Hacking the S7 - improving efficiency through minor hardware manipulation

[IITravel01]

OK, here's something interesting I've found with the S7-LN.  Others can verify.  Since the exhaust of the Enermax 1000w Gold power supply is pointing forward I switched the fans of the S7-LN around so it also exhausts to the front.  The temp. sensor must me in the back because before I let it run for 15 minutes at 512.5M is settled between 47 and 48 degrees C and the fans ran at 2,880 rpm.  Now that they're facing the other direction the temp sensor is showing 40-43 degrees C and therefore the fans are only running at 2,400 rpm which is noticeably more quiet, which probably shaves a watt off the usage also.  If you want it quieter, I suggest you try it.  Does it lower the temp, probably not and it's probably the temp sensor at the rear so with the fans in the original position the hotter air is going over the sensor, with them reversed the temp. sensor is probably reading the cooler air.

Also switching the 1000w Gold Enermax to a 1000w Titanium EVGA is running about 11w lower at the wall.  From 508w to 497w.

Update:  The 1st S7-LN runs stable at 525M, anything higher starts throwing errors and losing ASICs.  At 525M with the EVGA 1000w Titanium it's pulling about 509w at the wall at about 2.4TH/s with the fan at 2,520rpm and the temp. showing 41-45 C with no HW error after 15 minutes.

Update 2:  The 2nd S7-LN runs stable at 512.5M, but it's running about 0.036 HW errors at that speed.  On the Enermax 1000w Gold power supply it was pulling about 501w at the wall at 512.5M, switching to a 850w Gold power supply it pulling 519 watts at the wall (461w after the power supply according to a display showing 90.7 efficiency), and the S7-LN is still showing about the same error rate so it's not the power supply.  Same temps and fan speed as the 1st with the fans turned around.  (Noticeably more quiet, might be more than just the slower fan speed, I think in the rear there's space between the fan the heatsinks, in the front the heatsinks are right behind the fan, this space may be causing the lower fan volume)  It's doing about 2,260 TH/s for the first half hour.  Update:  I dropped this one down to 506.25M and the HW errors are about 0.025 with it doing about 2.25 TH/s pulling 453w after the power supply.

So both pull about 1,030w-1,035w at the wall with 4.650 TH/s.  Not bad as this is about the same as four S5's that would pull about 2,400w or close to one 3 blade S7 pulling about 1,430w at the wall (if both could do 525M with 0.00% HW errors then it would match a full S7 but only pulling about 1,040w, that's 400w less or almost a 30% reduction in power used).

I ordered a SilverStone 800w Titanium power supply for the 2nd S7-LN ($170 from Newegg.com, I've ordered this power supply twice from Amazon.com and they shipped the 750w Platinum version so I think they may have the UPC code mixed up in their system) so it should drop the wattage down around 10w less.

As for manually setting the fan speed.  If you do that, will the fan speed increase when the ambient room temperature goes up?  With the auto mode, it does.  For example when I did the initial settings the fans ran at 2,520 but now the room has gone up to 90 F and the fans have increased to 2,640 rpm.

[sidehack]

There's also an option to manually set the fans to whatever speed you want.

[in2tactics]

There's also an option to manually set the fans to whatever speed you want.

True, I generally just lower the fan speed until I notice an increase in temp. Trial and error, it works most of the time.

[philipma1957]

There's also an option to manually set the fans to whatever speed you want.

have you received  second batch from dhl?

[sidehack]

Yeah, customs hangups. I got it squared away this morning but, even if it had been in time to make the truck, I wouldn't have been at the shop to receive them. Units should be in my hands tomorrow morning and going out Thursday afternoon. Sorry about the delays.

[philipma1957]

Yeah, customs hangups. I got it squared away this morning but, even if it had been in time to make the truck, I wouldn't have been at the shop to receive them. Units should be in my hands tomorrow morning and going out Thursday afternoon. Sorry about the delays.

happens at least they are near ready to ship out.

[sidehack]

Yeah, but the last tracking update is still "shipment on hold" from yesterday morning and the local route usually would have the boxes dropped off at least an hour ago. Looks like the ball is dropped again.

UPDATE - apparently word that fees were paid yesterday at 8AM didn't get to the depot until this morning at 10AM after the truck was already out. So, packages will be delivered tomorrow and miners will go back out Friday.

Sometime in the next couple weeks I think I'm gonna work on a snap-in hardware solution to changing voltages. I don't know enough about the software on the S7 controller to have an idea how to write a program tapping into the TX lines, but I'm confident I can do it externally. Could be interesting.

[HagssFIN]

I got a Pickit 3 from Ebay and did a little tweaking to use a Enermax Revolution 87+ 1000W PSU (from my old S2) with Antminer S7, B9.

Here are the results:
Antminer S7
Hash rate: 4.390Th/s (40min avg)
Voltage setting: 640/670mV  
Frequency: 650 Mhz

Power supply: Enermax Revolution 87+ 1000W
Voltage: 227 VAC
Current: 4,9 A
Power (at wall): 1112,3 W
Power (from the 12VDC, approximately, with 0.87 efficiency): 967,701 W

Meter: HT Instruments HT7011, used for both voltage and current measuring

[IITravel01]

What's the original default hot starting voltage/cold running voltage for the blades on the S7 (not the S7-LN)?

[citronick]

@sidehack

I have an existing S7-LN, can I apply your mod to get the lowered wattage?

Is it a firmware mod?

If HW mod, then I will out of luck with getting this S7-LN down to lower power consumption.

Thanks

[sidehack]

This thread contains all the answers to those questions.

[aarons6]

@sidehack

I have an existing S7-LN, can I apply your mod to get the lowered wattage?

Is it a firmware mod?

If HW mod, then I will out of luck with getting this S7-LN down to lower power consumption.

Thanks

most of the how-to info is on this post.
https://bitcointalk.org/index.php?topic=1504228.msg15298837#msg15298837

[percy_tc]

would be great to have a video about the process.

[BryanC]

Sidehack -
You are a gentleman and a scholar. Applied your firmware to my S7-LN, with these results:

Before:
~2700 GH/S
Temp: 48-49
Fans: ~2800
Watts: 724


After:
~2050 GH/S
Temp: 42-44
Fans: 2280
Watts: 430

Watts measured with a belkin in-line meter.

That is brilliant. Daily power cost drops from $1.91 to $1.14 and overall profitability more than doubles.
[edit]Check your tip jar![/edit]

[BryanC]

So, additional question: Intermittently, Chain 1 has shown up with all xxxxx, or with 1 chip x'd. Disconnecting the ribbon cable was a real mother*ucker - could damage to the control cable account for that? Again, it's intermittent. Power cycling will result in one of those error states, or a non-error state. It seems that if I power down and "fiddle" with the cable, it comes back up in the non-error state.

So, that's a point for the DIYers on the S7-LN - it would be better to remove the power supply and have plenty of room to work the control cable off while supporting the socket to ensure you don't damage the cable or the socket.

[BryanC]

... And now it's gone totally wonky, with one board showing 33 chips and the other showing 43. Looks like I have some more work to do!

[alh]

... And now it's gone totally wonky, with one board showing 33 chips and the other showing 43. Looks like I have some more work to do!

Can you revert to the original Bitmain setup? I'd try back there to see if it's related to your changes, or something else.

[BryanC]

Yes - I dumped the existing firmware before writing the new one. Will give that a shot tonight.

... And now it's gone totally wonky, with one board showing 33 chips and the other showing 43. Looks like I have some more work to do!

Can you revert to the original Bitmain setup? I'd try back there to see if it's related to your changes, or something else.

[BryanC]

Ohhhhkay.
Reverted to the original firmware, and everything behaved normally.
Programmed to 650/620, and the strange behaviour returned (showing less than 45 chips per chain - not showing xxx's, but just showing fewer ooo's)
Programmed to 660/630 @ 550m, and it was wonky for the first power cycle (hundreds of HW errors), but after another power cycle it's running with zero errors (one error after another 10 minutes), about 2500 GH/s drawing 560 watts.

I guess that, for whatever reason, these boards won't tolerate the 650/620.

(I would post screenshots, but I'm assuming I don't have that feature until I'm out of "Newbie" status)

... And now it's gone totally wonky, with one board showing 33 chips and the other showing 43. Looks like I have some more work to do!

Can you revert to the original Bitmain setup? I'd try back there to see if it's related to your changes, or something else.

[sidehack]

I know 610 and 600 are specifically labeled "unstable" in the chart, but it's also noted in places that 620 isn't always reliable either. A lot of the group-buy miners weren't stable until 630 or 640. I did each of those on a per-board basis. I've got an S7 right now that has been dropping one board at like 670 for 550MHz so it really depends on the hardware. If a board runs unstable or drops out, the voltage is too low.

Know that the S7LN restarts cgminer. It runs for about a minute to warm up the chips (during which time you'll see hundreds of errors) and after that it should kill and restart clean. The firmware shifts to cold running about 30 seconds into the restarted process. At least that's how it should work - sometimes cgminer doesn't restart and you're left with the jillion errors.