After 15 years of using Intel-based processors for his Linux-building powerhouse computers, Linus Torvalds switched to an AMD Threadripper 3970x-based “frankenbox” for building the world’s most important operating system, Linux. Now, months later, Torvalds is glad of the move and wrote that he’s “very happy with AMD these days.“
That’s because Torvalds explained in a Real World Technologies discussion forum, you get more bang for the buck from AMD processors. But what about Intel’s high-end Xeon CPUs? Torvalds said frankly, “I used to look at the Xeon CPUs, and I could never really make the math work. The Intel math was basically that you get twice the CPU for five times the price. So for my personal workstations, I ended up using Intel consumer CPUs.”
As for AMD, on the other hand, “The AMD Threadripper pricing is much closer to ‘twice the price for twice the CPU.’ Yes, you end up paying more for the accouterments (MB and cooling), but that’s pretty much in line too. So yes, it ends up being more expensive, but if CPU power is what you want and need, the expense is pretty much in line with what you get.”
Of course, AMD has its own server CPU line, the Epyc processor family and, as Torvalds admitted, “You do pay more for that privilege, but at least AMD doesn’t try to screw you over and limit their non-server parts. So you do get ECC for Threadripper (and plain Ryzen) too, even if it’s not necessarily ‘officially verified.'”
ECC? This stands for Error-correcting code memory. ECC memory chips can detect when a memory error happens. There’s long been an assumption that memory errors are extremely uncommon. Torvalds, who knows a thing or two about memory, vehemently disagrees.
Torvalds snarled, “The ‘modern DRAM is so reliable that it doesn’t need ECC’ was always a bedtime story for children that had been dropped on their heads a bit too many times.”
“We have decades of odd random kernel oopsies that could never be explained,” added Torvalds. These “were likely due to bad memory. And if it causes a kernel oops, I can guarantee that there are several orders of magnitude more cases where it just caused a bit-flip that just never ended up being so critical.”
Today, ECC memory is hard to find and expensive when you do find it. Torvalds puts the blame squarely on Intel for this sad state of affairs. “Intel has been detrimental to the whole industry and to users because of their bad and misguided policies wrt [with regards to] ECC. Seriously.”
Torvalds adds that isn’t just his thinking. “And, if you don’t believe me, then just look at multiple generations of rowhammer [A memory security attack], where each time Intel and memory manufacturers bleated about how it’s going to be fixed next time. Narrator: ‘No it wasn’t.’
The root cause, according to Torvalds, was entirely about Intel’s “misguided and arse-backward policy of ‘consumers don’t need ECC,’ which made the market for ECC memory go away.”
The memory manufacturers also come in for their fair share of blame too. “The memory manufacturers claim it’s because of economics and lower power. And they are lying bastards – let me once again point to rowhammer about how those problems have existed for several generations already, but these f*ckers happily sold broken hardware to consumers and claimed it was an ‘attack,’ when it always was ‘we’re cutting corners.'” Still, Torvalds puts the lion’s share of the responsibility on “Intel was pushing shit to consumers.”
This isn’t a new problem. Torvalds reminds us that “You can find me complaining about this literally for decades now. I don’t want to say ‘I was right.’ I want this fixed, and I want ECC.” AMD, which unofficially supports ECC memory, “did it. Intel didn’t.”
Moving back to CPUs from memory issues, Torvalds wrote, “I’m personally very happy with AMD these days. I used to absolutely despise their horrible bulldozer cores, but I think they’ve had a home run with their Ryzen series and their chiplet approach. Not just because they fixed their cores, but because their chiplets made it so much easier to do the scaling they do and offer close to that ‘twice the cores for twice the price’ model.”
Would Torvalds consider going back to Intel-chip based workstations? That would be not just no, but hell no. Torvalds concluded, “Intel with their HEDT and Xeon chips that required different boutique silicon (and thus the excessive pricing) is dead to me unless they seriously fix their sh*t. I’ve been complaining about their ECC policies here on this forum for about two decades by now. Good effing riddance – because once Intel stopped offering the best bang, there was absolutely no advantage to staying with them.”