The Haswell chip (and early Broadwell CPUs) are flawed

[Pentad]

From Tech Report:

"The TSX instructions built into Intel's Haswell CPU cores haven't become widely used by everyday software just yet, but they promise to make certain types of multithreaded applications run much faster than they can today. Some of the savviest software developers are likely building TSX-enabled software right about now.

Unfortunately, that work may have to come to a halt, thanks to a bug—or "errata," as Intel prefers to call them—in Haswell's TSX implementation that can cause critical software failures.

An Intel spokesperson has provided Tech Report with a brief statement on the TSX erratum, confirming that Intel has "addressed the issue" and "disabled the TSX feature on affected products.""

Using Intel's CPU ID application for the Mac(which you can grab here) shows that HLE and RTM are present in the MBPr CPUs.

I'm not sure I would consider turning this feature off a fix.

I guess I feel like Intel owes its customers more. We were charged more for this chip than a lesser Intel chip because TSX was one of the hyped features on the higher spec Haswell CPUs. Even if I never use it, it was available and I paid for it. It just doesn't work, it will crash your machine if used.

I feel that Intel owes its customers more than 'we fixed it by turning it off!'


You can see the article here.


-P

 

[X-Ravin]

It stinks, but it happens. Take a look at the notorious Pentium Pro FDIV bug or the AMD Phenom TLB bug, those were far worse. As computer engineer who has worked with CPU architecture, I'm amazed at how few bugs make it into production chips. They are just too complex anymore for full validation.

 

[Good User]

What are HLE and RTM?
I just checked the intel core i5 4258u that is in rMBP and It has no TSX instructions. http://ark.intel.com/products/75990/Intel-Core-i5-4258U-Processor-3M-Cache-up-to-2_90-GHz

 

[dusk007]

AMD's TLB bug affected everyone and meant 10% less performance on all CPUs that had it disabled. This TSX bug affects virtually nothing anybody here should care about.
It is just an easier locking feature which helps usually software on servers with many cores that doesn't scale all that well because of too many locks.

It basically lets independent threads assume they can edit all data and just do their thing instead of each thread locking some amount of memory while it is working on it. If there is an issue, as in there should have been a lock, the CPU just goes back and starts over and excutes the whole thing again with locks. Since most of the time there won't be any issues the threads don't have to wait for locks and the software developers can use less fine tuning on their locking scheme as it will not hinder performance as much.

The feature is nice to have at best. I doubt there is much software that would run any slower at all much less noticeably slower that people use on notebooks. Don't forget software like video encoding scales well with many cores and has no need of TSX. Just because something uses lots of cores doesn't mean it benefits. Only software that is written for multiple cores but doesn't scale well because it isn't perfectly optimized would benefit. Software that either scales well or is mostly single threaded in nature anyway (or was never programmed for parallel execution) just doesn't get anything from TSX.

It really is a none issue. If you bought an expensive server cluster with muliple expensive Xeon CPUs and a workload that benefits from TSX things would be different.