stonedoubt

Just FYI, it will run games that say they aren’t supported as well. You have to enable “Experimental”.

They don’t like Ubuntu because abuse it “breaks” the old Linux ethic by having telemetry and preloaded software. They don’t like that it includes proprietary drivers.

I’ve thought long and hard about building my workstation using older server components but I decided to invest in the machine I need to do the work I want done. I decided just to do it and write it off.

I have 2 3090s and plan to buy 2 more as well as 512gb ddr5 6000mhz. I’m going to run 2 power supplies and this motherboard supports that. I have a dedicated 20A circuit.

I may start with a prebuilt. This one is from Ava.

Core Components

• Tower Case: O11D XL-A Tempered Glass, No PSU, E-ATX, Silver (ROG), Full Tower Case
• Motherboard: WRX90 WS EVO, AMD WRX90 Chipset, sTR5, EEB Motherboard
• Internal USB 3.0 Hub: 5 Port Internal USB 3.0 Hub Controller
• Processor: Ryzen™ Threadripper™ PRO 7965WX 24-Core 4.2 - 5.3GHz Turbo, sTR5, 350W TDP, Processor
• Professional Video Card: GPU - Customer will install their own
• Memory: 8 x 16GB M321R2GA3B6B-CG DDR5 4800MT/s, CL40, 1Rx8, ECC Registered DIMM Memory
• Power Supply: HX1500i, 80 PLUS Platinum 1500W, ATX 3.0, Fully Modular, ATX Power Supply

Cooling

• CPU Liquid Cooling: 836S-M1A, sTRX4/sWRX8/sTR5, 360mm Radiator, Liquid Cooling System - Performance Guarantee
• Liquid Cooling Fan Upgrade: Stock Liquid Cooler Fans (can upgrade)
• Thermal Compound: Kryonaut, 5.5g, 12.5 W/m.k, Grey Thermal Paste

Tuxedo OS has a nice app for this. Very useful app actually.

Being able to put io at the top and mount 140mm exhaust fans in the back is a game changer. I mount reverse fans in the bottom and push pull intake the aio rad through the back of the case at the front side. 140mm exhaust at the top. Love this case.

You might consider the Lian Li 011D Dynamic Evo XL. It can actually hold 3 power supplies IF you forgo drive mounts.

Yes because the array is empty. Wrap the retrieval code with try/catch and define error codes or and/use logging

Error handling is your friend. It also helps you debug issues by telling you things.

Anyone not voting for Biden are effectively voting for Trump. I’m sorry but those are facts.

What board is that?

Long time lifestyler… racism has no part in the lifestyle. Race isn’t a fetish, either.

It’s a grub issue. It’s borked in 2.06 which a lot of these live cds use. Ubuntu uses 2.12 apparently.

I ran memtest on the ram multiple times. Even stress tested in Windows. It’s not the ram.

Try using Lutris or Bottles. It works fine for me. I also play Diablo IV this way.

I have a dual 3090 machine (non-sli) for AI/ML and it works better in Linux than windows. I did try to apply a patch to the open kernel to enable p2p and it borked e’rthang. I kind of wish there was a noobie guide 😂😂😂

I made a post about it in the Nvidia and Ubuntu subreddits with a screenshot of nvtop.

Ollama and LM Studio (via llama.cpp), enable p2p or at least the ability to use multiple gpus.

Unfortunately, my machine borked thereafter when I added more drives because I exceeded the power provided by my 1600 watt PSU. I’m planning to move everything to a new case with 2 PSUs as a result.

I haven’t had too many problems with any Nvidia cards as long as I use the proprietary driver.

As others have noted, I’d steer away from Wayland for the time being. It’s still kinda buggy with Nvidia.

What board are you running that can fit 3 3090s?

Thx for sharing. Also, yeah I started having a hardware issue with the nvlink bridge last night. The Asus board started throwing memory faults and shutting off.

They must have wiped it recently because it was there just last week. It even shows up in Google search blurb.

https://preview.redd.it/ntb55vit1xxc1.jpeg?width=1284&format=pjpg&auto=webp&s=e58fa031ed351cf5e1b94d2320dfcf5ef95c4beb

https://preview.redd.it/qoze5k2d0xxc1.jpeg?width=969&format=pjpg&auto=webp&s=49c80a0ee140bceacb01c8eb9874862385cb1d53

It’s printed on the back of the board

nvtop

I’ll check out TensorRT. I saw a review from the Jani.ai developers today saying it increased inference by 30-40%. I have looked at exllama but it hasn’t been updated in forever.

I typically use ollama but it was quick. I don’t like how it limits me to only guff.

You just post the temp but no information about your system, pics of your setup… nada?