cleaner and easier, and about the same cost.

If you want to go that route, I'll gladly make and ship cables to both of you guys if you pay for the cost of the cable. Let's wait for fpgaminer's confirmation that it works, though.
sr. member
Activity: 445
Merit: 250
November 30, 2011, 02:36:33 PM
With fpgaminer's imprimatur, I, too, would love to ditch the Xilinx cable & take up your offer. I'm very pleased with my X5000, but looking forward to the X6500's  Smiley
donator
Activity: 2772
Merit: 1019
November 30, 2011, 11:12:22 AM
No code improvements for higher hashrate yet?

Sorry for the very late reply. Unfortunately, no improvements yet. Once these boards are finally on their way out to customers, we expect to be able to devote a lot more time to that work.

When updating software, can you keep in mind your first customers that have x5000?

Yes, absolutely. Unfortunately, I really don't know much about the X5000, because it was before my time. We need fpgaminer's help, but he's mostly MIA lately because he is swamped with work and moving to a new apartment right now. Hopefully he will be able to work on this stuff soon.

I do know that if you could switch from the Xilinx platform cable to an FT232R breakout board, the hardware and software interface would be identical to the X6500, so all code would be completely compatible. This would also eliminate the need for ISE.

Actually, this cable soldered to the correct connector would probably be perfect. If you were willing to pay for the cable, I'd be happy to solder it up and send it to you. At that point, I think you would just plug it in and run the X6500 software (with the X6500 bitstream) and you'd be off and running.

fpgaminer, can you confirm that my idea makes sense? Do you have other ideas for how to maintain support for the X5000?

fizzisist, I would happily pay for the cable if that method works. I hate that "xilinx platform cable" anyway.

You would have to ship it to germany, but it should probably fit in a letter, right? If shipping is prohibitively expensive, I can solder myself given you tell me what goes where.

fpgaminer, would that work?
hero member
Activity: 720
Merit: 525
November 30, 2011, 09:18:53 AM
No code improvements for higher hashrate yet?

Sorry for the very late reply. Unfortunately, no improvements yet. Once these boards are finally on their way out to customers, we expect to be able to devote a lot more time to that work.

When updating software, can you keep in mind your first customers that have x5000?

Yes, absolutely. Unfortunately, I really don't know much about the X5000, because it was before my time. We need fpgaminer's help, but he's mostly MIA lately because he is swamped with work and moving to a new apartment right now. Hopefully he will be able to work on this stuff soon.

I do know that if you could switch from the Xilinx platform cable to an FT232R breakout board, the hardware and software interface would be identical to the X6500, so all code would be completely compatible. This would also eliminate the need for ISE.

Actually, this cable soldered to the correct connector would probably be perfect. If you were willing to pay for the cable, I'd be happy to solder it up and send it to you. At that point, I think you would just plug it in and run the X6500 software (with the X6500 bitstream) and you'd be off and running.

fpgaminer, can you confirm that my idea makes sense? Do you have other ideas for how to maintain support for the X5000?
Pages:
Jump to: