MP 1,1-5,1 Cooler Northbridge on The Mac Pro 5,1 Dual CPU

[h9826790]

Just a quick update with Fans speeds achieved, which is "nearly normal" as per "auto" settings

View attachment 919371

Some recommendation about your fan setting.

NEVER base on the heatsink temperature, but make it base on the diode temperature. e.g. Exhaust base on CPU A diode, but not CPU A heatsink temperature.

Because, if the cooling system failed (e.g. the NB rivet broken), your heatsink will be very cool (because no heatsink transfer onto it anymore), but the diode will be cooking itself.

With MFC, your manual setting will completely override the system protection. Which means, even your NB / CPU A diode is overheating, but as long as the heatsink is cold. The fan won't spin up.

 

[tommy chen]

.

 

[TzunamiOSX]

@h9826790 Yes that is indeed a great idea.

I've also been thinking of a way to get more air over the NB Heatsink. I was just brute forcing the problem for a few days with a Macs Fan Control driven brute force attempt at pushing more air through the A CPU ( at rear of the Mac ). I'm just generally unhappy with this approach.

Why?
1. Using the CPU A Boost fan method from my previous post was getting on my nerves as the small fan inside the Heatsink was running at mid speed give a great mosquito pitched whine.
2. moving back to "auto" settings and therefore without the whine the Northbridge temperatures were just too high in mid 60ºC for my comfort and despite ignoring my Northbridge for 5 years I'd like to not have the processor board fail from inaction.

So to Action!
Get McGuyver prototype, what ever was handy and could be interated before dinner was due on the table ( my turn to cook ).

Prototype
Here's my 15 minute prototype in A4 80 GSM A4 and Cello tape. which I left installed and running under test while cooking pasta bolognese from a frozen base sauce with fresh everything else.

ta da' THE Mac Pro Air Baffle ( origami edition ) is born.

View attachment 919346View attachment 919347View attachment 919348
MacsFan settings


Back story: I've spent a bit of research time from my olden days at IBM. we introduced a Racked server "back in the day" and it had an innovative Air baffle as part of a memory and CPU up-scalable feature that allowed some similar scaling as the cMP ( Classic Mac Pro series ). There was one part in particular that I was enamored with. 'The Air Baffle', IT was both a gift and a impediment to servicing and adding RAM but if left out it was well known by field engineers to be a required part to avoid auto shutdown conditions.

Therefore I posit that while Apple's very capable engineers configured the processor board for a laminar flow they did not win the marketing / design battle to add a air flow guide ( baffle) to direct air over the North Bridge Heatsink Fins.
Bear in mind that IBM x346 had a similar power density but with 16 redundant fans and no effective sound output limits. X346 26.44 litres with 2 x 650 Watt Power Supplies giving 51 Watts per litre versus Mac Pro 5,1 with 43 litres and 980 Watts 22.8 Watts per litre or less than half the density. No criticism intended but there are therefore weaker spots in the design, aka the NB thermal problem!
What did IBM do ?

View attachment 919344

Testing
Initial testing Using iCore System Load running full load on 2 x 6 Cores of x5680 with 8 x 4 GB RAM slots 1-8 Full.
It seems the Inlet fan can provide sufficient total pressure to keep the velocity high enough to actually reduce the Northbridge temp check the attached 45 minute Full load test graph captured using Hardware Monitor.
View attachment 919349
Next Steps
Given the Hypothesis seems supported by the experiment I will now try fabricating a fit test model on my hobby 3d PLA Printer, It wil take some more time to get a heat resistant version.

Time Frame
Possibly 3 weeks for an testable engineering prototype.

If your'e interested just pop a reply.

Same idea, different constructions

With my construction i bring down the temp of the Tdiode from 82°C to 79°C, without any other fan speedups.

This is not enough for me, so i work on a wind tunnel like this (not to scale) with a 40x40mm fan. I think, I will use PVC to built the tunnel, because it looks exactly like the other plastic parts inside the MacPro.

 

[Sean__Wong]

Same idea, different constructions
View attachment 919627

With my construction i bring down the temp of the Tdiode from 82°C to 79°C, without any other fan speedups.

This is not enough for me, so i work on a wind tunnel like this (not to scale) with a 40x40mm fan. I think, I will use PVC to built the tunnel, because it looks exactly like the other plastic parts inside the MacPro.
View attachment 919628

I look forward to seeing your new shroud and Fan, don't forget your RAM needs much cooler air than the Northbridge.

 

[TzunamiOSX]

I look forward to seeing your new shroud and Fan, don't forget your RAM needs much cooler air than the Northbridge.

The lower part of the tunnel gives a free way to bring air to the RAM

 

[TzunamiOSX]

My first test constuction without the 40x40 mm fan. Northbridge temp is at 78°C (normal at 82°C) RAM also looks good. Next test is insert a fan inside the tunnel.

 

[flyproductions]

My first test constuction without the 40x40 mm fan. Northbridge temp is at 78°C (normal at 82°C) RAM also looks good. Next test is insert a fan inside the tunnel.

You know that you are taking away about half of the cooling of the already hotter rear CPU A that way?

 

[h9826790]

You know that you are taking away about half of the cooling of the already hotter rear CPU A that way?

Why?

 

[flyproductions]

Why?

The construction is covering three quarters of the front facing part of CPU A heatsink not already covered by CPU B's heatsink, blocking most of the direct airflow from the intake fan.

 

[h9826790]

The construction is covering three quarters of the front facing part of CPU A heatsink not already covered by CPU B's heatsink, blocking most of the direct airflow from the intake fan.

No it isn't. It just re-direct the air flow. All air will still flow through Heatsink A

 

[TzunamiOSX]

The construction is covering three quarters of the front facing part of CPU A heatsink not already covered by CPU B's heatsink, blocking most of the direct airflow from the intake fan.

No difference at the moment, but this is the test phase. CPU A is 10 degrees hotter than CPU B (same as before).

 

[flyproductions]

No it isn't. It just re-direct the air flow. All air will still flow through Heatsink A

Yes, but just trough the lowest part, flowing along no more than three or four of about 20 fins of the heatsink.

Anybody can do with their machine as they like. But i don't think that's a good idea to sacrify the cooling of a CPU to get better cooling for more heat resistant parts like the northbridge or not very heat generating parts like the ram modules.

 

[h9826790]

Yes, but just trough the lowest part, flowing along no more than three or four of about 20 fins of the heatsink.

Anybody can do with their machine as they like. But i don't think that's a good idea to sacrify the cooling of a CPU to get better cooling for more heat resistant parts like the northbridge or not very heat generating parts like the ram modules.

Doesn't matter, still roughly the same amount of air flow through a heatsink that has high thermal conductivity.

In fact, the re-directed air may be cooler than before, because those air now won't touch the side wall of heatsink B.

Anyway, as per TzunamiOSX reported, there is no difference, which fits the expectation.

 

[TzunamiOSX]

Every CPU has his own Fan and the Exhaust Fan also garant the airflow to the fins. Without an Exhaust Fan the airflow where significant reduced, but it shouldn't really matter with it.
.

 

[flyproductions]

Doesn't matter, still roughly the same amount of air flow through a heatsink that has high thermal conductivity.

Of course the amount of surface this airflow gets in touch with matters! This is how heatsinks with heatpipes and fins work. Otherwise this thing could have 3 instead of 20 fins.

In this situation it would be even better, this thing had vertical fins. So the air, coming out of the tunnel at least had the chance to expand upwards. But as it is, the upper ~15 fins are almost completely locked out of the airflow coming from the front. So my expectations would have been "a bit" different. But TzunamiOSX may be right estimating the exhaust fan together with the booster fan inside the sink still generating enough airflow in the upper part to prevent disaster.

No offense, and nice work done. But i sometimes do not understand the amount of attention this northbridge gets, which is a part that, like the pcie bridge of later AGP-graphicscards, generates but can also tolerate a lot of heat on cost of the cooling of the already hotter running CPU.

 

[h9826790]

Of course the amount of surface this airflow gets in touch with matters! This is how heatsinks with heatpipes and fins work. Otherwise this thing could have 3 instead of 20 fins.

In this situation it would be even better, this thing had vertical fins. So the air, coming out of the tunnel at least had the chance to expand upwards. But as it is, the upper ~15 fins are almost completely locked out of the airflow coming from the front. So my expectations would have been "a bit" different. But TzunamiOSX may be right estimating the exhaust fan together with the booster fan inside the sink still generating enough airflow in the upper part to prevent disaster.

No offense, and nice work done. But i sometimes do not understand the amount of attention this northbridge gets, which is a part that, like the pcie bridge of later AGP-graphicscards, generates but can also tolerate a lot of heat on cost of the cooling of the already hotter running CPU.

You can't see my point.

Heatsink size - same
Airflow - same
Heatsink surface area - same

Your example, from 20 to 3 fin - heatsink size changed - heatsink surface area changed

It just re-direct the air flow through different part of the heatsink in this case. Shouldn't make CPU A run noticeably warmer. And it's proved already.
[automerge]1591871466[/automerge]
There are lots of things that can affect cooling.

e.g. Air speed (even same air mass), the higher speed, the bigger the wind chill effect (of course, we only consider subsonic airflow here).

Or the air flow though which part of the heatsink.

e.g. Flow through the bottom of the heatsink is the best (which is closest to the CPU, and should be hottest part of the heatsink, has biggest thermal gradient to the airflow).

If we also put this into consideration. Now, the airflow are accelerated to flow though the bottom part of the heatsink. It may actually make CPU A cooling better.

Anyway, from TzunamiOSX reported, there is no change. No point to debate anymore, we already know the result.

 

[flyproductions]

Heatsink size - same
Airflow - same
Heatsink surface area - same
You example, from 20 to 3 fin - heatsink size changed - heatsink surface area changed
...
Or the air flow though which part of the heatsink.

Sure my thoughts are somewhat "theoretical" So you are perfectly right: the upper fins will not even with the best working heatpipes ever get as hot as the lower ones. So effect of blocking the upper part will not be a linear calculation.

But i am also not so sure that "we already know the result" as i think as higher the temp of CPU A gets from load pushing the intake fan to higher rpm and for as longer periods this scenario persist as more there would be a noticable increase of CPU A temp, compared to a free airflow. So i would be much interested in a "tunnel in front" <-> free airflow comparision of i.e. 2 houres of full CPU load with all fans set to "auto" and no additional fans in the system.

 

[TzunamiOSX]

Here are results.
One hour Prime95 with and without tunnel. I have marked the improtant temperatures. Red: higher, Green: same or lower. All temps are in normal range.

 

[Sean__Wong]

@flyproductions some wise engineers have trodden this path before us. IBM's X346 xeon based machines had an AIR baffle to funnel increased velocity air over the ram and then the CPU in a tunnel configuration, I posted about them as they also had the X56xx series chipset

Their Lineage included the x3650 that sported the Xeon X5680 chips we have in the cMP5,1
The new air baffle was in a configuration that bifurcated air for the RAM and CPU due to the temperature differential.
Look for Items 27a and 27b on the diagramme to see how they did this in only 2 Rack Unit height!

System service parts - IBM System x3650 M3 (4255, 7376, 7945)

System service part numbers for the IBM System x3650 M3.

www.ibm.com

[automerge]1591880259[/automerge]

Here are results.
One hour Prime95 with and without tunnel. I have marked the improtant temperatures. Red: higher, Green: same or lower. All temps are in normal range.

View attachment 923237

I think some extra Intake FAN POWER is required in your design.
Menawhile
After a week My system has been happily doing its quiet business and my stress level is lower as the NB Is below 60 and more over lower under load.

CPU A Heat Sink

CPU B Heat Sink

NorthBridge Heat Sink

This post #172 has the full background including a ghetto prototype.

 

[flyproductions]

Here are results.
One hour Prime95 with and without tunnel. I have marked the improtant temperatures. Red: higher, Green: same or lower. All temps are in normal range.

Thanks for so quickly performing the test!

So a short conclusion could be something like that:
Good news: It has no significant negative effect to CPU A temperature
Bad news: It has no significant positive effect to the northbridge temperature


No offense again. But 3 celsius is what you get of an inaudible speedbump of about 100rpm to the intake which even has a positive effect to both of the CPUs. Strange, by the way, that one of the ram modules gets even warmer with the tunnel installed. And sure results will be different with a fan mounted to the tunnel. But this will be one thing in the first place: One more fan.
[automerge]1591912672[/automerge]

@flyproductions some wise engineers have trodden this path before us. IBM's X346 xeon based machines had an AIR baffle to funnel increased velocity air over the ram and then the CPU in a tunnel configuration, I posted about them as they also had the X56xx series chipset

Yes, but i think rack mount cases are "little" of a different story. Sure, if there is no height available at all, your only choice to get proper cooling is very fast, concentrated airflow. But that's what these things sound like.

By the way: I gave 1.200rpm (double of what i usually have) a short test-run too. It got the northbridge down to 65 (and CPUs to 34/38) within minutes without anything additional installed.

 

[TzunamiOSX]

Thanks for so quickly performing the test!

So a short conclusion could be something like that:
Good news: It has no significant negative effect to CPU A temperature
Bad news: It has no significant positive effect to the northbridge temperature

The fan is missing in the test. second step is to insert a 40x40 Noctua fan inside the tunnel.

By the way: I gave 1.200rpm (double of what i usually have) a short test-run too. It got the northbridge down to 65 (and CPUs to 34/38) within minutes without anything additional installed.

I know that, but this will reduce the lifespan of the fans. All parts of the Mac Pro are special parts and were harder to find then a 3$ PVC tunnel and an standard fan

 

[MIKX]

Northbridge Tdiode & general cooling

De-Dusting & Northbridge temp reduction in Dual CPU cMP 4,1 & 5,1

( NOTE : I feel that both Northbridge temp reduction AND de-dusting/de-clogging the cheesegrater DUAL CPU cMPs deserves a separate thread to embrace the concepts. Northbridge cooling requires that all that hot air be efficiently extracted from the CPU case...

forums.macrumors.com

 

[TzunamiOSX]

I have found an old 40x40 mm fan and a long 7V adapter cable, so i have a chance to test the effect of the system while waiting for the Noctua fan.

I am aware that the exit of the tunnel is too big and also the angle on the upper plate is too strong. I will modifiy this by change the upper plate.

Here are the preliminary results

 

[fatespawn]

I think a better way of looking at your Northbridge temps are what they are at idle since most of the time, that's how your computer is sitting (I assume you aren't running Prime 95 all day long!).

With no exotic baffling (nor am I running any fan control software) my Northbridge die sits in the low 70's - all day long. But when I run a Handbrake encode obviously my CPU package spikes and the fans ramp up. When the fans ramp up, the Northbridge Die drops almost 15 deg C.

The Northbridge under load is much cooler than it is at idle. So, all the baffling in the world won't really change your long term temperature exposure if that's your concern. If you think your idle Northbridge is too hot, you need more mass airflow. Software is the only solution there.

 

[TzunamiOSX]

View attachment 923796View attachment 923795

I think a better way of looking at your Northbridge temps are what they are at idle since most of the time, that's how your computer is sitting (I assume you aren't running Prime 95 all day long!).

With no exotic baffling (nor am I running any fan control software) my Northbridge die sits in the low 70's - all day long. But when I run a Handbrake encode obviously my CPU package spikes and the fans ramp up. When the fans ramp up, the Northbridge Die drops almost 15 deg C.

The Northbridge under load is much cooler than it is at idle. So, all the baffling in the world won't really change your long term temperature exposure if that's your concern. If you think your idle Northbridge is too hot, you need more mass airflow. Software is the only solution there.

My last post was a test in idle. In idle i get the highest temperatures, because the intake and exhaust fans are working with minimal speed.