Antminer s5 increasing HW error with overclock

[dsh]

Hello

It is preferential to run Antminer at 52C-56C degrees. However, at overclock of 381.25 MHz it is only 52/53 and already it spikes in HW error up to 0.0324%. All parts purchased used.

Is this an issue with the PSU? Are the HW error standard at this frequency?

Below are more details.......
Cheers!

Miner: Antminer S5
PSU: Corsair RM750, Model #75-001937, Part #CP-902005S
UPS: APC Back-UPS XS-1500, 1500VA, 865 Watts
One stock fan operating at 3000 RPM. 2x 1080 RPM fans pushing air up from below the machine.
Fan is set to operate at 40% via miner configuration because of noise (placement in living space).
Ambient temperature is currently 67F.

At
350 MHz and below HW error is 0.0000%

At
356.25 MHz
Diff1#.0027% HW error after 35 minutes
UPS: ?? (did not check load values.....)
Temp: 50/51C

At
368.75 MHz
Diff1#.0092% HW error after 30 minutes
UPS: 590watts ~68% load
Temp: 51/52C

At
381.25 MHz
Diff1#.0324% 54 HW error after 15  minutes
UPS: 610watts ~70% load
Temp: 52/53C

At
387.5 MHz
Diff1#.0553% 247 HW error after 25  minutes
UPS: ?? (did not check values.....)
Temp: 52/53C
Fan speed: 2880 stock fan, 1080 2x support fans (decreased)

[VirosaGITS]

Hello

It is preferential to run Antminer at 52C-56C degrees. However, at overclock of 381.25 MHz it is only 52/53 and already it spikes in HW error up to 0.0324%. All parts purchased used.

Is this an issue with the PSU? Are the HW error standard at this frequency?

Below are more details.......
Cheers!

Miner: Antminer S5
PSU: Corsair RM750, Model #75-001937, Part #CP-902005S
UPS: APC Back-UPS XS-1500, 1500VA, 865 Watts
One stock fan operating at 3000 RPM. 2x 1080 RPM fans pushing air up from below the machine.
Fan is set to operate at 40% via miner configuration because of noise (placement in living space).
Ambient temperature is currently 67F.

At
350 MHz and below HW error is 0.0000%

At
356.25 MHz
Diff1#.0027% HW error after 35 minutes
UPS: ?? (did not check load values.....)
Temp: 50/51C

At
368.75 MHz
Diff1#.0092% HW error after 30 minutes
UPS: 590watts ~68% load
Temp: 51/52C

At
381.25 MHz
Diff1#.0324% 54 HW error after 15  minutes
UPS: 610watts ~70% load
Temp: 52/53C

At
387.5 MHz
Diff1#.0553% 247 HW error after 25  minutes
UPS: ?? (did not check values.....)
Temp: 52/53C
Fan speed: 2880 stock fan, 1080 2x support fans (decreased)

Probably the PSU. My crap Gold Raidmax bring the error rate to 0.3% at 393.75mhz and the hashrate is very low, like 1.23TH/s. My corsair does 0.01% HW and 1.27/TH/s. My EVGA g2 and platinium server PSU does 0HW and 1.3TH/s.

I also have one S5 that does 400Mhz 0HW and 1.32TH/s.

[dsh]

Probably the PSU. My crap Gold Raidmax bring the error rate to 0.3% at 393.75mhz and the hashrate is very low, like 1.23TH/s. My corsair does 0.01% HW and 1.27/TH/s. My EVGA g2 and platinium server PSU does 0HW and 1.3TH/s.

I also have one S5 that does 400Mhz 0HW and 1.32TH/s.

Nice....
Can you recommend the EVGA G2 for overclocked Antminer? Thinking that it would be nice to get 400-500 MHz without HW errors...... 

[VirosaGITS]

Probably the PSU. My crap Gold Raidmax bring the error rate to 0.3% at 393.75mhz and the hashrate is very low, like 1.23TH/s. My corsair does 0.01% HW and 1.27/TH/s. My EVGA g2 and platinium server PSU does 0HW and 1.3TH/s.

I also have one S5 that does 400Mhz 0HW and 1.32TH/s.

Nice....
Can you recommend the EVGA G2 for overclocked Antminer? Thinking that it would be nice to get 400-500 MHz without HW errors......  

EVGA G2/GS, P2 too but its pretty expensive. (Not G1)

Most Seasonic PSU's, Seasonic OEM, superflower OEM. I dont think you can do 408-500Mhz without overvolting, which you would need to do at the PSU. So you would need to get something else.

I don't even know why one of my miners take 400mhz. Logically this means the chip voltage is higher, but i'm not feeding it any current any higher than 12.0X~ I also don't know if there is a danger to raising the voltage too high, aside from more heat. Maybe RichBC can answer that one.

[-ck]

You increase the clockspeed past its specified speed and you're surprised that it misbehaves progressively more the further you push it? Temperature isn't everything...

[VirosaGITS]

You increase the clockspeed past its specified speed and you're surprised that it misbehaves progressively more the further you push it? Temperature isn't everything...

No, we know HWE% go up because the core voltage is too low for the clock. So we know why its "misbehaving". To drop the HWE% drop back to 0, you raise the core voltage.

So the question is, beside more heat, what other negative consequence there could be by raising the voltage too high? Theory; some components may not be build to handle the extra current, thus even if the chip would be just fine, something else could break.

[RichBC]

Figures look good to me and what I would expect from a typical S5. As you go above the Default / Specified frequency you are finding out to what degree your S5 exceeds it's specification, some are better than others. I find HW error rates of up to 0.005% to be acceptable with no measurable effect on the Hash Rate.

As VirosaGITS has said Overclock performance also depends on the exact voltage your PSU is supplying and how well it holds that voltage with the increased current taken by overclocking?

If you have a multimeter measure the voltage at the PCI-e connectors, ideally for overclocking you need a solid 12.1V - 12.2V if running at 400MHZ and above. You also then need to keep a close eye on the temperature.


Rich

[chiliminer]

I can tell you this, I had a gold rated Corsair 850W PC PSU and my S5 would throw out a lot of errors, still acceptable, but just a lot of errors.  I switched the PSU to HP 379123-001 Proliant DL380 G5 DPS-800GB, the 800W server PSU's, you can get them for about $15 on eBay.  Now I overclock mine to 375M with a hash rate of around 1.24Th/s and I see just a couple of errors a day, so my actual error rate shows up as 0.00000%.  So the PSU does make a big difference.

I couldn't believe it myself, as I thought the PC PSU would be fine, but the error count goes up into the 1000's over a couple of days, and with that DPS-800 I can count the number of errors in a couple of days on two hands.

Chili

[RichBC]

Some Miners will be more Voltage sensitive than others. For instance the S1 & S3 use Buck Converters to generate the Core Voltage, because of this they will compensate for variances in the supply voltage & protect you from problems due to voltage.

Other Miners like the S5, S7 & Avalon6 supply the ASIC chips connected in a string (Like Christmas Tree lights) The Core Voltage is then determined by the Supply Voltage divided by the Number of Chips in the string. So now dependant on the exact supply voltage you will get a different core voltage.

That supply voltage although nominally 12V is dependant on several factors. First is the voltage it is set to with no load. Next as you load the Power Supply that voltage will often change dependant on how good the PSU is and it's ability to regulate the voltage. Third and often forgotten the wires and connectors connecting the PSU to the miner have resistance and the higher the resistance the more the voltage will drop with load.

So the bottom line is. If you play with Miners buy a Multimeter. Then you can check the voltage that is actually at the PCIe connectors. I think that many Miner problems are PSU / Voltage related, and when presented with HW errors and other unexplained problems it's one of the first things to check.


Rich

[philipma1957]

Some Miners will be more Voltage sensitive than others. For instance the S1 & S3 use Buck Converters to generate the Core Voltage, because of this they will compensate for variances in the supply voltage & protect you from problems due to voltage.

Other Miners like the S5, S7 & Avalon6 supply the ASIC chips connected in a string (Like Christmas Tree lights) The Core Voltage is then determined by the Supply Voltage divided by the Number of Chips in the string. So now dependant on the exact supply voltage you will get a different core voltage.

That supply voltage although nominally 12V is dependant on several factors. First is the voltage it is set to with no load. Next as you load the Power Supply that voltage will often change dependant on how good the PSU is and it's ability to regulate the voltage. Third and often forgotten the wires and connectors connecting the PSU to the miner have resistance and the higher the resistance the more the voltage will drop with load.

So the bottom line is. If you play with Miners buy a Multimeter. Then you can check the voltage that is actually at the PCIe connectors. I think that many Miner problems are PSU / Voltage related, and when presented with HW errors and other unexplained problems it's one of the first things to check.


Rich

When running at freq above norm (overclocking)  The s-5 the s-7 the avalon 6 all like hot psu's  ie 12.1 volts not 11.9 volts .

If you mine with the 3 above or a s-5+  keep that in mind.

BTW of all these units only the avalon 6 tells you the input voltage  so it is the only one that you do not need a meter to know.