304 North Cardinal St.
Dorchester Center, MA 02124
Linus Torvalds, who made Linux, has said that Intel is stopping people from using error-correcting memory on a large scale and “helping to kill the whole ECC industry with its terrible market segmentation.”
ECC stands for “error-correcting code.” ECC memory adds extra parity bits to make sure that the data read from memory matches the data written to memory. Without this check, memory could get messed up if, for example, background radiation caused a bit to flip on its own. By quickly reading the same memory locations over and over again, a technique called “Rowhammer” can also be used to change the state of nearby memory locations.
ECC memory solves these problems—well, sort of, in the case of Rowhammer—but even though it’s been around for more than 50 years, most personal computers don’t use it. Cost is a factor, but what really bothers Torvalds is that Intel has made ECC support part of its Xeon line, which is for servers and high-end workstations, but not in other lines, like the Core line.
“With today’s fast CPU and memory speeds, soft memory errors are happening more and more often,” says a promotional video on Intel’s website. When important data gets messed up, it can be bad for business and the company’s reputation.” ECC memory and Intel’s Xeon processors work together to automatically find and fix errors in soft memory.
“Every year, one in three systems has one or more fixable memory errors,” says an Intel ad. There is ECC memory on servers, but not on desktop PCs.”
The topic came up in a discussion about AMD’s new Zen 3 Ryzen 9 5000 series processors on the Real World Tech forum. Most AMD processors have ECC support, but this isn’t really official. A contributor who hadn’t thought it through said, “I don’t think AMD’s unofficial support for ECC is that big of a deal.”
“ECC is very important,” said Torvalds in response. “Intel’s bad and misguided ECC policies hurt the whole industry and the people who use it. Seriously. And if you don’t believe me, look at how Intel and memory makers have been complaining about how it will be fixed next time… Again, this was all because of the stupid and backward policy that “consumers don’t need ECC,” which killed the market for ECC memory.
The accusation is important, especially now when everyone is worried about security. Even though rowhammer is just one type of attack that could be used, the idea is that Intel’s marketing choices have slowed the spread of a technology that makes users safer and PCs more stable.
“The arguments against ECC were always nonsense. “Now, even the people who make memory are doing ECC on their own because they finally admitted that they have to,” Torvalds said.
Torvalds said that people didn’t use Xeon because of how much it cost. “I could never figure out how Xeon CPUs worked when I looked at them. Intel’s math showed that for five times as much money, you get twice as much CPU. So, I used Intel’s consumer processors in my own workstations.” He said that Ryzen and Threadripper made prices go down last year, but it was “too little, too late.”
As a way to make up for it, he said, “[Intel’s] consumer offerings were fine, except for their position on ECC.”
Intel’s marketing strategy may be working against it now since ECC could be a selling point for AMD, even though ECC memory needs support from both the motherboard and CPU. “There was a lot of pent-up demand for alternatives because of Intel’s marketing tricks. “The ARM people would have loved to fill that need, but for now, AMD stepped up,” Torvalds said.
ECC support in AMD’s Ryzen line is unofficial at the moment. This is a problem because many motherboards don’t support it or don’t say they do, so business users will be hesitant to rely on it.
Intel sent us a statement that said, “Intel supports and verifies ECC on products where this feature is recommended by the industry. The industry has decided that ECC memory is most important for keeping data safe in servers and workstations with Intel Xeon processors.
READ MORE ARTICLES: