SPEC says Intel's Xeon processors were using a compiler that artificially inflated the results of its industrial benchmark by as much as 9%.
Intel accused of inflating CPU benchmark results::SPEC says Intel’s Xeon processors were using a compiler that artificially inflated the results of its industrial benchmark by as much as 9%.
But I suppose it’s not that different to when they rigged an industrial phase-change cooler onto one of their CPUs, then pretended it could perform like an AMD CPU lol
Everyone does it, it’s basically expected at this point that any manufacturer will announce that their new chip will run an infinite loop in fewer microseconds than the neighbouring one would.
Very mixed on this. If the compiler is for example using stuff like constexpr in C++ to do major calculations at compile time instead of at runtime then yeah its cheating.
But on the other hand if its doing some Microarchitecture specific optmizations like reordering instructions or replacing certain instructions with others - as long as these are available to the public its fair game i.m.o…
The article states that the compilation was tailored for this specific task, which means it doesn’t represent how the CPU would normally perform it.
So it’s definitely cheating.
A benchmark is not supposed to be a compiler optimization competition. If they showed both, and revealed it was optimized, it would be another matter.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !technology@lemmy.world
This is a most excellent place for technology news and articles.
English
Again?!
Again?
But I suppose it’s not that different to when they rigged an industrial phase-change cooler onto one of their CPUs, then pretended it could perform like an AMD CPU lol
Well, it could*!
*with an industrial cooling system
i will only care if they ever see consequences for it.
Oh, I thought that was assumed, because Intel always does that. At least I noticed it very clearly already back in the 90’s.
Everyone does it, it’s basically expected at this point that any manufacturer will announce that their new chip will run an infinite loop in fewer microseconds than the neighbouring one would.
Ah yes, the Snake oil merchants tried to sell snake oil…again
Very mixed on this. If the compiler is for example using stuff like constexpr in C++ to do major calculations at compile time instead of at runtime then yeah its cheating.
But on the other hand if its doing some Microarchitecture specific optmizations like reordering instructions or replacing certain instructions with others - as long as these are available to the public its fair game i.m.o…
The article states that the compilation was tailored for this specific task, which means it doesn’t represent how the CPU would normally perform it. So it’s definitely cheating.
A benchmark is not supposed to be a compiler optimization competition. If they showed both, and revealed it was optimized, it would be another matter.
Even if the compiler was available to the public most software doesn’t use it, so the benchmark is still not representative of real world performance.