3900x intresting findings for TIM

[bignaz]

So 3900x is a great cpu. But its hot. Even with a custom loop it gets hot. And with PBO just pushing a single cores power draw to the moon it gets hot even when using limited cores.

 

Few things to note.

 

7nm makes for a very focused heat spot.

Offset chiplets help spread the heat out but only so much you can do.

Ryzen has a bunch of temperature sensors and reports the highest one. 

Most heatsinks and blocks are not ideal for this type of heat load.

 

So i tested out an IC Graphite pad. And i saw a 4c drop against Noctua NT-H1. It worked pretty good. The heat dissipation in the x and y helped put more of that heat across the block. Why i was disappointed in it on my i9 it worked pretty well on my 3900x and i would have no problem using this.

 

IC is not really telling you the full specs. The reason why most reviews on a tradition cpu show it to be low to mid when compared to tim's. Real world its only 13 W/mK.

 

 

Now i work in the semiconductor industry. One of the perks is i have access to a bunch of materials and contacts with most of the larder manufacturers who make them.

 

 

So i now want better. So i made a call to Laird and had them send out some samples of Tgon 9000. So i got some 17um, 25um, 40um, 70um and 100um.  So you dont want too thin and you do not want too thick.  So i tested out the 70um today and over Noctua i seen a 10c drop.  Not IC is saying they do 800 W/mK in the x and y. The Laird is at 1900 W/mK. And this was a ASTM test we don't know how IC got there numbers.

 

 

I'll test the other samples and see what thickness offers the best performance on the 3900x.  For the thermals i was using a Corsair Hydro X loop and ran Prime for 2 hours to heat soak the loop. Water temp was also monitored. Ambient was in an controlled room at 70f.

 

 

I still wanna do more testing before i tell everyone to run out and buy them. I saw some respectable results from the IC pad and the temp drop i seen vs paste made sense since its pulling heat away from the hot spots and dumping it into more of the microfins.

 

 

 

So if you have a Ryzen 3000 series and one of these pads let me know if you seen any improvement vs paste. Im curious to see how it works on the single chiplet  Ryzans.  Like i said. On my i9 i really was not impressed with the IC pad. It was meh. But on my 3900x it worked pretty well.

 

 

[porina]

I was speculating elsewhere if tower coolers might not be the most effective if a chiplet is only under 1 or 2 of the side heatpipes, whereas previously it was more centralised and may cover more of them. Does a typical multiple heatpipe design spread between heatpipes much too? I was wondering if this might show up in IR imaging by one side of a tower being warmer than another.

 

If the above assumption is true, then materials spreading the heat better could help a lot. I'd be interested in seeing more results.

[bignaz]

We have an thermal camera at work I just need to figure out when I can get some time in it. 

 

Heatpipe coolers should be fine. Your making contact with the ihs. What I'm interested in is taking the heat and spreading it out more. 

 

 

It's still one test on one waterblock. I need to do more testing. I'm happy with the ic pad. And very happy with the meterial I got from Laird.  I don't have any air coolers. I do have a H115i pro I can test out. 

 

 

Now my concern with a cooler with direct touch heatpipes would be spacing between the base and pipes. It's been years since I used air and last direct touch heatsink I had has some tiny gaps. This your going to take a thermal hit from that. I seen some with gap filling propertys but my test were done on a flat waterblock. And man is that XC7 block flat. So this is a kinda best case scenario. Full ihs contact. Full block contact. The pads really show the benefit of spreading that heat out. 

 

 

It will take a bit but I'm going to post all my data using NT-H1 as a thermal paste baseline see how the thickness as well as other brands preform. 

 

I have a limit supply of carbon nano tube thermal paste. I emailed them to see if I could get more. But we used it to evaluate it so I have maybe enough left to do 3 cpu's. 

 

[For Science!]

6 hours ago, bignaz said:

-

 

I think I am slightly dubious at least from a theoretical stand point and would appreciate if you could explain to me a few things:

 

- Heat generated from the dies/chiplet (as in any CPU) is transferred to the IHS via the TIM underneath it (solder, in the case of Ryzen). Within the IHS, a sollid block of nickel coated copper (like a waterblock), there is pure metal-to-metal contact in X, Y, and Z which has very high thermal conductivity. Why does then applying a layer of graphene sheet over this improve the thermal spread within the IHS.

 

- As mentioned before, metal-to-metal contact has superior thermal to conductivity over any contact interfaced by TIM. My understanding is that a good thermal paste performs usually higher than a pad is because with pastes you maintain the metal-to-metal contact and the paste acts only "where it needs to" (i.e. in a gap) whereas with a pad you cover everything and therefore lose metal-to-metal contact in exchange for "reusability". 

[porina]

13 hours ago, bignaz said:

What I'm interested in is taking the heat and spreading it out more. 

That's what I'm interested in too. I see I can get one via Amazon for tomorrow, so could try this...  not too expensive (although not cheap compared to good paste) I'm willing to give it a go too. And ordered.

7 hours ago, For Science! said:

- Heat generated from the dies/chiplet (as in any CPU) is transferred to the IHS via the TIM underneath it (solder, in the case of Ryzen). Within the IHS, a sollid block of nickel coated copper (like a waterblock), there is pure metal-to-metal contact in X, Y, and Z which has very high thermal conductivity. Why does then applying a layer of graphene sheet over this improve the thermal spread within the IHS.

I think the way to look at it is that the pads have a greater conductivity within the plane, and act as like a 2nd HS and give the cooler more of a heat spread to work with. If this can be repeatably proven is another matter.

 

When it arrives, I'll try:

1, Noctua D9L with MX-4 (if I can find some spare NH-T1 I might use that instead)

2, Noctua D9L with IC graphite pad

 

This will be on a 3600 and I'll aim to do it in a short time (while still leaving it to stabilise) to reduce environmental variation influences. Past experience with the D9L on that CPU showed I can reach steady state in minutes with Prime95 6x 128k FFT as my preferred load.

 

I also have an IR camera so could try that, although as the heatsink surface is reflective that has given interesting results in the past. I might have to put some tape on it for a better IR reading surface for example.

[bignaz]

Just ordered some Indium foil.  I'll be testing that. 

 

New tubing should be in tomorrow to replace my temparty home Depot tubing I had to use.

[AngryBeaver]

15 hours ago, bignaz said:

Snip

 

 

I'm not sure this makes sense. So when it comes to thermal dynamics. Metal is the best conductor of heat. The problem with metal to metal is the air which can be between them when bare. Now air is a terrible conductor of heat... which is why TIM was born.

 

Now TIM doesn't conduct heat as good as metal, but does so much better than air. So it is there to fill those small gaps and imperfections so that heat can transfer effectively. 

 

So a pad or TIM isn't there to do anything but pass the heat from IHS to Block. The PAD shouldn't do this any more effectively than TIM outside of the rating for what it can move. 

 

So now the question is if it is spreading the heat out better (not likely) or if it is just the higher heat transfer rating. I would be curious to see a similar test with liquid metal. I actually will probably be looking for a nickel plated block for that very purpose in the future.

 

The results you are seeing could be your IHS and block are both very good surfaces. More so than most.

[bignaz]

I hope I can have some thermal image's uploaded by Saturday.  I'm using the $10k thermal camera at my work but because I'm taking them at work security needs to sign off on the pictures.  So until I get the email back with the permission I can't transfer the images "Joys of being a defense contracter".

 

 

Indium foil I have a bunch coming in. I forgot about this until today when I was working on one of our lasers and I needed some for that. So I got a bunch of different thickness.  They said it should be here by Wednesday.

 

 

That said using the thermal camera. The NT-H1 did have more pronounced hot spot on it. The pad it was more spread out.  Temps were no change but the camera did show more heat spread across the block but that's not really a surprise. Plan is when I can get some time with the camera some more to do a full video from boot with the loop at rt to the loop fully heatsoaked the. Run it at 5-10x and do the same for the NT-H1. I'm down to a little under 1/2 a tube so I need to stay planning stuff out more as I want to use the same stuff. I won't use any from another batch as I do not know how consistent there product is from batch to batch and I'm going for data.  And again that video will have to be approved by security. 

 

 

 

My expectations are not that it will be this end all to thermal paste on a 3900x. But see how the meterials thermal propertys translate to this cpu/block combo. I do not expect this to be the best. But from what I seen so far I'm expecting it to rank in the upper teir of thermal paste. What is nice considering it preformed pretty poor on the Intel and 2700x when I was first given it. 

 

 

[porina]

Just done the testing... results are:

 

Wraith Prism: 91.8C

IC Graphite: 87.5C

Arctic MX-4: 85.8C

 

CPU: Ryzen 3600 stock, with 2666 ram

Cooler: Noctua D9L

Stress: Prime95 6x 128k in-place FFT running 10 minutes.

 

So, no improvement from the IC graphite here. Tests were done after each other, I don't think ambient would have significantly changed but can't rule out some small change. I would comment that the D9L isn't a direct heatpipe contact design, but has its own heat spreader. Maybe that negates some of the possible benefit from using such a pad.

 

I tried using thermal imaging, but didn't do anything about possible reflections on the metallic surfaces so results inconclusive.

[For Science!]

On 8/7/2019 at 6:15 PM, bignaz said:

They said it should be here by Wednesday.

Any follow-up on this?