Hacking The KNC Firmware: Overclocking - page 10.

Elenelen

full member

Activity: 203

Merit: 100

Quote from: greghawk on June 29, 2014, 09:40:00 AM

Quote from: Elenelen on June 29, 2014, 08:16:42 AM

Quote from: AFox on June 29, 2014, 06:20:16 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

Yes. I pealed the yellow sticker from the VRM, and put the heatsink on top. See pictures in next posts....

is it even necessary to peal the yellow stickers off?

Don't know...I just did

greghawk

newbie

Activity: 37

Merit: 0

Quote from: crashoveride54902 on June 29, 2014, 10:28:31 AM

Quote from: greghawk on June 29, 2014, 09:40:00 AM

Quote from: Elenelen on June 29, 2014, 08:16:42 AM

Quote from: AFox on June 29, 2014, 06:20:16 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

Yes. I pealed the yellow sticker from the VRM, and put the heatsink on top. See pictures in next posts....

is it even necessary to peal the yellow stickers off?

on neppy probably...on nov...not so much as they don't heat up as bad as neppys do

yes I meant neppy. But than I assume, warranty is gone...hm

crashoveride54902

hero member

Activity: 784

Merit: 504

Dream become broken often

Quote from: greghawk on June 29, 2014, 09:40:00 AM

Quote from: Elenelen on June 29, 2014, 08:16:42 AM

Quote from: AFox on June 29, 2014, 06:20:16 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

Yes. I pealed the yellow sticker from the VRM, and put the heatsink on top. See pictures in next posts....

is it even necessary to peal the yellow stickers off?

on neppy probably...on nov...not so much as they don't heat up as bad as neppys do

greghawk

newbie

Activity: 37

Merit: 0

Quote from: Elenelen on June 29, 2014, 08:16:42 AM

Quote from: AFox on June 29, 2014, 06:20:16 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

Yes. I pealed the yellow sticker from the VRM, and put the heatsink on top. See pictures in next posts....

is it even necessary to peal the yellow stickers off?

butjust41day

full member

Activity: 131

Merit: 100

Quote from: Elenelen on June 29, 2014, 09:25:39 AM

Quote from: butjust41day on June 29, 2014, 09:16:48 AM

Considering some of your VRMs were over 100 degrees (based on pictures you posted earlier). I think you could safely say you achieved as high as 30 degree reduction on some VRMs. I think the VRMs and the fact that they are they are the BMR4641002 (max 40A) instead of the BMR4641008 (max 50A) as you mentioned earlier are the most disappointing aspect of the current design (KNC that's not a criticism; but a critique). I do understand that KNCminer was under a lot of pressure to ship before the end of the month. I wonder if they were expecting far greater power savings and that is why the board does not have the 1008s.

I think delivery dates written in stone are a two edged sword. They give targets to be achieved; but, can result in shipping before every important detail is fully dealt with. Hopefully we'll see revised boards coming.

I agree: KnC should have used more robust VRM's.

And indeed I had 107 degrees with some of the slot-1 VRM's due to a gap between pad and VRM... see this picture:

The edges of the aluminium indicate they use a mechanical shear to cut the plate. At least if memory serves me right it's been over 30 years since I used a shear. It's brute force and tends to twist the aluminium plate.

Elenelen

full member

Activity: 203

Merit: 100

Quote from: butjust41day on June 29, 2014, 09:16:48 AM

Considering some of your VRMs were over 100 degrees (based on pictures you posted earlier). I think you could safely say you achieved as high as 30 degree reduction on some VRMs. I think the VRMs and the fact that they are they are the BMR4641002 (max 40A) instead of the BMR4641008 (max 50A) as you mentioned earlier are the most disappointing aspect of the current design (KNC that's not a criticism; but a critique). I do understand that KNCminer was under a lot of pressure to ship before the end of the month. I wonder if they were expecting far greater power savings and that is why the board does not have the 1008s.

I think delivery dates written in stone are a two edged sword. They give targets to be achieved; but, can result in shipping before every important detail is fully dealt with. Hopefully we'll see revised boards coming.

I agree: KnC should have used more robust VRM's.

And indeed I had 107 degrees with some of the slot-1 VRM's due to a gap between pad and VRM... see this picture:

butjust41day

full member

Activity: 131

Merit: 100

Quote from: Elenelen on June 29, 2014, 08:27:56 AM

And this is the result: 20 degrees lower on the VRM's :

Considering some of your VRMs were over 100 degrees (based on pictures you posted earlier). I think you could safely say you achieved as high as 30 degree reduction on some VRMs. I think the VRMs and the fact that they are they are the BMR4641002 (max 40A) instead of the BMR4641008 (max 50A) as you mentioned earlier are the most disappointing aspect of the current design (KNC that's not a criticism; but a critique). I do understand that KNCminer was under a lot of pressure to ship before the end of the month. I wonder if they were expecting far greater power savings and that is why the board does not have the 1008s.

I think delivery dates written in stone are a two edged sword. They give targets to be achieved; but, can result in shipping before every important detail is fully dealt with. Hopefully we'll see revised boards coming.

crashoveride54902

hero member

Activity: 784

Merit: 504

Dream become broken often

nice work elenelen

looks good

Elenelen

full member

Activity: 203

Merit: 100

And this is the result: 20 degrees lower on the VRM's :

Elenelen

full member

Activity: 203

Merit: 100

Here some more:

Elenelen

full member

Activity: 203

Merit: 100

OK, here is my DIY cooling-project:

more pictures will follow

Elenelen

full member

Activity: 203

Merit: 100

Quote from: AFox on June 29, 2014, 06:20:16 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

Yes. I pealed the yellow sticker from the VRM, and put the heatsink on top. See pictures in next posts....

AFox

hero member

Activity: 539

Merit: 517

Quote from: tzortz on June 21, 2014, 12:18:20 PM

Quote from: crashoveride54902 on June 21, 2014, 11:58:06 AM

Is it possible to peal of the yellow sticker from the VRMs ?
Is it safe the put aluminum or cooper heatsinks directly on the VRMs with only thermal paste (link) and not thermal adhesive pads (link) ? As VRMs convert electricity and heatsinks are electrical conductors.

rograz

hero member

Activity: 575

Merit: 500

Quote from: fubly on June 25, 2014, 01:16:11 AM

Hashfast squeez out 500 Gh/s on this design, with 28nm. For this must it be possible to squeez out 1,5 Th/s per 20nm chip.

KNC have to redesign the board and the controller.

There are other metrics than process node that determines final performance, as a matter of fact process node mostly impacts efficiency and density and the only real effect on "performance" it has is allowing higher operational frequency. A transistor does not perform better just cause it's smaller than it's predecessor, yes you can pack more of them in the same area but performance/transistor at the same frequency stays the same from one node to the other.

Elenelen

full member

Activity: 203

Merit: 100

I have removed the Neptune aluminum plate. Now I can see the Ericsson VRM's..... this is bad: they are the BMR4641002 (max 40A) instead of the BMR4641008 (max 50A). Sad

...... no OC possible !

The 1002 has a OverCurrentProtection of 48A and looking to my Advanced-page: nearly all the VRMs are already around 46-47A (and this is KnC's default setting).
They should have used the 1008 instead, which has 50A and an OCP of 62A.

fubly

hero member

Activity: 561

Merit: 521

Trustless IceColdWallet

Powering up the chip! Thats KNC's problem.

Hashfast squeez out 500 Gh/s on this design, with 28nm. For this must it be possible to squeez out 1,5 Th/s per 20nm chip.

KNC have to redesign the board and the controller.

Ski72

member

Activity: 84

Merit: 10

Kinex - The New Frontier

Quote from: crashoveride54902 on June 24, 2014, 05:20:11 AM

Quote from: Elenelen on June 24, 2014, 12:35:29 AM

No, sorry Ski72....

You need something like this:

wonder if that is the same size as a cpu cooler mounting bracket?

Yes. I import it (the pic) into AutoCAD and the board is the same size as the Jup boards (I compared them side by side, dimension wise, in AutoCAD), and the holes are in the same places, just a different layout of the VRMs.

Ski72

member

Activity: 84

Merit: 10

Kinex - The New Frontier

Quote from: Elenelen on June 24, 2014, 12:35:29 AM

No, sorry Ski72....

You need something like this:

That's how they are mounting it now, yes, with the airflow in that direction.
The way it is/was mounted on the 28nm board, is with the radiator turned 90^o clockwise, and the front/face of the radiator in the direction of the PCIE and ten pin connectors.

Basically you'd take the board out of the box, mount the radiator as above, and put a fan on the radiator. You're still getting the same airflow across the radiator for cooling purposes, and take away all the foam, more airflow around the VRMs. Also, looking at the other pics from crash and tzortz, you could then put little heat sinks on the VRMs.

Just a few thoughts ...

crashoveride54902

hero member

Activity: 784

Merit: 504

Dream become broken often

Quote from: Elenelen on June 24, 2014, 12:35:29 AM

No, sorry Ski72....

You need something like this:

wonder if that is the same size as a cpu cooler mounting bracket?

Elenelen

full member

Activity: 203

Merit: 100

No, sorry Ski72....

You need something like this:

Topic: Hacking The KNC Firmware: Overclocking - page 10. (Read 144378 times)