144 Spartan6 LX150 FPGA cluster! 28.8 GH/s!!!

Turbor

legendary

Activity: 1022

Merit: 1000

BitMinter

Quote from: Inspector 2211 on February 04, 2012, 01:26:54 PM

So, are you alluding to SUCCESSFUL designs that went beyond 205 MHz and thus used more power, or to UNsuccessful designs that somehow wound up using more power?

If the former, that would be big news on this forum, considering the fact that right now I'm mining at a mere 208 MHz on a ZTEX board (with a shitty $2.79 fan cooler from China, though, not ZTEX's luxury cooler).

my first fan from this luxury cooler just died (started to make inacceptable noise). Not to much of a tragedy but earlier as i expected. Ordered some kaze minis from Scythe. They say 30000 hours average lifetime... let's see

Inspector 2211

sr. member

Activity: 448

Merit: 250

Quote from: ArtForz on February 03, 2012, 07:37:45 PM

Well, 7.5A/core was ok for a 192-205MHz 122-round 2-stage-per-round pipeline design. Just my later designs needed more power.

So, are you alluding to SUCCESSFUL designs that went beyond 205 MHz and thus used more power, or to UNsuccessful designs that somehow wound up using more power?

If the former, that would be big news on this forum, considering the fact that right now I'm mining at a mere 208 MHz on a ZTEX board (with a shitty $2.79 fan cooler from China, though, not ZTEX's luxury cooler).

[Tycho]

hero member

Activity: 742

Merit: 500

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

Btw, one thing that made me go WTF when looking at interior pics of copacobana a few years ago... a single massive AC->1.2V converter for vccint and really heavy cables + busbars to route it around...

Why they don't need any heatsinks on their FPGAs ?

DeathAndTaxes

donator

Activity: 1218

Merit: 1079

Gerald Davis

Quote from: rjk on February 02, 2012, 06:04:30 PM

I wonder how recent that price is though. I bet if someone contacted them, it could be had for cheaper.

Probably not. They are custom built and have tons of high speed connectivity, and banks of ram. Totally useless for Bitcoin but expensive and useful for other applications.

RandyFolds

sr. member

Activity: 448

Merit: 250

Quote from: zefir on February 03, 2012, 06:35:24 PM

Quote from: Dexter770221 on February 03, 2012, 06:22:31 PM

No sign of high current regulators for VCCINT and no sign of holes to mount a heatsink, so this boards are poorly designed for mining and you will not achieve more than 50MH/s.

... knew there is a catch. Anyhow my BitForces are to ship soon Wink

See mousepotato's comment...

ArtForz

sr. member

Activity: 406

Merit: 257

Quote from: Inspector 2211 on February 03, 2012, 07:03:22 PM

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

My early LX150 miner prototypes had a 60A switcher for vccint for 8 chips -> 7.5A each... and that later turned out to be not enough.

So, how come the German guy (Stefan) of ZTEX gets by with a [very compact] 8 Amp converter for VCCINT?
His dynamic clocking now achieves > 200 MH/s on a -3 device.
Impressive.
(Even more impressive considering the fact that he has less than half of the Xilinx-mandated amount of bypass capacitors surrounding the Spartan-6.)

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

Btw, one thing that made me go WTF when looking at interior pics of copacobana a few years ago... a single massive AC->1.2V converter for vccint and really heavy cables + busbars to route it around...

Here's a picture: http://www.copacobana.org/photos/photo_b4.jpg

Well, 7.5A/core was ok for a 192-205MHz 122-round 2-stage-per-round pipeline design. Just my later designs needed more power.
Also, unless I misremember, he uses ultra-low-esl 0306 caps for vccint, xilinx recommended #s are for caps with ESL values similar to 0402s... so... prolly fine.
Yep, that's the picture... Still don't get why on earth they did it that way. A AC->12V PSU and half a dozen or so point-of-load converters would've saved a whole lot of copper (and prevent accidentally welding holes in your chassis...)

Inspector 2211

sr. member

Activity: 448

Merit: 250

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

My early LX150 miner prototypes had a 60A switcher for vccint for 8 chips -> 7.5A each... and that later turned out to be not enough.

So, how come the German guy (Stefan) of ZTEX gets by with a [very compact] 8 Amp converter for VCCINT?
His dynamic clocking now achieves > 200 MH/s on a -3 device.
Impressive.
(Even more impressive considering the fact that he has less than half of the Xilinx-mandated amount of bypass capacitors surrounding the Spartan-6.)

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

Btw, one thing that made me go WTF when looking at interior pics of copacobana a few years ago... a single massive AC->1.2V converter for vccint and really heavy cables + busbars to route it around...

Here's a picture: http://www.copacobana.org/photos/photo_b4.jpg

zefir

donator

Activity: 919

Merit: 1000

Quote from: ArtForz on February 03, 2012, 06:45:16 PM

First one has a shared 10A reg for 2 LX150s -> max of 5A vccint each.
Second has a 12A for 4 LX150s -> 3A each
My early LX150 miner prototypes had a 60A switcher for vccint for 8 chips -> 7.5A each... and that later turned out to be not enough.
Btw, one thing that made me go WTF when looking at interior pics of copacobana a few years ago... a single massive AC->1.2V converter for vccint and really heavy cables + busbars to route it around...

At least it's good to know that you get what you pay for (even if the price premium might be slightly disproportional). Don't really know what they charge for the Copacobana, but speculate it should be in the 6-digit range.

ArtForz

sr. member

Activity: 406

Merit: 257

Quote from: zefir on February 03, 2012, 06:13:47 PM

Anyone noticed this one:
http://enterpoint.co.uk/products/spartan-6-development-boards/xc6slx150-x2/
Shop: http://www.enterpoint.co.uk/shop/en/93-xc6slx150-x2-coprocessor.html
2 XC6SLX150, 400 MH/s, £300.00 tax excl. => ~1.2US$ / MH/s

This one is less cost-effective, but higher integrated:
http://enterpoint.co.uk/products/asic-development-high-performance-computing/merrick-3/
Shop: http://www.enterpoint.co.uk/shop/en/98-merrick-3.html (sold out)
24 XC6SLX150, 4.8 GH/s, £6,500.00 tax excl. => ~2.2US$ / MH/s

The first one got my heads-up, since it comes for a price comparable to Icarus, which currently is the most cost-effective and commercially available FPGA miner.

First one has a shared 10A reg for 2 LX150s -> max of 5A vccint each.
Second has a 12A for 4 LX150s -> 3A each
My early LX150 miner prototypes had a 60A switcher for vccint for 8 chips -> 7.5A each... and that later turned out to be not enough.
Btw, one thing that made me go WTF when looking at interior pics of copacobana a few years ago... a single massive AC->1.2V converter for vccint and really heavy cables + busbars to route it around...

zefir

donator

Activity: 919

Merit: 1000

Quote from: Dexter770221 on February 03, 2012, 06:22:31 PM

No sign of high current regulators for VCCINT and no sign of holes to mount a heatsink, so this boards are poorly designed for mining and you will not achieve more than 50MH/s.

... knew there is a catch. Anyhow my BitForces are to ship soon Wink

Dexter770221

legendary

Activity: 1029

Merit: 1000

No sign of high current regulators for VCCINT and no sign of holes to mount a heatsink, so this boards are poorly designed for mining and you will not achieve more than 50MH/s.

zefir

donator

Activity: 919

Merit: 1000

Anyone noticed this one:
http://enterpoint.co.uk/products/spartan-6-development-boards/xc6slx150-x2/
Shop: http://www.enterpoint.co.uk/shop/en/93-xc6slx150-x2-coprocessor.html
2 XC6SLX150, 400 MH/s, £300.00 tax excl. => ~1.2US$ / MH/s

This one is less cost-effective, but higher integrated:
http://enterpoint.co.uk/products/asic-development-high-performance-computing/merrick-3/
Shop: http://www.enterpoint.co.uk/shop/en/98-merrick-3.html (sold out)
24 XC6SLX150, 4.8 GH/s, £6,500.00 tax excl. => ~2.2US$ / MH/s

The first one got my heads-up, since it comes for a price comparable to Icarus, which currently is the most cost-effective and commercially available FPGA miner.

stcupp

full member

Activity: 209

Merit: 100

Quote from: [Tycho] on February 02, 2012, 06:35:52 PM

Quote

Power/cooling to handle up to 5W per FPGA

Not enough. Full-speed mining would require more than 8W per FPGA.

Also, it's at least 3 times more expensive than currently available Spartan-6 mining solutions.

I noticed this too.... its about $660 per spartan 6 there are cheaper solutions in the cards built just for mining

God I just wish i had the $ to play around with a huge fpga cluster..... Sad

[Tycho]

hero member

Activity: 742

Merit: 500

Quote

Power/cooling to handle up to 5W per FPGA

Not enough. Full-speed mining would require more than 8W per FPGA.

Also, it's at least 3 times more expensive than currently available Spartan-6 mining solutions.

stcupp

full member

Activity: 209

Merit: 100

Quote from: kais3r on February 02, 2012, 06:03:32 PM

Quote from: stcupp on February 02, 2012, 05:55:52 PM

lol yeah mine too

http://www.fpga-faq.com/FPGA_Boards.shtml

this site lists each pcie board as $7950.00

$7950.00 * 12 = $95,400

so i'd say the whole cluster is around $110,000

wow, I believe with 110K you could make a much more efficient mining rig

Maybe more hashes but not more efficient... FPGA's don't use as much power

each card with 12 fpga's is probably only running around 1800mh/s at 120 watts

while a gpu runs around 750 mh/s at 400 watts or 350mh/s for 200 watts

also theres no where near as much heat output on these cards compared to gpu's

so FPGA's are alot more efficient less money in power, more mh/s per watt, and less money in cooling

rjk

sr. member

Activity: 448

Merit: 250

1ngldh

I wonder how recent that price is though. I bet if someone contacted them, it could be had for cheaper.

kais3r

newbie

Activity: 85

Merit: 0

Quote from: stcupp on February 02, 2012, 05:55:52 PM

lol yeah mine too

http://www.fpga-faq.com/FPGA_Boards.shtml

this site lists each pcie board as $7950.00

$7950.00 * 12 = $95,400

so i'd say the whole cluster is around $110,000

wow, I believe with 110K you could make a much more efficient mining rig

Timbo925

sr. member

Activity: 352

Merit: 250

Quote from: Mousepotato on February 02, 2012, 05:59:53 PM

Ships in 4-6 weeks.

Made me LOL Grin

Mousepotato

hero member

Activity: 896

Merit: 1000

Seal Cub Clubbing Club

Ships in 4-6 weeks.

stcupp

full member

Activity: 209

Merit: 100

lol yeah mine too

http://www.fpga-faq.com/FPGA_Boards.shtml

this site lists each pcie board as $7950.00

$7950.00 * 12 = $95,400

so i'd say the whole cluster is around $110,000

Topic: 144 Spartan6 LX150 FPGA cluster! 28.8 GH/s!!! (Read 16284 times)