trying the near impossible... most powerful GOD BOX in a miniITX case

[MrGenius]

Major fail on the using pads instead of the tape provided with those heat spreaders. That tape is probably the best possible solution. And there's not a snowball's chance in hell those pads perform better(they're mediocre pads to begin with, Fujipoly are WAY better). I've got those same heat spreaders on about a dozen sticks of RAM. Some of which have been removed and reused. Your guess about it ripping ICs off is a bad one. Not even close. For reuse I've replaced the provided tape with the 0.11mm BCP thermal tape(stuff on the blue rolls) you can get all over the place. It's a tiny bit thicker than the stock tape. But it's also a bit wider, and can be cut to length to fit the spreaders better. Resulting in more coverage on the ICs in some instances as well(particularly DDR 1).

I should also mention the fitment issue with those though. Using any kind of tape with dual-sided DIMMs doesn't fit correctly. You need to cut the tabs off the tops to get both spreaders to mount closer to each other. Or you won't be able to get full contact on the RAM ICs. But you'd have figured that out if you'd tried using the tape.

 

[lexluthermiester]

Major fail on the using pads instead of the tape provided with those heat spreaders. That tape is probably the best possible solution. And there's not a snowball's chance in hell those pads perform better(they're mediocre pads to begin with, Fujipoly are WAY better). I've got those same heat spreaders on about a dozen sticks of RAM. Some of which have been removed and reused. Your guess about it ripping ICs off is a bad one. Not even close. For reuse I've replaced the provided tape with the 0.11mm BCP thermal tape(stuff on the blue rolls) you can get all over the place. It's a tiny bit thicker than the stock tape. But it's also a bit wider, and can be cut to length to fit the spreaders better. Resulting in more coverage on the ICs in some instances as well(particularly DDR 1).

I should also mention the fitment issue with those though. Using any kind of tape with dual-sided DIMMs doesn't fit correctly. You need to cut the tabs off the tops to get both spreaders to mount closer to each other. Or you won't be able to get full contact on the RAM ICs. But you'd have figured that out if you'd tried using the tape.

That seems a bit harsh. I have to disagree with you. The spreaders @venturi is using do very well for RAM modules that tend to get hot.

 

[venturi]

Mr genius,
The pad 8material has higher thermal conductance than the tape.
But I also want to be able to remove the heatspreaders later without damaging the dimms, the tape is way too dangerous for these expensive dimms.
I’ve seen tape rip chips off the pcb.
Lastly, I appreciate your opinion, but I’ve tested my way up to this process, it was not arrived by accident,

The clay also goes down between crevices making full contact and limiting the thermos effect.

Again, thank you for your opinion.

I was able to drop the temps 13C, so I am content

 

[akee99]

very nice Pc omg

 

[venturi]

I made some optimizations. Here is my Cinebench R20 score. Not bad for a small cute little build.

20143

 

[Mussels]

This isnt the Godbox anymore, needs 2x64 core threadrippers

 

[lexluthermiester]

I made some optimizations. Here is my Cinebench R20 score. Not bad for a small cute little build.

20143

View attachment 128795

Good Grief mate! Very nice!

 

[venturi]

This isnt the Godbox anymore, needs 2x64 core threadrippers

If it was on core count alone, then even that 2x64 you refer to would be outdone by any rack box with 4 or 8 cpu sockets

The intent was the most workstation with Audio, 4 way sli or 4 gpu, raid, be very quiet, and run off a single household power outlet

I’m not aware of any 2x64 motherboards that can do that

 

[Vario]

I made some optimizations. Here is my Cinebench R20 score. Not bad for a small cute little build.

It's an impressive build, and the case might be limited to ITX only. but the P1 case is anything but small at 422 x 332 x 380 mm (16.6 x 13.1 x 15 inch) and 20 lb. For comparison, my full ATX case is 19x15x8 inch.

 

[venturi]

It's an impressive build, and the case might be limited to ITX only. but the P1 case is anything but small at 422 x 332 x 380 mm (16.6 x 13.1 x 15 inch) and 20 lb. For comparison, my full ATX case is 18x15x8 inch.

Very true. The intent was to build off thermaltake’s only making the P1 a miniITX case. With some will power, a motherboard almost the size of the whole case did fit on it (and 2 cpu, 768gb of ram, 4x Titan V, 10x ssd raid, and power, and very quiet)

 

[Cybrnook2002]

I wasn't even sure where to post it as I have some questions that don't fall into any particular category.

2. I like using the latest bios, but the prior bios without the microcode mitigation is a faster machine, is there a way to edit out a microcode mitigation in the bios and keep the latest one?

I scanned the pages, and did not see mention of it. Look into a tool called "UBU" : https://www.win-raid.com/t154f16-Tool-Guide-News-quot-UEFI-BIOS-Updater-quot-UBU.html
Q/A Thread- https://www.win-raid.com/t4531f16-Discussion-UBU-Tool-related-Questions-Reports-and-Suggestions.html

I use it normally to go the other way, on my X99 systems, I use it to keep my MC up to date, since no vendors still supports X99. The tool gives you the ability to slice out and replace MC from your UEFI BIOS.

One word of caution, M$ also deploys MC updates through windows update. So, what WAS happening for me on some of my X99 boards, is that myself (and users around the world) lost the ability to OC our chips. The reason was that newer MC was deployed via M$ update, yet our boards were still on outdated MC so there was a conflict. Just something to keep in mind.

 

[venturi]

I scanned the pages, and did not see mention of it. Look into a tool called "UBU" : https://www.win-raid.com/t154f16-Tool-Guide-News-quot-UEFI-BIOS-Updater-quot-UBU.html
Q/A Thread- https://www.win-raid.com/t4531f16-Discussion-UBU-Tool-related-Questions-Reports-and-Suggestions.html

I use it normally to go the other way, on my X99 systems, I use it to keep my MC up to date, since no vendors still supports X99. The tool gives you the ability to slice out and replace MC from your UEFI BIOS.

One word of caution, M$ also deploys MC updates through windows update. So, what WAS happening for me on some of my X99 boards, is that myself (and users around the world) lost the ability to OC our chips. The reason was that newer MC was deployed via M$ update, yet our boards were still on outdated MC so there was a conflict. Just something to keep in mind.

I did find where if you’re running MS server - you can using MS provided registry entries set 2 specific variable to not make use of any mitigations, MS provided it as many folks on server were taking a tremendous performance hit.

So in MS server 2019 you can disable / opt out of any mitigations

2nd
I have found a way in this board bios to bypass microcode mitigation in a performance setting


There are several tools around the net to test if you’re protected against the speculative and read ahead mitigations
The tools report when the bios has them off and when the OS has them off. In my case both are off now

 

[Mussels]

If it was on core count alone, then even that 2x64 you refer to would be outdone by any rack box with 4 or 8 cpu sockets

The intent was the most workstation with Audio, 4 way sli or 4 gpu, raid, be very quiet, and run off a single household power outlet

I’m not aware of any 2x64 motherboards that can do that

just a joke, but i do look forward to whatever monster you make after this

 

[Wavetrex]

I somehow doubt the new Threadrippers will go to 64 cores.

However, 2x EPYC 7702 (128 cores) would kick this build's nuts so painfully...

 

[Mussels]

until now i didnt realise epyc and threadripper were different things

learn something new every day

 

[venturi]

just a joke, but i do look forward to whatever monster you make after this

Me too.
That day, I go to turn off the pc after doing some emails and a voice says: “Wait....please don’t!”

 

[Vya Domus]

until now i didnt realise epyc and threadripper were different things

They sort of aren't actually, the physical dies and packages are the same. The only differences come from how they configure them and how the socket is wired, so these differences are pretty much imposed and not necessary.

There is no reason to believe Zen 2 TRs wouldn't go up to 64 cores as well.

 

[Wavetrex]

They sort of aren't actually, the physical dies and packages are the same. The only differences come from how they configure them and how the socket is wired

Having half the memory channels and half the PCI-e lanes is quite a difference in wiring, don't you think ?
Basically, a large amounts of pins on the socket are not used on TR but are used on Epyc.

There is no reason to believe Zen 2 TRs wouldn't go up to 64 cores as well.

Market segregation.

The 28c Xeon W-3175X vs 32c Threadripper 2990WX was a dick-measuring contest between Intel and AMD for the "HEDT" platform.
Edit: Corrected error in Xeon model number.

With this generation, AMD has won and Intel doesn't have anything faster.
Why would AMD offer a 64-core TR and cut into their EPYC sales when they can stop at 48-core and not cut into their high-margin sales?

 

[Vya Domus]

Having half the memory channels and half the PCI-e lanes is quite a difference in wiring, don't you think ?
Basically, a large amounts of pins on the socket are not used on TR but are used on Epyc.

I just said the main difference is the wiring. I didn't denied that.

Market segregation.

The 28c Xeon W-3275 vs Threadripper 2990WX was a dick-measuring contest between Intel and AMD for the "HEDT" platform.

With this generation, AMD has won and Intel doesn't have anything faster.
Why would AMD offer a 64-core TR and cut into their EPYC sales when they can stop at 48-core and not cut into their high-margin sales?

People said the same thing about how a 32 core TR would cut into EPYC sales. But really it didn't and it wont happen this time either, HEDT is a spec of dust compared to the actual server market.

 

[lexluthermiester]

I somehow doubt the new Threadrippers will go to 64 cores.

There supposed to and AMD has been delivering what they promise.

 

[Deleted member 178884]

However, 2x EPYC 7702 (128 cores) would kick this build's nuts so painfully...

Not if you take a look at the use case.