Ryzen Decision to Make

[mackerel]

I saw a claim on another forum that the graphite pads worked for a user better than regular thermal compound. These are claimed to have a high thermal conductivity in its plane, more so than vertically, so the spreading action might help here. Maybe I'll let more try it before I jump on the bandwagon myself.

 

[pgdeaner]

Problem is with Ryzen architecture, not really with coolers. I was talking with Noctua about it in the last days. The difference between some smaller and top air cooling is not so big. It's because of too much heat condensed on a small surface. The same issue is with Intel chips like 9900K. In the same time there is a BS specification saying that Ryzens are 65-105W TDP or that 9900K is 95W TDP. In real 150W TDP coolers can't keep them below 90C.
As an effect of Ryzen architecture, max temps on 6 cores are about the same as on 8, and 12 cores are also not so far away. Also 8/12 cores can work at a lower voltage for the same clock as 6 what clearly shows core binning.

The 9900K suffers from the thickness of the silicon. Der8auer has demonstrated some gains by de-lidding and lapping but to your point, the chip is struggling to shed heat.

 

[bignazpwns]

I saw a claim on another forum that the graphite pads worked for a user better than regular thermal compound. These are claimed to have a high thermal conductivity in its plane, more so than vertically, so the spreading action might help here. Maybe I'll let more try it before I jump on the bandwagon myself.

Might of been me.


But testing an IC graphite pad i seen a 4c drop over Noctua NT-H1 paste. But this is based on 1 3900x and one block. I dont know how it translates to a say 3700x or a 3900x with a heatsink. But in my case i seen a pretty good temp drop. I have some higher end stuff as well as indium foil that should be here on Wed. The other pads i seen a bigger decrease in temps. IC's numbers for there pads are not very accurate to what it is in real life. The stuff i got from one of our vendors is around 15 w/mk in the z and 1900 in the x/y. I had a few diff versions of it i wanna test. I know for my waterblock they work great.


But i need to run more test. I can repeat the results. But im going to try another block and see if the temp drop is the same. It could be just a block/pad combo that works. Im using a new block the corsair XC7. But i did turn the system on with no water in the loop just the block. Using a thermal camera at work i could see what side the cores were on pretty clear. Then with the pad it spread that heat out more across the block.


But i do have a very small amount of carbon nano tube thermal paste what kills liquid metal. This was left over from some grad students who were working on this for a project they did the testing in our lab. The application was for high thermal performance for laser modulus on satellites. But i think indium foil would be a good choice. We use it on our lasers. ARC is sending me out a bunch along with an engineering survival kit because another lab got sent some but ours did not so i cried to them lol. But indium foil is really soft and will work pretty good.


Here is the bad part. Its expensive. Like really expensive. 4 inch X 4 inch around $60. So around $15 per 40X40. We get bigger sheets at work. And this wont be reusable. Its way to soft. But i'll test it out test out some diff thickness's and post my findings and enough people want a 40x40mm square i'll see about picking up some direct from them and try to get the same price they give my work.


This is how i been delidding and re soldering Ryzens and intels.

 

[bassnut]

Might of been me.


But testing an IC graphite pad i seen a 4c drop over Noctua NT-H1 paste. But this is based on 1 3900x and one block. I dont know how it translates to a say 3700x or a 3900x with a heatsink. But in my case i seen a pretty good temp drop. I have some higher end stuff as well as indium foil that should be here on Wed. The other pads i seen a bigger decrease in temps. IC's numbers for there pads are not very accurate to what it is in real life. The stuff i got from one of our vendors is around 15 w/mk in the z and 1900 in the x/y. I had a few diff versions of it i wanna test. I know for my waterblock they work great.


But i need to run more test. I can repeat the results. But im going to try another block and see if the temp drop is the same. It could be just a block/pad combo that works. Im using a new block the corsair XC7. But i did turn the system on with no water in the loop just the block. Using a thermal camera at work i could see what side the cores were on pretty clear. Then with the pad it spread that heat out more across the block.


But i do have a very small amount of carbon nano tube thermal paste what kills liquid metal. This was left over from some grad students who were working on this for a project they did the testing in our lab. The application was for high thermal performance for laser modulus on satellites. But i think indium foil would be a good choice. We use it on our lasers. ARC is sending me out a bunch along with an engineering survival kit because another lab got sent some but ours did not so i cried to them lol. But indium foil is really soft and will work pretty good.


Here is the bad part. Its expensive. Like really expensive. 4 inch X 4 inch around $60. So around $15 per 40X40. We get bigger sheets at work. And this wont be reusable. Its way to soft. But i'll test it out test out some diff thickness's and post my findings and enough people want a 40x40mm square i'll see about picking up some direct from them and try to get the same price they give my work.


This is how i been delidding and re soldering Ryzens and intels.

Something like this …… https://www.dhgate.com/product/arsy...568626601986#cpmfvh-2-5|null:null:r1474235553

If so for $20 a sheet (30 X 30cm) I'll take a gamble and order one.

 

[bignazpwns]

Something like this …… https://www.dhgate.com/product/arsy...568626601986#cpmfvh-2-5|null:null:r1474235553

If so for $20 a sheet (30 X 30cm) I'll take a gamble and order one.

I can get way higher quality stuff for pretty much the same price but 18” x 24” sheets. Hell a 18" x 24" sheet of Tgon 820 is only $11. But i would go with a Tgon 9070

https://assets.lairdtech.com/home/brandworld/files/THR-DS-TGON_9000_032415.pdf

I got all the Tgon 9XXX im checking out playing with the different thickness. It takes a little bit to get all the data. But thinner could be better or it could be worse. I'll be finding out that and if there is a sweet spot for that thickness. Then i'll switch out and test some indium foil.


Some are just graphite powder on a sheet. You want the in plane mono crystal structure. Some of the cheaper stuff they just take the common powdered graphite put it on some pet film and sell it like that. We want graphite only.


Larid makes good stuff. There a billion dollar company and make some great stuff.


So i have to balance this between work, stuff i need to do for home and the home life. So this week i have a few experiments i need to be watching for most of the day and the free time i have tomorrow i gotta fire up the 3d printer and print out a 90 fitting to drain my loop at home...i should just use the SLS and make a stainless steel one lol. But resin will work. So my free time tomorrow will be spent modeling the part up and then cleaning it because resin is just so fun to work with. But i should have some time Saturday and be able to get some time on the thermal camera.



Fun bit of information. We have a 1kW laser at work. We have bigger ones but in the lab i work in we only have a 1kW. But we got a new took to take some measurements of the beam. It puts out 674 micrograms of force. So for some laser propulsion 5,489,614,243.32w laser that. We could then hit the speed of light and keep it going by bouncing the beam back and forth and just seeding the beam to keep it going. This whole conversation started about having a mirror that was big enough and millions of light years away so we could look at it with a telescope and see dinosaurs.

 

[mackerel]

Might of been me.

It was you, didn't make the connection in username until you said this. You may note a similarity in avatar pics there... I ordered an IC graphite pad and that should arrive today. Currently I have a Prism on the 3600 which is rather hopeless for temps and noise. I was going to switch back to the D9L anyway, but can test it with the pad and also my regular MX-4 most likely.

 

[Zerileous]

There have been a couple of notable youtube videos testing out the graphite pads, however not with Ryzens. IIRC their results seemed to be fairly marginal, however that doesn't account for the theory that Ryzen's]/s] Zen2 CPUs have a much smaller surface area generating heat and the lateral conductivity of the graphite. Here they are for reference

If you used indium foil wouldn't that basically solder the cold plate to the IHS? Also would this not permanently damage both metals?

 

[bignazpwns]

There have been a couple of notable youtube videos testing out the graphite pads, however not with Ryzens. IIRC their results seemed to be fairly marginal, however that doesn't account for the theory that Ryzen's]/s] Zen2 CPUs have a much smaller surface area generating heat and the lateral conductivity of the graphite. Here they are for reference

If you used indium foil wouldn't that basically solder the cold plate to the IHS? Also would this not permanently damage both metals?

Nah. Indium has a melting point of 156.6c.
This is why it's used on cpu's. Low melting point but higher then the typical temps that a CPU will be exposed to.

This is why I started looking into it for Ryzen. It was never tested on the new Ryzens. It was quickly written off. And I even did it on my 9900k and my 2700x. It just did. Not work very well. It was acceptable performance. Now when I tested it on the 3900x I found it to preform pretty well. It went from slightly below what I was getting with the current paste I was using to a bit better. . So that's why I started to look into other meterials for this.

Here's Indium when exposed to temps.

 

[pgdeaner]

There have been a couple of notable youtube videos testing out the graphite pads, however not with Ryzens. IIRC their results seemed to be fairly marginal, however that doesn't account for the theory that Ryzen's]/s] Zen2 CPUs have a much smaller surface area generating heat and the lateral conductivity of the graphite. Here they are for reference

If you used indium foil wouldn't that basically solder the cold plate to the IHS? Also would this not permanently damage both metals?

Are you maybe thinking of gallium? It bonds rather nicely with other metals like copper. It's used in liquid metal products.

 

[mackerel]

But testing an IC graphite pad i seen a 4c drop over Noctua NT-H1 paste. But this is based on 1 3900x and one block. I dont know how it translates to a say 3700x or a 3900x with a heatsink. But in my case i seen a pretty good temp drop. I have some higher end stuff as well as indium foil that should be here on Wed. The other pads i seen a bigger decrease in temps. IC's numbers for there pads are not very accurate to what it is in real life. The stuff i got from one of our vendors is around 15 w/mk in the z and 1900 in the x/y. I had a few diff versions of it i wanna test. I know for my waterblock they work great.

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.

 

[bignazpwns]

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.

Kinda figured the results might only be good for the 3900x and 3950x.

But also ordered a new tube of NT-H1 just in case the tube is junk and i'll retest.

 

[Woomack]

My Noctua D9L with 1 fan in a closed ITX case has about 6-8°C worse results than with two fans. Noctua itself said that the difference in their results in the lab was lower and it doesn't matter much if in use is one or two fans. I guess it's because of limited airflow in the ITX case but still worth to mention.
You may not see big gains from graphite just because contact with the IHS is not the issue here but how dies are heating up on a small surface. As an effect, most small and large air CPU coolers will have +/- 10°C. I already said that but I tested it on 4 coolers so far and Noctua confirmed it with their internal tests. It's like you go with good water cooling or just accept 90°C or more under load.
Btw. my Ryzen was not losing stability up to 115°C. Usually, stability issues or lower OC because of overheating are starting just above 95°C.

 

[mackerel]

My suspected stability loss running stock was happening around 90C, and my criteria for stability might be tougher since my normal workload is comparable to Prime95. I'm doing actual work on PrimeGrid, where there is a challenge currently running. By benchmarking, I found the best throughput was with 2 workers (2 tasks) using 3 cores each. I don't think it is possible to specify this in P95 stress test, as it only works one core per worker. FFT size was 1280k but I hear reports some are now 1440k. These fit in each CCX's local cache, without crossing cache boundary and resulting significant performance loss if you do. Temps are slightly lower than my 6x 128k FFT test used here, presumably due to inefficiencies in the multi-threading code. Tasks results are all double checked, so it is easy to see if there has been any instability. I have 2 known bad units, two inconclusive units which I expect to be confirmed bad in due course, 15 known good units, and 3 unchecked so far. Each task takes over 6 hours. Very roughly, I had a 1 in 4 error rate occurring within a 6 hour window per CCX. Running Prime95 stress test at a single FFT range for 24 hours would not necessarily detect this error rate. When Skylake launched I had detectable error rates of around 1 a month until I figured out it was fussy with ram. Tests back then were not multi-threaded and longer ones could take a day or more, which is a lot of compute time wasted from errors.

Anyway, I left the graphite pad in as it was the last one I fitted, and have resumed crunching. Let's see if the new units are without error from the somewhat reduced temperatures compared to the Prism.

I would like to see if temperature related instability at stock on Zen 2 is a thing on more than just this observation e.g. with more systems, but it isn't something I can set up easily. I never observed instability on stock Intel CPUs due to CPU temperature. When power efficiency considered overclocked (modest clock increase, lower voltage setting) I found stability drops around 80C or so, so that remains my target maximum for 24/7 OC running on Intel to this day. I can tolerate higher stock temps.

 

[Zerileous]

Thanks guys, yeah I was thinking of gallium.

For the indium, I'm guessing it's not plastic enough (although it does seem to deform fairly easily under those temperatures).

 

[Woomack]

I would like to see if temperature related instability at stock on Zen 2 is a thing on more than just this observation e.g. with more systems, but it isn't something I can set up easily. I never observed instability on stock Intel CPUs due to CPU temperature. When power efficiency considered overclocked (modest clock increase, lower voltage setting) I found stability drops around 80C or so, so that remains my target maximum for 24/7 OC running on Intel to this day. I can tolerate higher stock temps.

I haven't seen any stability issues on 3600 and 3700X at stock settings when CPUs were under full load, AVX etc. CPU is designed to work at 95°C and at least at auto, it will throttle to keep max 95°C. So performance will drop, not stability (at least shouldn't).

One thing regarding stability. Open hwinfo64 or windows logs and run anything 3D related, 3DMark or even open Afterburner. I don't know if you can confirm that but so far everyone that I asked and has X570, said they see pcie hardware errors. I also couldn't pass 3DMark stability test on X570 and the best result was 96% while typical was more like 92-94%. Nothing is crashing but these errors are weird.

Btw 280 AIO/2x 1000RPM fans, in tests on 3700X... max temp so far 83°C, 88°C was on Scythe Fuma2, 94°C on Noctua D9L w/2 fans (this one inside the ITX case), Cryorig C7 Cu and C1 couldn't handle this CPU, the same as stock AMD cooler, all 3 were throttling at 95°C+. All coolers tested with Noctua H1 TIM.

 

[Bustos]

Looks like I’ll be investing in an aftermarket air cooler. I was hoping to get by with stock on PBO.

 

[EarthDog]

3DMark stability test

What is this, actually? Are you talking that new PCIe bandwidth test?

 

[Woomack]

What is this, actually? Are you talking that new PCIe bandwidth test?

Nope. It's called "stress test". Have to pass 97% to be stable. I get randomly 88-94% on X570. On 2 mobos with different graphics cards is the same. On Z390 I get ~98%.

Looks like I’ll be investing in an aftermarket air cooler. I was hoping to get by with stock on PBO.

If I find time then I will test Noctua U12A next week. Right now it's on 9700K. However, I count it will be in the mid '80s. Manual overclocking helps to drop temps a bit and overclock in the same time. With Fuma 2 cooler I was able to set 4.3GHz 1.37V on 3700X and it was slightly below 95°C. That's max temp so for gaming etc. it would be more like ~75°C max.

 

[mackerel]

I haven't seen any stability issues on 3600 and 3700X at stock settings when CPUs were under full load, AVX etc. CPU is designed to work at 95°C and at least at auto, it will throttle to keep max 95°C. So performance will drop, not stability (at least shouldn't).

Shouldn't doesn't mean doesn't. How close to the limit are AMD boosting these CPUs? Still, I only have a sample of one, and there's more testing that could be done to confirm before I really shout about it.

One thing regarding stability. Open hwinfo64 or windows logs and run anything 3D related, 3DMark or even open Afterburner. I don't know if you can confirm that but so far everyone that I asked and has X570, said they see pcie hardware errors. I also couldn't pass 3DMark stability test on X570 and the best result was 96% while typical was more like 92-94%. Nothing is crashing but these errors are weird.

Reminds me of the follow link, although that specifically refers to SSDs. The -ABB bios was released for my Asrock B450 earlier this week and I just installed it but not really explored it yet.
https://www.techpowerup.com/257817/...resses-several-issues-affecting-3rd-gen-ryzen

Nope. It's called "stress test". Have to pass 97% to be stable. I get randomly 88-94% on X570. On 2 mobos with different graphics cards is the same. On Z390 I get ~98%.

From memory, it loops the 3DMark test you select something like 10 times. Wonder how it determines stability short of an outright crash?

 

[EarthDog]

From memory, it loops the 3DMark test you select something like 10 times.

Ick... if it does that, I don't have time to do it now.. I will run it later after I get some other testing done........


....... on quite possibly the most ridiculous little board ever in the X570 Phantom ITX board............this little thing uses Intel 115x mounts instead of AMx!!!!! WTH?!!!!!