Cairnsmore1 - Quad XC6SLX150 Board

[kano]

makomk's dcmwd2 190MH/s to all but ICA 9 which is running the 180 one.

...
Note that boards with HW: > 7074 are showing some strange error since all of them, at a certain point in time, went from a normal error count to 7074 and from them kept increasing it.

spiccioli.

End of last week I wrote a proper HW: detection code change and then had that same problem occur ... but only once.
... I've been trying to make it happen again with a whole bunch of debug added, but it's only happened for me that once.

Somehow the Icarus driver gets in a loop of processing a single work a few thousand times (very quickly) - and then it escapes the loop.
It actually doesn't cause much of a problem - but of course when you have a counter like that inside the loop, the counter goes up by the thousands of times it loops.

If you have time it would be good if I could give you a version of cgminer to debug the cause of the problem so I can fix it.

The change I wrote last week (that won't be going into cgminer) is:
https://github.com/kanoi/cgminer/commit/0df64344407180353bec4b7c612e8385c6bac928
There's nothing wrong with it, it's just that it needs to be written to support GPU in the findnonce.c code also, so I need to make it a little different and then edit all the other devices.

[1714009513]

1714009513

[1714009513]

1714009513

[1714009513]

1714009513

[1714009513]

1714009513

[1714009513]

1714009513

[spiccioli]

makomk's dcmwd2 190MH/s to all but ICA 9 which is running the 180 one.

...
Note that boards with HW: > 7074 are showing some strange error since all of them, at a certain point in time, went from a normal error count to 7074 and from them kept increasing it.

spiccioli.

End of last week I wrote a proper HW: detection code change and then had that same problem occur ... but only once.
... I've been trying to make it happen again with a whole bunch of debug added, but it's only happened for me that once.

Somehow the Icarus driver gets in a loop of processing a single work a few thousand times (very quickly) - and then it escapes the loop.
It actually doesn't cause much of a problem - but of course when you have a counter like that inside the loop, the counter goes up by the thousands of times it loops.

If you have time it would be good if I could give you a version of cgminer to debug the cause of the problem so I can fix it.

The change I wrote last week (that won't be going into cgminer) is:
https://github.com/kanoi/cgminer/commit/0df64344407180353bec4b7c612e8385c6bac928
There's nothing wrong with it, it's just that it needs to be written to support GPU in the findnonce.c code also, so I need to make it a little different and then edit all the other devices.

kano,

yes, if you have one please tell me where I can download it, it has to be compiled without ADL/OpenCL and it will be running on a PC with very little available disk space.

I can build it myself, if you're better giving me the source code.

Code:

$ df /
Filesystem     1K-blocks    Used Available Use% Mounted on
/dev/sda1        1995048 1459284    435824  78% /

spiccioli

[ebereon]

@Kano:
I use your windows binary 2.7.0a and I don't get HW: counting. When do you think we will have the HW detection in your windows binary too?

eb

[kano]

...
kano,

yes, if you have one please tell me where I can download it, it has to be compiled without ADL/OpenCL and it will be running on a PC with very little available disk space.

I can build it myself, if you're better giving me the source code.

Code:

$ df /
Filesystem     1K-blocks    Used Available Use% Mounted on
/dev/sda1        1995048 1459284    435824  78% /

spiccioli

I'll make a branch in my git tomorrow, sorry don't have time tonight (it's now 8pm)
I'll have to sort out a way to make it turn on debug when it happens rather than having a large debug log
(maybe when it gets over 1000 sounds like an easy test)
And let you know the git clone command when it's ready
Thanks!

[kano]

@Kano:
I use your windows binary 2.7.0a and I don't get HW: counting. When do you think we will have the HW detection in your windows binary too?

eb

Certainly not until after this bug is sorted out.

I'll then need to rewrite the share checking for all the drivers so that might be a while yet.
... I'm rarely quick at getting stuff done - since most of it is free

[yohan]

Controller Rev 1.5 is now available on the support page. Programmer is fixed and it offers the dualling up of 2 boards onto 1 USB cable.

[LazyOtto]

dualling up of 2 boards onto 1 USB cable.

Is there anything 'magic' about the ribbon cable between boards?

IOWs, could I just take an old IDE or floppy ribbon cable, chop off the extra wires/connections, and use that?

[ebereon]

dualling up of 2 boards onto 1 USB cable.

Is there anything 'magic' about the ribbon cable between boards?

IOWs, could I just take an old IDE or floppy ribbon cable, chop off the extra wires/connections, and use that?

It's a easy 1to1 wire. The dimension of the IDE or Floppy connectors(holes) are to big.

I done them my self with these connectors and cables:
- connector 0.08€ -> post connectors
- cable 10 pol 3 meters 1.95€ -> ribbon cable

That's 0,225 € / CM1 Ribbon cable  

The left one is my own made cable(0,225€+2minutes time), the right one is from enterpoint(5.20€):



More photos here

eb

[yohan]

dualling up of 2 boards onto 1 USB cable.

Is there anything 'magic' about the ribbon cable between boards?

IOWs, could I just take an old IDE or floppy ribbon cable, chop off the extra wires/connections, and use that?

You first problem is the connector size is all wrong. The length may also affect results as well. Also IDE ribbons use a strange double wire setup which tends not work in anything else.

It's possible to make these things and that is probably better but unless you have the right kit results can be patchy. There many hairless engineers around as the result of trying to make up ribbon cables.

[testconpastas2]

It's a easy 1to1 wire. The dimension of the IDE or Floppy connectors(holes) are to big.

I done them my self with these connectors and cables:
- connector 0.08€ -> post connectors
- cable 10 pol 3 meters 1.95€ -> ribbon cable

Nice ribbon cable and much more cleaner    your detergent is way better than ep's 

And nice shop added to my bookmarks.

Thank you.

[ebereon]

...  your detergent is way better than ep's  ...

 

[ebereon]

Update Power Consumption:

10 x CM1 (mix of makomk's wd4c 200/210Mh bitstreams)
1 x PDB
1 x LogiLink 10 Port USBHub (12,95€)
1 x ASUS Netbook (Win7 32bit, 1GB, cgminer 2.7.0a)
1 x PSU Super-Flower SF650P14XE Golden Green Pro 80plus gold - 54 Ampere max. on 12V (79,95€)
-----------
424W @ 224V @ 8.1GH/s
Power Factor 0.98

eb

[ebereon]

Update on latest bitstreams:
Total 40 FPGA's = 10 boards SN#400+.

Makomk's wd4c oc bitstreams 2 day's running:
15x 210Mh, 25x 200Mh. That means 100% of all fpga's running @200Mh+ with <0.8% invalids!

Working bitstreams in percent:
210Mh oc = 37,5%
200Mh oc = 100%
190Mh oc = 100%
180Mh oc = 100%
170Mh = 100%


Update Glasswalker's bitstream:
175Mh oc = 100% (only short test about 2 hours)

eb

[steamboat]

Update on latest bitstreams:
Total 40 FPGA's = 10 boards SN#400+.

Makomk's wd4c oc bitstreams 2 day's running:
15x 210Mh, 25x 200Mh. That means 100% of all fpga's running @200Mh+ with <0.8% invalids!

Working bitstreams in percent:
210Mh oc = 37,5%
200Mh oc = 100%
190Mh oc = 100%
180Mh oc = 100%
170Mh = 100%


Update Glasswalker's bitstream:
175Mh oc = 100% (only short test about 2 hours)

eb

What's the actual mh/s you're seeing from the 200mh bitstream? I'm avg'ing ~770 per board.

[ebereon]

What's the actual mh/s you're seeing from the 200mh bitstream? I'm avg'ing ~770 per board.

Update Power Consumption:

10 x CM1 (mix of makomk's wd4c 200/210Mh bitstreams)
1 x PDB
1 x LogiLink 10 Port USBHub (12,95€)
1 x ASUS Netbook (Win7 32bit, 1GB, cgminer 2.7.0a)
1 x PSU Super-Flower SF650P14XE Golden Green Pro 80plus gold - 54 Ampere max. on 12V (79,95€)
-----------
424W @ 224V @ 8.1GH/s
Power Factor 0.98

eb

 

[steamboat]

What's the actual mh/s you're seeing from the 200mh bitstream? I'm avg'ing ~770 per board.

Update Power Consumption:

10 x CM1 (mix of makomk's wd4c 200/210Mh bitstreams)
1 x PDB
1 x LogiLink 10 Port USBHub (12,95€)
1 x ASUS Netbook (Win7 32bit, 1GB, cgminer 2.7.0a)
1 x PSU Super-Flower SF650P14XE Golden Green Pro 80plus gold - 54 Ampere max. on 12V (79,95€)
-----------
424W @ 224V @ 8.1GH/s
Power Factor 0.98

eb

 

gah, sorry 

[LazyOtto]

Shipments are going fairly smoothly and about half of September's promised deliveries have been shipped already. We will be asking for the balance to confirm during this coming week as we have to decide both on how much extra silicon to buy and if we can accept any more orders in the near term.

I got the email to confirm and responded with a yes. Haven't heard back with how to make payment (not via BTC for me - I'm BTC poor) and firm shipping date(s) yet.

Can you provide any update on how 'the second half of September' orders are being processed, please?

[LazyOtto]

It's a easy 1to1 wire. The dimension of the IDE or Floppy connectors(holes) are to big.

Thank you, ebereon.

I've got other connectors/ribbons left over from USB to motherboard and SPDIF to motherboard, but my recollection is that those all had the same pin spacing as IDE/floppy connectors.

Probably worth my while to just throw a few dollars at Enterpoint and avoid any parts procurement hassles for this since I only need one atm.

--

Guess all my junk box stuff is getting really old. Anyone need SCSI cables/terminators?

[ebereon]

It's a easy 1to1 wire. The dimension of the IDE or Floppy connectors(holes) are to big.

Thank you, ebereon.
I've got other connectors/ribbons left over from USB to motherboard and SPDIF to motherboard, but my recollection is that those all had the same pin spacing as IDE/floppy connectors.

Oh well, I was wrong! I checked it again and it seems they have the same dimensions, sorry about that! It could work if you cut the connectors, but make sure you connect it correctly, I don't know if it can damage the board when you connect it in the wrong direction.

eb

[yohan]

Shipments are going fairly smoothly and about half of September's promised deliveries have been shipped already. We will be asking for the balance to confirm during this coming week as we have to decide both on how much extra silicon to buy and if we can accept any more orders in the near term.

I got the email to confirm and responded with a yes. Haven't heard back with how to make payment (not via BTC for me - I'm BTC poor) and firm shipping date(s) yet.

Can you provide any update on how 'the second half of September' orders are being processed, please?

We are still basically processing in order but we are ahead currently. What we are doing now is really to check that we have the right amount of Spartan-6 silicon. As we are getting towards the end of the pre-orders it is a bit like how airlines fill seats. They overbook seats on the basis some peipole won't show for the flight. We underbuy silicon on the basis some orders will drop out. So to avoid customer dissappointment we need to check now if we have enough silicon. The lead times to get extra Spartan-6 silicon are about 6 weeks ands we are already marginal on timing to get that in and build boards in September. This does mean we could be late on the last 100 boards or so if people tell us last minute they do want their order and everyone wants their entire order. Whilst we can carry a small risk of potentially overbuying a small amount of Spartan-6 100 boards worth would be a serious cashflow issue at best and it could be a lot worse for our business.

So before the end of the week we will decide on how much Spartan-6 silicon we need to buy and place orders accordingly. If we don't get responses from customers on the email that went out then we will assume they are not going ahead. We won't wipe their order but if they come back late then their order might be late as we will give priority to the people who actually responded.

We will be asking more people this week for payment. Our production numbers are only about 20-50% of a normal month currently so bigger orders like ebereon's can take most of a weeks production in one hit. The next 2 weeks are likely to be worse as there is a big public holiday here on Monday and many people take holidays that week or even the next one. We are very down on staff numbers over those 2 weeks.