Threadripper Pro - for local AI and other adventures

I finally did it - took the plunge into Threadripper right in the deep end. I have a few Ryzen 9000 series builds and I just got sick of having limited PCI lanes (in addition to wanting a serious AI setup) . Why did AMD and Intel restrict their PCI lanes? I have gone from 24 to 124 available PCI 5.0 lanes in one giant expensive step.

Here is an opening snapshot. notice that there is no case except one that is wonderful but not big enough (HS 420).

Everything is so dark I think I will use another white case.

Parts you may be able to id:

massive ASRock 1650 watt PSU
Silverstone XE360-TR5
ASUS Pro WS WRX90-SAGE SE
128GB (4x32) Kingston Fury DDR5 5600 cl28 ECC
Threadripper Pro 9975WX
ASUS PRO R9700
HAVN HS 420 (not for this build :frowning: )
Custom IKEA table with wheels
our Hero describing the WRX90-SAGE when it was first released

OK time to get organized and try for a first boot

2 Likes

“A Threadripper build? At this time of year? At this time of day? In this economy?”

“Why, yes!”

“May I see it?”

“…No. The Fed raised rates again.”

“Ah.”

“And inflation ate the GPU budget.”

“Good Lord!”

“The motherboard is technically affordable, but only because I financed it over 84 months.”

“84 months?!”

“Local economists call it ‘consumer resilience.’”

“Is that what they call it?”

“Yes. The unemployment rate is 4.1%, real wages are getting mugged in an alley, housing is consuming half of a typical family’s income, and the national two-bedroom housing wage is $34.73 an hour.”

“.. but what about Steamed hams?”

“Threadripper waste heat for cooking.”

“At this time of year?”

“At this time of year.”

“In this economy?”

“…Adjusted for inflation, absolutely not.”

10 Likes

Torrent? Define 7 XL? Antec 900? Something else?

I am looking seriously at a white Fractal North XL but I am afraid of not having enough room

I’m slightly jealous… Nice rig! :slight_smile:

If you’re going to use all the slots for GPUs, then perhaps look at the Silverstone RM51, it has 8 PCI-e expansion slots, so you can mount four GPUs in there and it can be used both as a tower as well as a rack mounted chassis.

Cant believe ASUS website is down! Trying to get a new BIOS

1 Like

Thanks for the recommendation - I am only going to start with 2 R9700s and maybe pop a 9070XT in there temporarily. I looked at the Silverstone 2S and will take a look at this RM51 too

1 Like

Unintentional terminator reference?

:thinking:

1 Like

Meanwhile I was able to attempt a post only after plugging in ALL the MB and CPU power connectors - I think it trained memory and then I got to d6 "No Console Output Devices are found " - and the Video card is not blowing. I think I may need to find a switch for VGA or BMC if recall…

have to go check on the chickens for now

1 Like

Congrats on the build, it a rad system.

The BMC video error is probably caused by a annoying default setting ASUS prevent you from changing in the bios.
If the bios detects a pcie gpu got installed it automatically sets it as the default video output for the system.
The ipmi web console is unable to access the video out on a pcie gpu and thus giving the no output device error.
Did you have a monitor connector connected to any of the gpu’s btw?
Because if you have not screen connected to a gpu usually runs in standby like state.
Wich could explain the no fan spin on you gpu. The also might just have a 0rpm mode if they are cold enough.

Oke I messed up the replay somehow didn’t know how to fix it, so I just deleted

I took a look at the mini switches on the board.
While you can disable the integrated vga gpu and the entire bmc
Non of them are able to set the display output they moved this all to the bios I guess.

ASUS site is back online - got BIOS and several firmware updates

after switching back on 3 switches I had turned off (one affects video but not exactly how it works due to the terrible manual) and getting a different display port cable and replacing a Y PCIe cable for one of the power connections (PCIE_CPU_12V_1) with a single end, I have gotten past several more codes and now hung on Q-92 “PCI Bus Initialization is started”

It has been ther for 20 minutes so I will try on more time and them will flash the BIOS with the latest version 1317

I don’t even know what version I am on bot I do not think it is the earlier versions from 2 years ago since I have gotten this far in the POST.

SO the GPU blower comes on and it gets through RDIMM and yellow, white, goes to green but stops at 92 (PCI init)

Now it is getting to 98 “Console Input devices connect” and it was doing something but seems to be hung on that now - the blower had stopped running and has never ramped up.

So.. I will prepare the BIOS Flashback process and see if that will help.

Have you been able to get successful post at all from the board and access the bios?

No this is all an attempt to get to the BIOS - it seems to not consistently recognize or connect to the GPU - if the BIOS Flash does not get me in I will next move the GPU to a different slot.

Do you have a monitor connected any of the gpu’s?

And I think you are making things unnecessarily complex for your right now.
Usually it’s better to simple and add a variable to the process on at a time.

I suggest you take out all pcie cards including the gpu’s.
And see if you can get the system to boot and in the bios.
After you get a successful boot add one card to see if a specifiek one causes a failed boot

OK I am in the BIOS!

After updating to 1317 BIOS using FlashBack I tried again and it stuck on code 98 “Console Input devices connect”

when I clicked a few keys it seemed to react but did not really get further

So I moved the GPU to slot 4 from slot 1 - and that may have been an issue.

I reset my RDIMMS to the EXPO setting - relearned and came up successfully to 5600 cl28

Will look through the BIOS and probably put an SSD in and load Win 10 Pro tonight

1 Like

Alright, glad to hear you got it posting.
I Just remember something, can you check if above 4g encoding is enabled in the bios?
It has something to do with size limit on hardware addresses or the memory map.
32GB vram is just under the limit before the system has no spare address to assign,
You could probably guess 64GB of combine vram will have some issues.
The above 4g encoding opinion might just fix a lot of pcie issues in general.

Oh and one more thing before I leave to sleep.
In case you have issses with not being able to install windows,This might help.
Do you have a tpm 2.0 module installed in you system?
It’s a hard requirement for being able install Window you have a tpm enabled.
Until it is not, Microsoft was a special extended support edition of Windows 11 with no tpm requirement.

I suggest you try to tack down a Windows 11 IoT Enterprise LTSC iso.
Than you can sort out the tpm issue later.

I am trying to learn Threadripper Pro and this historically frustrating ASUS motherboard step by step in the real world after watching several videos - not too worried about eliminating every last variable as long as I am aware of the variables. Here at home I have that luxury while I watch out for my cat trying to use all these parts for toys, and make more coffee!

I am just learning about threadripper but know AM5 and lots of Microsoft corporate tech

  • I only have 1 GPU in the system so far but will add a second soon.

I use Win 10 Pro and just reinstall often - will probably recreate at least one Win 11 machine soon

thanks for the tip about above 4g encoding - as for tpm, yes it has it but will not matter for a while since I stick to Win 10 Pro for now.

I feel that I am very behind in the ROCm and Vulcan world - I have a friend who is helping me some with AI concepts but he is in the Mac world.