System76 Thelio Ramps Up AI & Creator Performance With New AMD Ryzen Threadripper CPUs

[sophisticles]

Why wouldn't you want to buy systems with the latest built-in AI accelerators for the task?

I actually agree with you one this. But selling AMD TR based systems does not accomplish this.

If System76 wants to target the hardware accelerated AI market then forget AMD and use Xeon Bronze at a minimum that feature hardware AI accelrator built in and are way cheaper than Threadrippers:

https://www.intel.com/content/www/us/en/products/sku/236639/intel-xeon-bronze-3508u-processor-22-5m-cache-2-10-ghz/specifications.html

And if you want to be promoting only an open source platform, then use Intel Arc GPUs that are excellent for AI and have a complete open source stack.

To me, saying we want to target open source AI then offering AMD processors with Nvidia GPUs sends a confusing message to the clients.

[sophisticles]

If anyone here thinks that they can make a successful business out of releasing a proprietary, closed source desktop environment and operating system in the year of 2024, go right ahead. You will not get very far because few will want to use it. Without the influence of a monolith like Apple or Microsoft, you will not be able to a convince anyone that it's worth investing into your platform over any of the existing platforms. Inability to access source code is a liability, not a strength.

You do know that since 2003 Microsoft has made the Windows source code available for inspection?

https://learn.microsoft.com/en-us/security/gsp/onlinesources

Since its launch in 2003, the Government Security Program (GSP) has provided governments and international organizations with the ability to access and inspect source code for a variety of Microsoft products. The Online Source offering provides greater security assurance evidence by enabling access to product source code using Code Center Premium (CCP), a secure web portal. Available products include Windows, Office, SharePoint, Exchange and SQL Server. Agency individuals using Microsoft Entra ID are granted read-only access to CCP to browse, search, and reference product source code.

And you did touch on something that could be a significant revenue generator.

If the Rust DE is any good, by keeping it closed source you could try to sell it to MS, Google or IBM for millions.

[sophisticles]

I don't understand why you keep throwing around "GPL" in your arguments. Why do you think there is an issue with using NVIDIA drivers with CUDA on Pop!_OS or Linux as a whole? There is no legal problems with packaging or installing NVIDIA drivers in Pop!_OS, or any other proprietary blobs included with every installation of Linux today. Additionally, we develop software using a handful of open source licenses, based on the purpose of the project in question. Including MIT, MPL-2.0, and GPLv3.

As i said you are sending a confusing message to you potential clients.

Here's your sales pitch: "We are an open source based company, our OS is open source, our firmware is open source, our software stack is open source and we target the hardware accelerated AI market".

Potential client: "Then why are you offering AMD TR CPUs that lack dedicated hardware accelerators and why are you using a GPU that requires a proprietary closed source driver"?

You: "What would you have us use"?

Potential client: "Xeon processors with dedicated hardware accelerators built in and, Intel Arc graphics that are supported by a completely open source driver stack".

You: "You are a troll, I refuse to waste my time with you!"

Potential client: "Hello, Dell?"

[spicfoo]

If the Rust DE is any good, by keeping it closed source you could try to sell it to MS, Google or IBM for millions.

I really doubt that. I see no reason for any of them pay a single cent for a DE. MS and Google certainly won't do it since neither is interested in desktop Linux. IBM has some investments in corporate workstations but they have an existing client base with an open source DE.

[qarium]

It's been explained multiple times in the past, but the tl;dr is that the justification is the same for why we created Pop!_OS and open source firmware/EC. We are uniquely positioned in the space because we sell systems directly to Linux desktop customers. For any issue that a customer experiences, they will create support tickets that are handled by our dedicated support teams. They also regularly fill out surveys, and provide a lot of positive and negative feedback on different areas of the experience. We've been in the business long enough that we've accumulated a lot data on what our customers want and need. Which then directly influences what problems we work on.

Pop!_OS gave us the ability to resolve many commonly-reported paper cuts and rough edges in the user experience for installation, first time setup, and configuration. The more we invested into Pop!_OS projects, the greater our capacity to enact changes to fix fundamental problems or improve the overall desktop experience. We performed a lot of packaging and optimization work to ensure that users have the best out of the box experience possible, whether they're playing PC games on a hybrid graphics laptop or running CUDA apps. We could backport newer kernels, drivers, firmware, Mesa, etc. and test every update against the hardware in our QA lab. As well as package commonly-requested applications such as Lutris, Slack, Discord, etc.. It's for that reason that Pop!_OS has become quite popular over the years.

Open source firmware and EC brought that same level of care to our laptops. It enables us to significantly enhance system integration with Linux, and address critical problems at the firmware level from the source code without needing to rely on a third party firmware vendor in Taiwan or China to listen to our Linux support issues. Now we can consistently develop and maintain firmware support for a wide number of systems we've ported to coreboot for years to come.

COSMIC is a culmination of all of that and more. It began as a set of GNOME Shell extensions to address pain points such as the lack of tiling in GNOME. Both the development team and many of our customers wanted tiling window management. Many customers have ultrawide and multi-monitor display setups, which were unusable with GNOME for those who've tasted tiling window management. As we continued deeper into addressing common pain points in the desktop environment with shell extensions, it was evident that we were merely biding time until we had the resources to develop a proper desktop environment from the ground up.

The need to create COSMIC increased as we became more and more frustrated with the limitations in extensions, the direction that GNOME is taking with their desktop, and the difficulty working with a platform built on legacy languages like C and JS. Meanwhile, the Rust ecosystem was advancing rapidly in the desktop GUI space. They only needed the helping hand of a company that's willing to glue everything together and push it across the finish line. Thus we were able to leverage top tier projects like Smithay, Iced, Winit, and WGPU to create COSMIC, and in turn we've been developing the missing pieces to the puzzle, such as cosmic-text, while building a modern Wayland desktop environment in Rust from the ground up.

this all sounds really really good. but honestly i thought tiling window management is something from the Xerox PARC ara in 1970 ?

but i admit something goes wrong with gnome if i install more apps then this childish app overview of gnome becomes a multipage wall of app icons without any order ...
other DEs like cinnamon or kde do nicely order the apps by topic.

gnome is really more like a andorid smartphone OS/DE than a serious Desktop DE.

i think this retro tiling window management movement comes from the same pain that these so called modern DE are childish and does sapotage any advanced workflow...

[qarium]

[URL]https://blog.system76.com/post/open-up-contributions-and-collaborations[/URL
This is three years old, but you should be able to get a glimpse of what we've done in the past, despite having only a few developers at the time. Anything that was possible to be upstreamed was upstreamed long ago. That doesn't mean that all of our work is possible to upstream though. And when we develop software from scratch, we are the upstream by definition. All of our software is open sourced, so other distributions are free to use our software. Some do package and use some of our software, such as system76-scheduler, system7-power, or the pop-launcher.
Our development team has grown since then, and our capacity to develop and contribute has grown significantly thanks to COSMIC. As I mentioned above, our cosmic-text library is now universally adopted by most GUI projects in the Rust ecosystem, including wgpu via glyphon, bevy, iced, lapce via floem, theo, vizia, cushy via kludgine, etc. We're now members of the winit windowing group, and have merged improvements for window creation in Rust. We've contributed a handful of improvements to iced since we started building our platform toolkit with. We hired Smithay's maintainer/developer, and all developments to cosmic-comp are also simultaneously boosting smithay's development.

the rust ecosystem is really fascinating. it looks like it plays by different rules than the rest.

the rest is very bound to legacy cruft people are not ready to left behind.

[qarium]

I actually agree with you one this. But selling AMD TR based systems does not accomplish this.
If System76 wants to target the hardware accelerated AI market then forget AMD and use Xeon Bronze at a minimum that feature hardware AI accelrator built in and are way cheaper than Threadrippers:

https://www.intel.com/content/www/us/en/products/sku/236639/intel-xeon-bronze-3508u-processor-22-5m-cache-2-10-ghz/specifications.html

And if you want to be promoting only an open source platform, then use Intel Arc GPUs that are excellent for AI and have a complete open source stack.
To me, saying we want to target open source AI then offering AMD processors with Nvidia GPUs sends a confusing message to the clients.

the ARC GPUs have 16gb vram max... people reported years ago that their radeon7 with 16gvram is not enough and they run out of vram with their AI workloads.... the last gpu i bought for AI workload was a AMD PRO w7900 with 48gb vram...

with only 16gb vram intel will not land any hit i promise this to you.

and your "AI accelerator" argument... first of all there are no "AI accelrators" in reality its just the ability to perform in hardware 4bit fload and 6bit float and 8 bit float datatypes called mini floats https://en.wikipedia.org/wiki/Minifloat

anything bigger than 8bit float is nothing special like my 6-7 year old Vega16 has already bfloat16 floating-point format in hardware.

so do you want to call my vega64 a AI Accelerator ? no of course not that why you talk about 4bit and 6bit and 8 bit float.

but the precision of 4/6/8 bit Float data types are to low for anything else than just AI-inference-mode

and a AI Accelerator is not (!=) AI-inference-mode-accelerator

its really funny that you think a Minifloat AI-inference-mode-accelerator is a AI-Accelerator what is clearly not the case.

[skeevy420]

Those who purchase System76 hardware are not interested in buying into yet another proprietary platform. They are here for the first class Linux support, and all that the open source ecosystem has to offer. They will tell this person with a "business background" that they can take their business and shove it where the sun doesn't shine. Instead, they will be shopping for a machine that has deep integration with open source software from the firmware to the desktop environment. And what better than getting a machine from a company that produces both the firmware and the desktop environment?

I've asked this before, but are y'all ever going to make a case that holds more than 4 2.5 SSDs? I see what y'all offer and when I compare it to something like this one it makes it hard to pick one of the Nebula cases. I say that as someone with limited funds that tries to make every dollar count. A case like that allows a person to expand as needed and even run two systems in one. Being able to have those kinds of capabilities with first class Linux support would be awesome.

Now, speaking as someone who played with the workstation configuration tools on the HP, Dell, and Lenovo sites, they have options for more than 4 drives with various RAID setups, drive docks, different fan options, disk drives, memory speeds, ECC, and more. I think y'all are shooting yourselves in both feet by not designing a more expandable case system that has mounting options that can be adopted for 2.5, 3.5, different fans and radiators, etc so y'all can increase the amount of configurations available to start selling $64K systems to whales like your competitors do with Ubuntu and RHEL systems.

I don't know any other way of stating this, but not offering motherboards for sale does make it feel like a bit of a proprietary platform. By that I mean it's no different than buying the stuff HP or Dell sell at a big box store where if something happens you have to go through warranty or buy another system if it's been too long. Open source from the firmware to the desktop isn't the same as open availability.

Please know I'm trying to say all of that in the nicest, most polite way possible. Feeling and intent don't necessarily go well through text and sometimes I unintentionally write in a rather impolite, rude manner when I'm not being silly without realizing it.