Several months ago, I built a computer with a Gigabyte motherboard and an AMD 9950x3d CPU. It worked for about a week and then stopped. As luck would have it, I was upgrading another system to a 9950x3d so I had an extra brand new CPU of the same model.

I tried the new CPU in the Gigaboard and the system started right up. I put the old CPU back in... and the system still failed.

I tried the old "bad" CPU in the other system with an Asus board... and it was fine.

Why would one CPU fail in one board but keep working just fine in another? I bought them pretty close together so I doubt there's a difference in revision. Both systems have been working without issue for over 2 months now so I'm certain the issue is resolved.

One suggestion I've read is that there are 2 makers of CPU sockets for motherboards (one being Foxconn, I don't recall the other), and it could be that the "bad" CPU didn't quite properly work in whichever socket the particular Gigabyte board has.

Apparently, the different boards of the otherwise exact same model can have have a socket made by other company, it's just luck of the draw. But maybe this is a red herring and it's another issue entirely?

While neither CPU burned up and there aren't any scorch marks in the CPU sockets, I wonder if my weird issue could be tangentially related.

I'm not suggesting anything remotely like that.

With VLIW, parallelism isn't abstracted by the CPU (you get to deal with parallelism directly in the ISA) as speculative execution is moved to the compiler. Unsurprisingly, compilers were never very good at this... But apparently you could get good performance out of Itanium with hand-tuned assembly, which isn't really practical to expect all software to have such optimizations (especially since need to recompile your software for newer VLIW CPU iterations much more than with traditional CPU designs).

Perhaps AI could determine the best set of instructions for the most amount of parallelism (performance). And likely it could do so in a deterministic manner to produce consistent binaries.

That being said, I would think that there are still some advantages to the CPU determining the best way to execute instructions runtime, because a compiler can't know the current state of the CPU.

With VLIW, the speculative execution is supposed to be done by the compiler. The parallelism is built directly into the instruction set rather than obscured by the CPU and then determined at runtime.

Reading about VLIW it's quick to see that this is one of those ideas that must've sounded great in academia, but wouldn't pan out in the real world (at least for general purpose CPU's).

Intel kept saying Itanium will perform, we just need better compilers! But this never really panned out which isn't at all surprising. But with AI, maybe you could actually determine the optimal instructions and parallelism for the best performance.

Apparently you could get pretty good performance (parallelism) with hand-tuned assembly, but to hand optimize all your software is pretty impractical.

I love IBM POWER! I own 2 POWER9 systems from https://raptorcs.com/. A Talos II based system and a Blackbird based system. I also intend to buy one of their new systems in 2025 if I can scrounge up the money.

However, at least POWER9 is vulnerable to meltdown (and I think aarch64 is as well). So far as far as speculative execution vulnerabilities go everything is vulnerable except like Itanium.

Maybe with AI guided compilers VLIW could actually be viable.

I've seen the prices on Samsung 990 Pro SSD's go up recently (across multiple retailers). Unsure if it's limited to just Samsung drives. Hopefully this means prices will go down soon..?

Might wait for Black Friday.

I think most 5000 series CPU's are Zen 3. The 5500 is specifically listed as Zen 3 here so you should be okay: https://en.wikipedia.org/wiki/...

Isn't Moon dust a serious problem? Not only would such caves would need to be pressurized, but all dust would to be cleared up.

https://www.esa.int/Science_Ex...

And it will be an absolute shit-show for Linux. Most things will work. The sound most likely will not.

Newer Lenovo laptops are mostly using Cirrus Logic amplifier chips to control audio. Most likely the CSC3551 (which is supported by the linux cs35l41 driver). The problem will be that Lenovo won't add the entries to the DSD table so that the Linux drivers know how to configure the amp chips: https://bugzilla.kernel.org/sh...

Lenovo has refused to do this for all of their laptops with otherwise supported Cirrus Logic amp chips.

Lenovo,
You have the model specific info needed by the amp drivers, because you have working windows drivers. Just add add this info to the DSD tables in BIOS updates to your various models that make use of such chips. There's really no reason not to at this point. The Linux drivers were written by Cirrus Logic themselves!

Other vendors are doing this. This is not even a Linux-specific thing. Asus has these requisite DSD entries for the CSC3551's in their laptops.

IIRC, The Steam Deck even uses CSC3551!

I'm not asking for official Linux support for your non-supported laptops. But there's no reason to not do the minimal work to get sound working outside of Windows. It's utterly ridiculous.

My point is we no longer need to depend on IBM for POWER. That being said, I really hope P11 works out, but P11 is still a good 1-2 years out I'd think.

I get where you're coming from, but the POWER ISA was opened up by IBM.

So while IBM could stop playing ball... Someone else would be free to implement the POWER themselves. See Libre-SOC as an example of just that.

No, I don't particularly trust IBM. But I do trust Raptor. I don't know if they ever released the full details of their audit... But I think they audited multiple chips/architectures and found POWER was safe.

Absolutely correct, POWER9 is vulnerable to both of these, but POWER9 coupled with Raptor CS's platform means you have a box that is completely owned and controlled by you. The firmware and the BMC are completely open. The schematics to the motherboards are available. You can enable secure boot (I have not), and the keys are controlled entirely by you (there are no "vendor" keys). Raptor has audited the chip for any backdoors and feel confident there's nothing malicious hidden. There's No Intel IME or AMD PSP or backdoors "for Hollywood".

The only thing that can compete in this space is RISCV which was fairly less viable when Raptor launched their Talos II over 4 years ago now. I don't know for sure there we have some completely open RISCV systems yet, but we probably do. May the best architecture/platform win!

Raptor is not providing POWER10 based systems as P10 doesn't meet the standards for openness. IIRC, there's at least 2 closed source firmware blobs required by POWER10 (one is for the memory controller IIRC). It sounds like P10 was a "casualty of COVID" and IBM went the route of paying for some 3rd party tech to get things done within a reasonable timeframe.

There's real hope that POWER11 won't have these issues, especially since IBM is a platinum member of the Open Power Foundtion

Well, POWER9 came out in 2018 IIRC and is 14 nm... So it's a bit behind to be sure. When these cheaps came out, I felt that the price/performance was fairly competitive compared to Xeon. If you look at the prices for just the CPU's alone, for the server market, it's not TOO terrible. I did a small amount of price comparisons back when these CPU's first launched to the Xeons of that day, and thought the prices were fairly good.

A lot of the cost comes from the motherboards. They were expensive in 2018 or so, but then we had tariffs, and then we had supply chain issues due to the global shutdowns and these have really driven prices up.

I built my own Talos II based box using a 2U server chassis to save several thousand dollars. Raptor CS validates the builds they make for you so you pay a lot more... But in the process of building my own, I certainly ran into a number of headaches... Hilariously, one of them is that the IO shield that came with the board was _really_ difficult to get to fit into the Supermicro chassis (same one they were using at the time). Speaking to Tim Pearson of Raptor on IRC, he said it seemed Supermicro's tolerances weren't quite right and it was a pain for them too.

Anyway, the motherboards for these systems are really niche so in addition to all the aforementioned issues contributing to the costs, they aren't sold in high enough quantities to bring the costs down so it's going to be very expensive. So unless you have a really strong interest in the platform (ie, fully audited/trustworthy, have a strong hobby or professional interested in POWER, hate money, etc) there's not much reason to buy. I really hope we can switch to POWER11 at some point, but it's still at least somewhat up in the air so who knows?

The open aspects of these systems are really nice. The BMC firmware didn't and probably still doesn't come with iptables. So I built my own image and added iptables and added a little firewall script. Similarly, the system firmware runs petitboot, which is just Linux as well. I built my own petitboot image to add support for a SAS controller (Adapter HBA 1100-4i) I wanted to boot from and that has been working great for years now.

And finally, I had a fair amount of fun teaching myself some PPC64 assembly.

Not really. If you want hardware in the modern era you can actually trust, this is the way to go though.

I'm a fan of using other architectures out of a hobby, and it did remind me of running old SPARC workstations back in the day.

You can still go with IBM POWER:
https://raptorcs.com/content/b...


I have 2 POWER9 based boxes with the motherboards, CPU's, and RAM purchased from Raptor. There were some growing pains at first, but with some people actually using Linux (and FreeBSD!) on POWER outside of IBM... the experience has become better.

For example, I found that some software didn't like the default 64k page size. I compiled a kernel with 4k page support, and while it fixed some issues... I found other software that had weirdness on POWER with 4k page sizes. Now everything works form e under 64k pages.

Too bad Polaris (the project to port OpenSolaris/Illumos to PowerPC) never went very far.

My only real issue is that we're still stuck with POWER9 has POWER10 has some non-open stuff in iit that can't be vetted... But there's hope POWER11 will work out.