Log Cravo CG7 Rebirth: X3D, A2000, water cooling and case cramming

[mishmash]

I'm finding it's time to upgrade my aging i7-4700/1650 combo mini HTPC that I use as a steam box in my living room in big picture mode. Some lag is appearing in some indie/older games I have started playing, and I'd also like to see what the VR fuss is about.

As usual, I have the horrible limitation of a 290mm depth in my rental apartment home theatre cabinet. This brings me back to an old favourite of mine: Lazer's Cravo CG7. This was originally my home server box, and was decommissioned for a slightly larger in volume 2U myElectronics case.

Anyway, lets start with some big parts that arrived on Friday (pic below). The build will be based around a 5800X3D and A2000 12GB. The CPU is overpowered for the GPU, but I have seen some excellent gains (especially 1% FPS) across the board for any GPU compared to its other 8/16 thread siblings. It also appears to be an excellent CPU for AI driven games like Civ6, Stellaris, etc.

More to come later!

 

[Factor Jay]

Looking the goods! Keep us updated

 

[mikejmcfarlane]

That is a really pretty and elegant case :-)

 

[mishmash]

I always knew this build would be tight, on paper it works, but seeing your plans laid out in real life can always become daunting!

Starting to test fit some components for layout. Here we have a GTX M160 HardwareLabs radiator, the GPU block that will be used for the A2000, and a VRM cooler that I will be using to cool the HDPlex GAN PSU that I have on the way.

Not seen in this pic - but my original plan was to use 25mm thick fans, unfortunately I was out by 2mm and will have to step down to 80x15mm fans to keep things comfortable. The radiator was originally planned to be further back against the case wall, but with the way the inlet and outlet radiator chambers are organised I will need to set the fans in a push direction to blow inside the case for optimum heat transfer. The final air cooling solution will then be 2x 80x15mm fans for the rad, and a single 120mm or 140mm for the top panel. The airflow inside the case should take care of the motherboard VRM and M2. The GPU will block most of the PSU, hence my desire to add additional cooling there.

Build will be soft tubed with Tygon A-60-G. Still to add in are quick disconnects, ModUltra LOBO CPU block, a filter, and an aquacomputer High Flow NEXT.

Quick pic:

 

[mishmash]

Today I added the shunt mod to my A2000 and also changed the cooler to the bartx waterblock (that I got unused with the card from the seller).

I've added 47 mOhm of shunt to the cooler side of the card. This should add another 7.4W of power (up from 70W normally) through the PCIe slot. The removed fan will also allow an extra half a watt to go to the GPU rather than the fan

Not my best looking solder job, but the joint appears to be holding and is not a cold solder. I also tested to see if the card still ran before unmounting the cooler:


Unmounting the cooler showed that the Nvidia factory was a bit overzealous with thermal paste.


Here is the prepared card with new thermal pads and some Noctua paste I had lying around:


Final assembled card:


I'll have to wait for the CPU block to arrive before continuing (which will be mid June).

 

[Snerual]

Awesome! I had no idea there were waterblocks for the A2000. And I absolutely loooooove the design of that case.

I know there are some 80x20mm fans out there but I don't know if they would be any good for a radiator application. Or 80x10 fans in a push-pull config?

Out of curiosity... how exactly do you arrive at that calculation for the shunt mod? I am also considering it and I think 75-80W max total power is exactly where I want to land.

 

[mishmash]

Awesome! I had no idea there were waterblocks for the A2000. And I absolutely loooooove the design of that case.

I know there are some 80x20mm fans out there but I don't know if they would be any good for a radiator application. Or 80x10 fans in a push-pull config?

Out of curiosity... how exactly do you arrive at that calculation for the shunt mod? I am also considering it and I think 75-80W max total power is exactly where I want to land.

I think the block I got from the seller is used for mining applications - the nice thing is that it covers the memory and a portion of the VRM area.

I'll look into the push-pull setup, that's a good idea. It will be millimetre tight I think.

Here is my output from the shunt mod calculator:


And the app is here:
https://github.com/bmgjet/ShutMod-Calculator

I have a few spare 47 mOhm resistors left, I can send you a couple for free (just pay postage). I'm located in France.
Edit: this is the part on RS-components

 

[Snerual]

Thanks a lot for the offer but I have some 50 mOhm lying around already.

I understand the basics of calculating the new value, but my assue is that the RTX A2000 actually has 2 shunts (there is a 5 mOhm on the bottom of the PCB as well) and I can't find any info online about how that one relates to the top shunt, other than people saying "you need to also mod that one to get max performance".

 

[mishmash]

Thanks a lot for the offer but I have some 50 mOhm lying around already.

I understand the basics of calculating the new value, but my assue is that the RTX A2000 actually has 2 shunts (there is a 5 mOhm on the bottom of the PCB as well) and I can't find any info online about how that one relates to the top shunt, other than people saying "you need to also mod that one to get max performance".

I was also having trouble with this.

I recall finding some videos from China indicating that modding both removes the power limit completely (actually it increases it to around 160W) and as a result they were also adding a +12V wire soldered to the rear resistor (with ground through PCIe slot).

The picture in this link suggests only modding the cooler side resistor.

The comments here also suggest only touching the front unless you will be adding wires to the rear resistor.

I can't find the link anymore about the 160W thing - perhaps it was in a video.

 

[Snerual]

Yeah I dunno... the people in those comments also sound like they are just winging it and doing whatever to make their hash rate go up a bit.

My assumption is that both shunts play "some" role and the people that solder 005 on both sides are simply halving the overall resistance thus doubling potential power draw (to 140W I guess no idea how someone would land at 160.)

By touching only one shunt, the effect of the mod is reduced. However it is impossible to really know by how much?

We could assume both shunts play an equal role, and then the effect of modding only 1 would be half of what you actually expect.

So in your case that would mean you didn't increase power draw by 11% but rather by only 5.5%. But again that is assuming the ratio/importance of both shunts is equal and I have NO idea about that...

 

[mishmash]

Yeah I dunno... the people in those comments also sound like they are just winging it and doing whatever to make their hash rate go up a bit.

My assumption is that both shunts play "some" role and the people that solder 005 on both sides are simply halving the overall resistance thus doubling potential power draw (to 140W I guess no idea how someone would land at 160.)

By touching only one shunt, the effect of the mod is reduced. However it is impossible to really know by how much?

We could assume both shunts play an equal role, and then the effect of modding only 1 would be half of what you actually expect.

So in your case that would mean you didn't increase power draw by 11% but rather by only 5.5%. But again that is assuming the ratio/importance of both shunts is equal and I have NO idea about that...

That's true - there is no solid understanding of how each resistor is operating. I'm ok with 5.5% also if your assumption is correct.

You could use some elmor labs kit to measure pcie slot power before/after modding one/both sides to see the power levels - this way you can work out how to set the correct resistance as well to optimise the power draw.

 

[hereforthefeast]

Today I added the shunt mod to my A2000 and also changed the cooler to the bartx waterblock (that I got unused with the card from the seller).

Ah, I really wished I had replied faster to that hws post but glad to see it went to a good home! Can't wait to see the finished build

 

[mishmash]

Couple of small updates:

I had to countersink the screws that hold the feet onto the bottom panel, I found that the existing screws were protruding out into the interior of the case, and would interfere with some of the radiator/PSU mounting:



My modultra arrived! I mistakenly ordered a 3 pin 4.2 DDC, but this isn't a big issue, as I will be running it at constant speed anyway. I re-terminated the connector and also added some sleeving so it is all nice and clean when mounted on the motherboard:




The HDPlex PSU cooling was tricky, the generic VRM cooler I ordered was too tall. Luckily I found a discontinued alphacool RAM cooler which sits very low profile and will do the trick with cooling the PSU:



This is the case layout so far, I've made the final order for fittings. The general idea is to run a closed loop: Rad -> CPU -> GPU -> PSU. The radiator is suprisingly thick, I'm not worried about the surface area though, more worried about the fans getting enough air flow through it. We'll see!

 

[rfarmer]

Couple of small updates:

I had to countersink the screws that hold the feet onto the bottom panel, I found that the existing screws were protruding out into the interior of the case, and would interfere with some of the radiator/PSU mounting:



My modultra arrived! I mistakenly ordered a 3 pin 4.2 DDC, but this isn't a big issue, as I will be running it at constant speed anyway. I re-terminated the connector and also added some sleeving so it is all nice and clean when mounted on the motherboard:




The HDPlex PSU cooling was tricky, the generic VRM cooler I ordered was too tall. Luckily I found a discontinued alphacool RAM cooler which sits very low profile and will do the trick with cooling the PSU:



This is the case layout so far, I've made the final order for fittings. The general idea is to run a closed loop: Rad -> CPU -> GPU -> PSU. The radiator is suprisingly thick, I'm not worried about the surface area though, more worried about the fans getting enough air flow through it. We'll see!

You can also use voltage control on the Gigabyte boards so even with a 3 pin connector you can still adjust it if you need to.

 

[mishmash]

You can also use voltage control on the Gigabyte boards so even with a 3 pin connector you can still adjust it if you need to.

Yes, that's definitely the idea. I was thinking 60% or so should be enough.

 

[Elaman]

I have this same radiator and a non-3D Ryzen 5800X, but I chickened-out and got a 140mm radiator which I ended up using.
Obviously I am very curious about what your temperatures will look like.
I have always liked this case, especially with the wooden panels.

 

[BaK]

The HDPlex PSU cooling was tricky, the generic VRM cooler I ordered was too tall. Luckily I found a discontinued alphacool RAM cooler which sits very low profile and will do the trick with cooling the PSU:

Liquid cooling the PSU, oh yeah!
Were you able to use the 3 screw holes I see on the VRM block to attach it to the PSU ?

 

[mishmash]

Liquid cooling the PSU, oh yeah!
Were you able to use the 3 screw holes I see on the VRM block to attach it to the PSU ?

It will be attached with a thermal pad in the middle of the contact area, and then adhesive thermal tape around the outside + middle to keep it all together. The screw holes unfortunately did not fit up perfectly...

 

[mishmash]

I had a few more fittings come in today, which let me finalise the layout. It was not as per my original vision, I wanted a longer uninterrupted run to the flow meter, but, we are on the SFF forum, so the gods of SFF did not allow it...

The PSU is directly connected to the GPU.

My main concerns is that with SFF and 16/10 tubing there is a lot of o-ring strain everywhere. I suppose we will find out at leak test time what happens, especially with some of these rotary fittings. Anyway, here are some pics of the test fitment, I still need to pull it all apart and rinse everything. Avoiding kinks was a challenge!





Next step is rinsing, making cables, and the leak test.

 

[morj]

Looks amazing. Why do you need a flow meter though? For science?