AMD fixes Ryzen bug on Linux

[cobito]

Just over two months ago, some Linux users who were testing their Ryzen processors noticed an issue when compiling programs under certain circumstances.

Read more...

[Yorus]

Oops, I was unaware of that flaw but it doesn't surprise me that it came out, because the Ryzen stand out in their multi-core work capacity and make them ideal for that type of task among others.

[palotes]

People think that processors are infallible devices, but the reality is quite different. If you take a look at the Haswell bug list, you'll see that it's basically a 5-page manual of failures that aren't even solved yet, where they tell you how not to do certain things to avoid problems. In the Piledriver architecture, the 256-bit AVX set introduces a delay of about 20 cycles in each execution. And beware of the patch for all of you who have an Atom prior to the Bay Trails because for every floating-point instruction that is executed, a NOP operation is introduced that keeps the micro doing nothing for one cycle. That means that the calculation of reals in the Atoms goes at half the speed of what it should, but since no one uses the Atoms for that, the topic isn't talked about much.

It's normal for there to be bugs in processors. Sometimes they are seen (like this one) and other times they go unnoticed or are only visible for software development. I find it strange that AMD hasn't said anything about this, especially if they supposedly solved the problem.