Avalon stable @ 98,986Mhash/sec

[SebastianJu]

*lol* 450MHz... and with such "old" technique. I calculated KNCMiner to be half the price (including vat) per GH instead using avalon chips + assembly. Now with this overclocking the advantage of KNCMiner is gone, taking into account the longer months until the miner can be received. BFL is fully out of the game and the their credits are worth nothing. And since the GH of bitfury is similar to kncminer its the same there like with kncminer.
Of course it has to be shown whats the overclocking capability of bfl, bitfury or kncminer chips. But its a bit funny that the old technique is nearly as good as the new one. Probably there is a reason why avalon and bitcountain chose to take the old techniques first.

[pikeadz]

This is definitely a game changer if they run stable at that rate.

[cardcomm]

Have anyone tried to completely submerge Avalon into mineral oil? That works very well for desktop computers.

Fluid would be too viscous for the fanz...

I've got a system that's submerged in mineral oil (quad E5-4650).  Transformer oil to be specific (lower viscosity, higher flash temps, mostly synthetic, no PBAs).   Fans don't really care, although that's not the primary circulation method (pump).

Nice. I've been wanting to do this to a computer for a while (I think my i7-2700k would over-clock like a MoFo.)

I'm pretty sure I'll be trying this with my K16 boards. Where did you source the mineral oil, if you don't mind me asking? Also, what type of container are you using? I've seen debates on using a plexiglass enclosure, and possible ill effects on it from the mineral oil.

Thanks

[johnyj]

was wondering how long it'd take people to notice ( and more importantly share the constant that we've released on github.)

the number you are all aiming for is 450 of course, that's not really possible on just air cooling.

Do you think that the batch 1 design power distribution can manage the increased power draw at those speeds?

This is a concern, normally for overclock you need strong power supply module, not just only a large PSU. Someone already burnt one of the hash submodule even under 282Mhz, but that could be caused by a bad component. Anyway, I found out that hash rate tends to jump between 50Gh and 100+Gh wildly when heavily overclocked (350Mhz), not a good sign

[Humax]

Batch 2.  Active cooling + overclock. Each tweak I apply bumps the GHash.

Now I'm shooting for 100Ghash/sec. Who is with me?

I'd love to join you, but I don't got anything to shoot with! 

[Tehfiend]

I am testing this and I am seeing only a very small increase.  Maybe 1 Ghs/s when temp 3 is dropped from the low 50's to low 40's C.  


Unit with A/C
1h 37m 58s  83896.29  temp3 43    freq(auto) 354


Unit without A/C
6h 54m 22s  83111.32   temp3 53    freq(auto) 353

You're forgetting to overclock when running the AC

[PuertoLibre]

What happens if you replace the original paste with Artic Silver 5 or other similar compounds such as high quality thermal pads?

[johnyj]

What happens if you replace the original paste with Artic Silver 5 or other similar compounds such as high quality thermal pads?

AS5 is conductive, if it flows back to the chip through those small tunnels, who knows what will happen. MX4 is non-conductive and also cheaper if you want to apply for all 240 chips (a 20g tube is enough)

[PuertoLibre]

What happens if you replace the original paste with Artic Silver 5 or other similar compounds such as high quality thermal pads?

AS5 is conductive, if it flows back to the chip through those small tunnels, who knows what will happen. MX4 is non-conductive and also cheaper if you want to apply for all 240 chips (a 20g tube is enough)

Any positive benefit to applying MX4 on the chips?

[67152398]

Congratulations

[el_rlee]

Have anyone tried to completely submerge Avalon into mineral oil? That works very well for desktop computers.

Fluid would be too viscous for the fanz...

I wonder if there is a water block out there that would fit...
100GH/s with good cooling, might be worthwhile. 

https://bitcointalk.org/index.php?topic=211825.0

[rograz]

Is there enough spacing between the modules to put gpu ram sinks on each individual chip? I have about 2 kg of them, yes kilograms  

Infact just checked how much the bag of them weigh, it was close to 4 kilograms so I should have plenty for 1x3 module avalon once I get it.

[scyth3]

Has anyone verified that 98GH/s is stable 24/7?

[jspielberg]

Has anyone verified that 98GH/s is stable 24/7?

I think the highest people have gotten stable with stock B2 is 350MHz ~ 84GH/s on 3 module system.
98GH/s on a 3module system is ~410MHz which I don't think anyone has been stable on.

98GH/s on a 4 module system is ~310MHz which should be fine.
I don't believe that MagicMike523 has stated in this thread what frequency the chips are running at or how many modules.

My guess is that the 98GH/s is lucky variance with a 3 module running at 375MH/s ~90GH/s with some lucky stability and lots of cooling.

[TheSwede75]

Water cooling? Mineral oil? Just wrap the blades in applewood smoked bacon and enjoy a slower but tastier Avalon! 

[Bitcoinorama]

Water cooling? Mineral oil? Just wrap the blades in applewood smoked bacon and enjoy a slower but tastier Avalon! 

Mmmm...nom, nom, nom.

[PuertoLibre]

Is there enough spacing between the modules to put gpu ram sinks on each individual chip? I have about 2 kg of them, yes kilograms  

Infact just checked how much the bag of them weigh, it was close to 4 kilograms so I should have plenty for 1x3 module avalon once I get it.

I was thinking people should replace the thermal compound used in the Avalon unit. Considering there are some people who found out some units have no thermal compound or had it improperly applied.

Assuming that the thermal compound is crap and/or low efficiency, then transferring the heat from the chips to the heatsink would probably help some at higher clocks. A good number of thermal pads with 6 or 7watt/mk should help alot if this is the case. (It would be the equivalent to MX4.)

The next problem after that is how (in)efficient the heatsink actually is at discharging the heat with the current design.

Then finally finding an air temperature that allows the heatsink to work at a reasonable pace. (not too cold because it costs $$$ to keep it at that temp)

etc

[rograz]

I was thinking people should replace the thermal compound used in the Avalon unit. Considering there are some people who found out some units have no thermal compound or had it improperly applied.

Assuming that the thermal compound is crap and/or low efficiency, then transferring the heat from the chips to the heatsink would probably help some at higher clocks. A good number of thermal pads with 6 or 7watt/mk should help alot if this is the case. (It would be the equivalent to MX4.)

Yes thermal pads would have been a better solution on such a large and uneven surface I'm quite sure. I'm still curious however to see how much individual heat sinks on the chips would make since then you don't have remove the original heat sink. Might take a while to apply them all though 

[johnyj]

I was thinking people should replace the thermal compound used in the Avalon unit. Considering there are some people who found out some units have no thermal compound or had it improperly applied.

Assuming that the thermal compound is crap and/or low efficiency, then transferring the heat from the chips to the heatsink would probably help some at higher clocks. A good number of thermal pads with 6 or 7watt/mk should help alot if this is the case. (It would be the equivalent to MX4.)

Yes thermal pads would have been a better solution on such a large and uneven surface I'm quite sure. I'm still curious however to see how much individual heat sinks on the chips would make since then you don't have remove the original heat sink. Might take a while to apply them all though 

https://en.bitcoin.it/wiki/Avalon#Others

"About install extra heat-sinks on each avalon chip: please do not do that. there is a air gap between the die and package top, install a heatsink on chip is useless. and will cause overheating. because the top PCB copper act as a heatsink too. do not cover them. " --NGZhang

[johnyj]

Is there enough spacing between the modules to put gpu ram sinks on each individual chip? I have about 2 kg of them, yes kilograms  

Infact just checked how much the bag of them weigh, it was close to 4 kilograms so I should have plenty for 1x3 module avalon once I get it.

I was thinking people should replace the thermal compound used in the Avalon unit. Considering there are some people who found out some units have no thermal compound or had it improperly applied.

Assuming that the thermal compound is crap and/or low efficiency, then transferring the heat from the chips to the heatsink would probably help some at higher clocks. A good number of thermal pads with 6 or 7watt/mk should help alot if this is the case. (It would be the equivalent to MX4.)

The next problem after that is how (in)efficient the heatsink actually is at discharging the heat with the current design.

Then finally finding an air temperature that allows the heatsink to work at a reasonable pace. (not too cold because it costs $$$ to keep it at that temp)

etc

The exposed metal area under the chips are shallower than the surface of the PCB, if you put a large piece of heatpad then maybe there will still be air gap between the metal part and heat pad, and you can't guarantee a good pressure on all the chips' contact area at the same time, so I think thermal compound is the solution: You put one big drop of the thermal compound on each metal area, and tighten the screws, the thermal compound will spread out depends on the gap distance between chip metal area and the heatsink, so that every chip will get a good contact with heatsink



The result of this handling is reduced temp of 3-4 degree under same fan speed, but since fan speed will always adjust according to temp3, the real effect is just lowered fan speed and maybe 1 degree lower temp. Anyway, you can be sure each chip get maximum cooling now. Then you can also add thermal compound between heatsink and bottom of the case, and cool the bottom of case with an external table fan, this can drop another 3 degree of temp

Another important part is putting screws around VRM area, there is a very hot component at this area, MOSFETS maybe, but not like chips, this component only contact the surface ground layer of PCB, so adding screws will transfer the heat from the surface ground layer directly into heatsink


Since Avalon heatsink are so large, maybe it works even better by taking out those modules and use a large table fan to directly blow on them, or mount many 120mm fans on heatsink like someone did on ASIC miner blades, as long as you reach a certain CFM, the cooling effect should be the same. I usually like the bigger fan solution since it will provide much more airflow and much less noise