GekkoScience has a new pod miner, just in time for Christmas

[cygan]

My only theory right now is that they get slower as the heat up and the heatsink becomes fully soaked. Not sure about that, though.

I had the same thought when I saw the hashrate increase after the slight undervolt at 550MHz. The avg hashrate is at 2.12Th/s by now, I think I will just leave it running like this for a few days to see if it goes up even further yet.

That's impressively high for just 550MHz. After a few hours of running 700MHz, mine seems to have settled on around 2.3Th/s, after starting at ~2.5Th/s... Still not sure if this is temp related.

my 2 r909 miners have been running for ten days now and the performance at 680 and 640mhz is impressive
both pod miners have settled between 2.34 and 2.45 th/s

Code:

4: GSF 10070009: BM1397:06+ 680.00MHz T:680 P:662 (3:2)    | 90.1% WU: 90% | 2.838T / 2.457Th/s WU:34332.8/m
5: GSF 10070003: BM1397:06+ 640.00MHz T:640 P:630 (3:2)    | 91.0% WU: 91% | 2.128T / 2.343Th/s WU:32744.4/m

[1701988064]

1701988064

[1701988064]

1701988064

[gergar]

Hello all, i want to share my experience and some advise on cooling as i found out points to improve:
...
3 inverted the fan to pull the air, why, the current setu makes hot air stays inside the case (measured with a temp camera)
...

Hello, what was the temperature difference you have measured with your temp camera when the (stock?) fan was blowing out? Would be interesting to know.

Hello,

to make easier some definitions first, 550Mhz as i tested with my gekko compactf settings temps in celcius

FP= Top Heatsink (front plate)
PCB
BP = bottom Heatsink (back plate)
EX = exhaust (stock = front / modified = back)

Stock
FP = 43 / with all bolts (50)
PCB = 54 / with all bokts (47)
BP = 27 / with all bolts (30)
EX = 48 / with all bolts (50)

Modified
FP = 53
PCB = 35
BP = 45 (temp at front = 21, two degrees lower then ambient 23)
EX = 44

what it means?, the heat was mostly staying at the PCB as the heatsinks were not fully operational, once i adjusted bolts temp lowered some how but still the lower heatsink was mot doing clean contact with the pcb (air in between. But the air as it was not strong enough to move out stayed there in the front near the cables. with the changes the backplate is draining more heat from the pcb, pcb is colder and top maintained temps so contact issue was the initial problem. Then all front part is about ambient temperature, as air gets higher speed it is 2 to 3 degrees lower just cos pressure changes, and the air going out of the fan and case is hot.

I am running it at 650 now and here the temps i just captured as i was intrigued on how much temp gets from 100Mhz more:
FP= 66 (some are hotter 69-72)
Pcb = 36-37 relative cold
BP= 50 (seems it still gets heat out)
EX = 49 - 52

I assume with current stock fan might get too hot to go higher on freq, not sure the asic temperatures down there but they should be extremely hot. 650 had maintained 24 hrs at 2.6T stable (21 degree temp intake, maybe cooler temps can allow it to go higher on freq. - mine anything higher than 650 starts to bounce back to 650-646 and i get lower hashrates.

[kano]

Actually, what matters most is the heat of the BM1397 chips.
If that heat is going away, to other components, then that's good.

One way to measure this is the W usage, since it will go up, at the same frequency, if the chip is heating up.

[gergar]

Actually, what matters most is the heat of the BM1397 chips.
If that heat is going away, to other components, then that's good.

One way to measure this is the W usage, since it will go up, at the same frequency, if the chip is heating up.

Yes and i have done that to calculate the expected BTUs etc from the asics, what i found was about a linear growth, 50w per 100Mhz of freq increase and about 10+ degree increase. i am preparing some thermistors to check temps at bottom on art of the heatsink that i assume might be closer to the asics if there is good thermal transfer.

have another question on the Api, i sent via python a post message with command=ascset parameter={0,freq,400} but it returns “Unknown Option: freq” is there any documentation. on the payload to be sent using json? i used the page that explains in summary the api but as it uses | i am not quite clear how will it read using json command/parameter format.

[n0nce]

@gergar What do you think about using 2 fans (1x intake, 1x exhaust)? I mean, the downsides of having ports on the back now is that it's not really possible, but maybe a small fan that covers part of the back side could help.
Or using 2 fans in the front. That did wonders on my Apollo.

[gergar]

@gergar What do you think about using 2 fans (1x intake, 1x exhaust)? I mean, the downsides of having ports on the back now is that it's not really possible, but maybe a small fan that covers part of the back side could help.
Or using 2 fans in the front. That did wonders on my Apollo.

it will help but in reality the way cables are today might need very small fans and to have a proper push-pull setup fans must be similar in cfm and pressure capacities (i tried some noctua 40x25mm and are large to fit) so i ordered som 6k 80mm fans from amazon one 25mm and other that is 38mm but cfm is 3x the current but noise will the the issue with these but i want to test how will it help.

this is the 38mm
https://www.amazon.com/Wathai-Cooling-80mm-Speed-Bearing/dp/B07QX3QCP5
this is the 25mm that fits with mo issues but you need in both an adaptor (link also included)
https://www.amazon.com/GDSTIME-80mm-25mm-Brushless-Cooling/dp/B00N1Y50QQ
the adaptor
https://www.amazon.com/Mainboard-Standard-Extension-Power-Cable/dp/B07Z8Z182S

once i get them netx week let me test it and i can report back so you don’t throw away money … i can try to set a push-pull using the original fans i just need to get some very large bolts/nuts to use the top bolts from the current fan to secure the front ones just i need to do some measuring to avoid bending too much the power and usb cable as i am using an atx 1600w pc psu and the pcie cable doesn’t bend too much.

hope it makes sense to you.

[sidehack]

Does push-pull really matter in a system without much flow constriction? My understanding was that helped maintain airflow in an environment with lots of obstruction, like close-packed heatsink fins, but the R909 is pretty open.

[100knot2dae]

I also replaced the stock fan with the Noctua NF-A8 today, and by that time I checked the heatsink screws as well. Almost all have been at least slightly loose and needed some re-tightening.

As a result of both, the unit is now hashing with 2.15TH/s with the undervolt at 550MHz, and the increased airflow helps to save a few more Watts. 

[n0nce]

Does push-pull really matter in a system without much flow constriction? My understanding was that helped maintain airflow in an environment with lots of obstruction, like close-packed heatsink fins, but the R909 is pretty open.

I'm not sure about that myself.
Earlier today I tried 550MHz with the NF-A8 installed and the stock fan placed in front of it, it seemed to make no difference. I remember someone trying this in a different application and they had the same results. Somehow, it was effective on the Futurebit Apollo, though. However that device was exhausting up top, with the fan pulling air through a pretty dense heatsink. So by the time it reached the top, the airspeed was quite slow. Pretty different scenario.

After those experiments, I placed the stock fan in the back, covering roughly half of the back area and set the miner to 700MHz. (I know; not very scientific. I should stick to 550 now, for further experiments).
Anyhow, I'm now seeing 2.6Th/s, something I have never experienced with this miner before.

Code:

0: GSF 10070017: BM1397:06+ 700.00MHz T:700 P:694 (3:2) | 91.6% WU: 93% | 2.507T / 2.613Th/s WU:36505.2/m

Let's see if it can keep this up. So far, that's a 4:30h average. I'm also interested to see if the hashrate drops off over the next 24h or if this double fan setup keeps it above 2.5.

[gergar]

Does push-pull really matter in a system without much flow constriction? My understanding was that helped maintain airflow in an environment with lots of obstruction, like close-packed heatsink fins, but the R909 is pretty open.

So far with the size and fan type I don't expect any benefit but swapping to exhaust the air vs push inside did a difference (also adding the thermal path to the backplate)... with High CFM/Pressure might help, BUT heatsink design on 909 is not for that type of work, so I doubt.

I will test it so we can see if works but what I decided based on the construction of 909 was to set a stronger pull fan as I see that will be enough... at the end these are 150W tops devices and to make it push-pull cable needs to be routed differently to make the fan on top no "gaps" and create a tunnel at the 909 chassis. Also power consumption won't be great if we add fans and 6k fans pulls a lot of power at full speed. To me is a balance on the lower power possible and the highest hashrate possible, which I just haven't found it yet...

I don't see it worth to change design for push-pull, I would better look for a pipe/type dissipator/heatsink with very close fins like the ones used at server CPUs, that will maximize the airflow that passes. I am trying to find in the market any CPU like heatsink that I can try on it before I convert them to water cool as I have the compactF (just learning if the current air-cooling can get improved, but my end point is water-cooling it as summer is very hot here and air density is quite poor).

But well it is all about learning and experimentation so couple of usd worth it (at least to me).

I am using some alpha cool memory water cooling same as for the Cf to cool them down, I am working on the holders to leverage the 6 bolts and the backplate which is a great add BTW!  so doing some design/testing to assure PCB does not bend. I also ordered some SSD and custom memory watercolors to see if they fit better than the small Alphacool NDXs. 

I will test before I move them to water and share if it improves... for now 650Mhz @ 2.6T for 48hr and heat is well managed by the current design just with the few mods I did. Since the new 1397AI chips I ordered  have not arrived yet for repairs I will stay away from 650+ speeds as I need these for testing.

you both You and Kano had made an amazing product !!! thank you!!

[gergar]

Does push-pull really matter in a system without much flow constriction? My understanding was that helped maintain airflow in an environment with lots of obstruction, like close-packed heatsink fins, but the R909 is pretty open.

I'm not sure about that myself.
Earlier today I tried 550MHz with the NF-A8 installed and the stock fan placed in front of it, it seemed to make no difference. I remember someone trying this in a different application and they had the same results. Somehow, it was effective on the Futurebit Apollo, though. However that device was exhausting up top, with the fan pulling air through a pretty dense heatsink. So by the time it reached the top, the airspeed was quite slow. Pretty different scenario.

After those experiments, I placed the stock fan in the back, covering roughly half of the back area and set the miner to 700MHz. (I know; not very scientific. I should stick to 550 now, for further experiments).
Anyhow, I'm now seeing 2.6Th/s, something I have never experienced with this miner before.

Code:

0: GSF 10070017: BM1397:06+ 700.00MHz T:700 P:694 (3:2) | 91.6% WU: 93% | 2.507T / 2.613Th/s WU:36505.2/m

Let's see if it can keep this up. So far, that's a 4:30h average. I'm also interested to see if the hashrate drops off over the next 24h or if this double fan setup keeps it above 2.5.

What I think can help is make the fan to pull / extract the air use the original as it is stronger than Noctua (Arctic label pointing the PCB will make it to extract --- I checked their specs as I have Noctuas too) and add thermal pads to the bottom heatsinks 1mm @ 13+w/k will have the highest impact in it, those few changes will give you 650@2.5-2.6T stable on a ~120W on stock voltage.

[n0nce]

I will test before I move them to water and share if it improves... for now 650Mhz @ 2.6T for 48hr and heat is well managed by the current design just with the few mods I did. Since the new 1397AI chips I ordered  have not arrived yet for repairs I will stay away from 650+ speeds as I need these for testing.

That sounds great! I will try rotating the fan, too then.

For what it's worth, my experiment with a loosely positioned 'exhaust fan' did not last long; 8h in, the hashrate dropped to 2.4Th/s again. I'm not sure why, but my chips perform well, for a few hours and then slowly drop further and further. It must be thermally related.

add thermal pads to the bottom heatsinks 1mm @ 13+w/k will have the highest impact in it, those few changes will give you 650@2.5-2.6T stable on a ~120W on stock voltage.

Just to be clear: the bottom heatsinks just cool the PCB? And by default are bolted on without any thermal interface (pads or anything else)?
I'd like to prevent touching the top heatsinks, since I believe there's a chance to rip a thermal pad if you don't screw the heatsink on perfectly straight, which would then connect grounds and lead to magic smoke.
But bottom heatsinks are a mod I may actually do. I even have 12W/mK 1mm pads here.

[tcj2001]

I have a compac-f getting 348.5 Gh/s and R909 getting 2.457 Th/s
What suggest-diff i should use for these individually.

I saw some calulation suggesting difficulty of 1 per 1 Gh/s / 1.8 and another one suggested  Gh/s * (71 * .05)^2

Any suggestion?

[os2sam]

I have a compac-f getting 348.5 Gh/s and R909 getting 2.457 Th/s
What suggest-diff i should use for these individually.

I saw some calulation suggesting difficulty of 1 per 1 Gh/s / 1.8 and another one suggested  Gh/s * (71 * .05)^2

Most pools adjust diff to 18 shares per minute regardless of what your suggest diff is.  It is just a suggestion after all.

[tcj2001]

So then I think its okay to set suggest-diff of 500 for compac-f and 5000 for r909, as pool will takes care of it and adjust it it accordingly. BTW i am using ckpool

[os2sam]

So then I think its okay to set suggest-diff of 500 for compac-f and 5000 for r909, as pool will takes care of it and adjust it it accordingly. BTW i am using ckpool

It will start at your suggest diff and then adjust to 18 shares per minute.  It doesn't really matter what you set it to, pools will adjust it automatically.

Pools also now have a minimum diff so that ASIC's won't flood them with diff1 shares.

[tcj2001]

So then I think its okay to set suggest-diff of 500 for compac-f and 5000 for r909, as pool will takes care of it and adjust it it accordingly. BTW i am using ckpool

It will start at your suggest diff and then adjust to 18 shares per minute.  It doesn't really matter what you set it to, pools will adjust it automatically.

Pools also now have a minimum diff so that ASIC's won't flood them with diff1 shares.

Pardon my ignorance, I have attached the screen shots of cgminer for both of these units
Compac-f https://imgur.com/Da9dfnc
R909 https://imgur.com/xPDxaDl

I believe the WU tells the diff1 and these are very high compared to 18.

[gergar]

I will test before I move them to water and share if it improves... for now 650Mhz @ 2.6T for 48hr and heat is well managed by the current design just with the few mods I did. Since the new 1397AI chips I ordered  have not arrived yet for repairs I will stay away from 650+ speeds as I need these for testing.

That sounds great! I will try rotating the fan, too then.

For what it's worth, my experiment with a loosely positioned 'exhaust fan' did not last long; 8h in, the hashrate dropped to 2.4Th/s again. I'm not sure why, but my chips perform well, for a few hours and then slowly drop further and further. It must be thermally related.

add thermal pads to the bottom heatsinks 1mm @ 13+w/k will have the highest impact in it, those few changes will give you 650@2.5-2.6T stable on a ~120W on stock voltage.

Just to be clear: the bottom heatsinks just cool the PCB? And by default are bolted on without any thermal interface (pads or anything else)?
I'd like to prevent touching the top heatsinks, since I believe there's a chance to rip a thermal pad if you don't screw the heatsink on perfectly straight, which would then connect grounds and lead to magic smoke.
But bottom heatsinks are a mod I may actually do. I even have 12W/mK 1mm pads here.

the pcb is taking the back heat from the asics, they come with no thermal interface on them, only air therefore the heat stays and gets into the pcb/components including asics.

what i usually do is change both up and lower but upper vs what you have are better 20w/k once you lose the bolts the bottom will fall, do 1 at time just be careful of not move them much.

overall that will help the Asics to get cooler same PCB as also power controllers getting hot will lose capacity and cause performance issues, so keeping all as cool as possible pays out in higher room to push hard the asics and stability of other components.

hope makes sense to you.

[100knot2dae]

What I think can help is make the fan to pull / extract the air use the original as it is stronger than Noctua (Arctic label pointing the PCB will make it to extract --- I checked their specs as I have Noctuas too) and add thermal pads to the bottom heatsinks 1mm @ 13+w/k will have the highest impact in it, those few changes will give you 650@2.5-2.6T stable on a ~120W on stock voltage.

Hm, why do you think the stock fan is stronger? When comparing the specs between the Noctua NF-A8 and the Arctic F8, I see the Noctua better in every discipline. Slightly higher RPM, better airflow and MUCH better static pressure. The latter of which makes it an even better candidate for blowing air into the unit (the original direction).

[os2sam]

Pardon my ignorance, I have attached the screen shots of cgminer for both of these units
Compac-f https://imgur.com/Da9dfnc


I believe the WU tells the diff1 and these are very high compared to 18.

Are you expecting the suggest-diff command to have an effect on the Worker Utility?

In your screen shot it looks like your suggest-diff of 500 is being honored.  So that equates to 9.94 shares per minute being submitted to the pool.