Discussion Zen 5 Builders thread

[Det0x]

My 8400 a-die profile have reached its final form in the meantime 💪
Cpu cant run higher memoryspeed with a-die GDM disable, and memory cant run tigher timings, so i'm pretty much "done" on this combo.

• Delidded 9950X ES running PBO
• ROG CROSSHAIR X870E APEX
• 2x16GB G.Skill Trident Z5 Royal Neo A-DIE running 8400MT/s CL34 @ daily 1.68 VDD / 1.55 VDDQ voltages
• Very large custom cooling with mora

Decided to complete all stability tests in a single boot to show that i dont cherry pick settings for spesific benchmarks 🤝

Manged to average 373mb/s over a 10 hour karhu run


Following completed:

• 10 hours karhu @ 373 mb/s
• ~350% runmemtestpro @ 36.5k mb/s
• 1 hour y-cruncher all tests enabled
• 1 hour OCCT large AVX512
• 1 hour Prime95 large FFT's AVX512
• Quick Testmem5 1usumus cfg to end it off

 

[Det0x]

Have also had a few days to play with my 2x24GB m-die

Shouldn't come as a surprise for anyone i hope, but a-die fully maxed out is alittle faster than maxed m-die, but infact the difference was alittle less than what i originally expected.

End result was around ~373mb/s for A-die and ~364mb/s for M-die in karhu, while latency was also a tiny bit worse for M-die. (same cpu that caps out at 8400MT/s used for both kits) The main performance-difference comes down to the inability to run GDM disable with m-die at high memory-speeds

If the small performance increase is worth 50% less memory is up to you to decide i guess 🤔

Onwards to some screenshots (y)

• Delidded 9950X ES running PBO
• ROG CROSSHAIR X870E APEX
• 2x24GB Trident Z 8400 M-DIE running 8400MT/s CL36-49-40-80 @ daily 1.56 VDD / 1.50 VDDQ voltages (core tuning level2)
• Very large custom cooling with mora

Managed 364mb/s average over 6 hours in karhu


Afterwards i also ran some ~9 hours runmemtestpro while i went to work


Also updated Y-cruncher 10B for hwbot before i went back to daily sticks

 

[Hulk]

So I took Igor's advice and tried the XTM70 thermal paste. Here's how the test went.
Set TDP at 200W. Should have set it higher but I'm not doing it again!
MX-6 was applied as pea size in the middle of the spreader. When I took it apart the paste was pretty much perfect.
For the XTM70 I followed the Corsair instructions making the grid on the AIO plate.
10 minutes of CB R23.
Scores were the same within statistical variation due to same 200W limit. 40,010 for MX-6, 40,165 for XTM70.

I'm curious as to how people here would interpret these results?

 

[Markfw]

So I took Igor's advice and tried the XTM70 thermal paste. Here's how the test went.
Set TDP at 200W. Should have set it higher but I'm not doing it again!
MX-6 was applied as pea size in the middle of the spreader. When I took it apart the paste was pretty much perfect.
For the XTM70 I followed the Corsair instructions making the grid on the AIO plate.
10 minutes of CB R23.
Scores were the same within statistical variation due to same 200W limit. 40,010 for MX-6, 40,165 for XTM70.

I'm curious as to how people here would interpret these results?

I would say that in general, thermal paste is overrated. What I mean by that is any paste from a recognized authority (Corsair, thermalright, etc...) are all the same. I am sure there are exceptions.

 

[DaaQ]

So I took Igor's advice and tried the XTM70 thermal paste. Here's how the test went.
Set TDP at 200W. Should have set it higher but I'm not doing it again!
MX-6 was applied as pea size in the middle of the spreader. When I took it apart the paste was pretty much perfect.
For the XTM70 I followed the Corsair instructions making the grid on the AIO plate.
10 minutes of CB R23.
Scores were the same within statistical variation due to same 200W limit. 40,010 for MX-6, 40,165 for XTM70.

I'm curious as to how people here would interpret these results?

I would say Kryosheet, unless you want liquid metal. I am running a 7900xtx sapphire, and never hit the temps others report. Highest I've seen is 80C mem junction.

I am going to use it on my 9800x3d.

EDIT: Large custom loop.

 

[IEC]

Once you get to "good enough" I haven't noticed a substantial difference from TIM. I happened to get a very good deal on a bunch of TG Hydronaut, so my default high performance paste is Hydronaut but I have MX4, MX5, Noctua, Scythe, and Thermalright TIMs for more temporary setups/testing and have not noticed any significant difference. Maybe maximally a 10MHz boost clock difference? Which is usually within margin of error... Though I have noticed a few tend to pull chips out of sockets if I don't heat up the system before removing a heatsink, LOL

I recently brought my Radeon VII back online for the Folding@Home race this month and put it under a Bitspower GPU block I imported from Japan - using Hydronaut as the TIM for the GPU/HBM area and included Bitspower 1mm thermal pads for VRMs and surface mount components the beast stays pretty tamed at hotspot temps under 75C max.

 

[igor_kavinski]

Absolutely hammering the 9950X right now with a 90 minute Vegas Pro encode using Voukoder to x265, slow setting.

I think try that again with the new paste?

 

[Hulk]

I think try that again with the new paste?

You were right Igor. Compared to the MX-6 at 175W, with the XTM70 I'm able to run 185W at around the same or lower temps. Seems like the higher the thermal load the more of a difference between the paste brands. CPU(TctlTdie) average MX-6 175W is 85.3C vs XTM70 at 200W at 85.0C. Same temp but dissipating an additional 10W. I'm set to 200W TDP as opposed to 175W with the MX-6 but the render is only pulling 182W average, where before it was pulling 172W average. Now I'm fully utilizing the stock 200W cap on the CPU. Average clock 110MHz faster.

This CPU game is one of inches sometimes.

I'll take it. Thanks for the recommendation!

 

[Seba]

Though I have noticed a few tend to pull chips out of sockets if I don't heat up the system before removing a heatsink, LOL

That's a more important factor (at least with PGA/AM4 CPUs).

 

[igor_kavinski]

Compared to the MX-6 at 175W, with the XTM70 I'm able to run 185W at around the same or lower temps.

How did you apply the paste?

 

[igor_kavinski]

Best thermal paste database and charts - Paste versus paste comparison, reviews and durability for CPU and GPU | igor´sLAB

We have incorporated the extensive feedback from the beta phase and are now finally putting the first final version of the database online. Operation has been greatly simplified once again and clarity…

www.igorslab.de

Corsair XTM70 vs Dow Corning DOWSIL TC-5888 | igor´sLAB

www.igorslab.de

That Dowsil one is even better (may net you additional 100 MHz?) but seems to only be available from Aliexpress: https://www.aliexpress.com/item/1005006200653459.html

Price seems very reasonable for 50g of that stuff.

TG-04 and TFX may be better but they need to be heated up before application. Complete no-no, at least for me.

 

[Hulk]

How did you apply the paste?

I used the template as per the instructions It was easy.

 

[mmaenpaa]

Best thermal paste database and charts - Paste versus paste comparison, reviews and durability for CPU and GPU | igor´sLAB

We have incorporated the extensive feedback from the beta phase and are now finally putting the first final version of the database online. Operation has been greatly simplified once again and clarity…

www.igorslab.de

Corsair XTM70 vs Dow Corning DOWSIL TC-5888 | igor´sLAB

www.igorslab.de

That Dowsil one is even better (may net you additional 100 MHz?) but seems to only be available from Aliexpress: https://www.aliexpress.com/item/1005006200653459.html

Price seems very reasonable for 50g of that stuff.

TG-04 and TFX may be better but they need to be heated up before application. Complete no-no, at least for me.

As always there are risks involved when buying online 😁

Fake factory vs. wheel of fortune: DOWSIL TC-5550 and TC-5888 from the China store and only one of them is genuine | igor´sLAB

Sourcing really good thermal pastes is a real challenge at the moment. The DOWSIL TC-5888 and TC-5550 are prime examples of the frustration that modern supply chains or manufacturers’ unwillingness to…

www.igorslab.de

 

[Hulk]

I would say that in general, thermal paste is overrated. What I mean by that is any paste from a recognized authority (Corsair, thermalright, etc...) are all the same. I am sure there are exceptions.

I agree. It's definitely "fooling around at the edges" kind of stuff.
It's hard to even tell if a degree or two is statistical variation during testing or due to application differences.
But I would rate the XTM70 as definitely as good and "probably" a little more effective than the MX-6 if I were comparing the two.
Yeah, I know, not a strong recommendation, especially for the price!

 

[alcoholbob]

I agree. It's definitely "fooling around at the edges" kind of stuff.
It's hard to even tell if a degree or two is statistical variation during testing or due to application differences.
But I would rate the XTM70 as definitely as good and "probably" a little more effective than the MX-6 if I were comparing the two.
Yeah, I know, not a strong recommendation, especially for the price!

Have you tried using PTM7950 based thermal pads? Apparently they get better with age (through multiple heating/cooling cycles) and also don't ever have to be replaced unlike traditional thermal paste.

 

[Hail The Brain Slug]

Have you tried using PTM7950 based thermal pads? Apparently they get better with age (through multiple heating/cooling cycles) and also don't ever have to be replaced unlike traditional thermal paste.

View attachment 115248

PCM based TIM can't be meaningfully used on an IHS application. While the core/chip might achieve high temps, the skin temp of the IHS rarely ever reaches temperatures that allow it to fully melt and flow. I and a few others have tried it with awful results. I even unplugged fans to heatsoak and force it to melt (several times actually, I did 10+ heat cycles this way) but performance was not great. As soon as I gave up and repasted with normal paste, temps were better.

It's really only useful in direct die applications. If it doesn't melt in routine use, it doesn't perform well.

 

[alcoholbob]

PCM based TIM can't be meaningfully used on an IHS application. While the core/chip might achieve high temps, the skin temp of the IHS rarely ever reaches temperatures that allow it to fully melt and flow. I and a few others have tried it with awful results. I even unplugged fans to heatsoak and force it to melt (several times actually, I did 10+ heat cycles this way) but performance was not great. As soon as I gave up and repasted with normal paste, temps were better.

It's really only useful in direct die applications. If it doesn't melt in routine use, it doesn't perform well.

Laptop makers seem to disagree...

 

[Hail The Brain Slug]

Laptop makers seem to disagree...

Laptop CPUs don't have an IHS. That's a direct die application.

 

[Seba]

If you want to use a graphical representation to visually show differences between some values, then you must start from zero on the relevant axis.

Otherwise, the visual representation is very extremely misleading.

Not to mention that comparing temperature values is problematic.

 

[Thunder 57]

If you want to use a graphical representation to visually show differences between some values, then you must start from zero on the relevant axis.

Otherwise, the visual representation is very extremely misleading.

Not to mention that comparing temperature values is problematic.

Yup. It's going to be twice as cold tomorrow. What does that mean? Unless using Kelvin, nothing.

 

[Markfw]

Have you tried using PTM7950 based thermal pads? Apparently they get better with age (through multiple heating/cooling cycles) and also don't ever have to be replaced unlike traditional thermal paste.

View attachment 115248

and why are you using a 14900k to represent anything in a Zen 5 builders thread ? apples and oranges.

 

[Hulk]

If you want to use a graphical representation to visually show differences between some values, then you must start from zero on the relevant axis.

Otherwise, the visual representation is very extremely misleading.

Not to mention that comparing temperature values is problematic.

Good point "0" is about 6 miles to the left on that chart.

 

[Hulk]

Anybody here doing direct-to-die cooling on their Zen 5? Was it worth it?

 

[Det0x]

Anybody here doing direct-to-die cooling on their Zen 5? Was it worth it?

It helps if you got the correct setup to use the extra headroom

Temp-spread while running Cinebench R23 MT PBO daily tune
(GPU showing idle temp)

 

[Hulk]

It helps if you got the correct setup to use the extra headroom

Temp-spread while running Cinebench R23 MT PBO daily tune
(GPU showing idle temp)

View attachment 115305

Keeping those die temps<80C while dissipating 300W is impressive.
If you have the time can you let me know what your cooling setup is and how hard it was to delid?