News for hackers. Stuff that matters.

A comment on Apple Preps Next Mac Chips with Aim to Outclass Highest-End PCs. Parent comment

Re:Apple Preps Next Mac Chips with Aim to Outclass Highest-E... by dcolkitt on Monday December 07, 2020 @11:08AM (#25333464)
> I think one problem for CPUs specifically is that people plainly just don't care as much about speed anymore. Modern processors are extremely fast.

I disagree. At least in the datacenter there are still lots of workloads that are CPU bound. And many of these workloads are not easy to parallelize. Right now there are about a bajillion startups in a Python codebase that are hitting scaling limits, where the single-core GIL limits are starting to hurt responsiveness.

Yes, they could re-write into Go and get higher performance and better concurrency at the same time. But most orgs don't have the bandwidth for re-building their software stack from scratch. They'd much rather throw money at hardware.

My guess is if you had a datacenter CPU with 2-3X the single-core threaded performance of top-level Xeons, it would sell like hotcakes.

[ Parent | on HN ]
Re:Apple Preps Next Mac Chips with Aim to Outclass Highest-E... by thu2111 on Monday December 07, 2020 @12:00PM (#25334004)
That particular issue is much better solved by a JIT compiling GIL-free Python implementation like GraalPython.
[ Parent | on HN ]
Re:Apple Preps Next Mac Chips with Aim to Outclass Highest-E... by dcolkitt on Monday December 07, 2020 @12:21PM (#25334287)
Better solved depends under the organizational and operational calculus you’re working under.

Many companies would rather throw money at beefier hardware than play around with experimental JITs that may or may not break standard behavior in unpredictable ways.

[ Parent | on HN ]
Re:Apple Preps Next Mac Chips with Aim to Outclass Highest-E... by cercatrova on Monday December 07, 2020 @03:10PM (#25336695)
That's taking slow software and complaining that it's slow due to hardware rather than making the software itself faster, either by rewriting it or optimizing it another way.
[ Parent | on HN ]