Date: Sat, 15 Sep 2007 23:32:09 +0200 From: Kris Kennaway <kris@FreeBSD.org> To: josh.carroll@gmail.com Cc: Maxim Khitrov <mkhitrov@gmail.com>, Aryeh Friedman <aryeh.friedman@gmail.com>, freebsd-questions@freebsd.org Subject: Re: Building a new workstation - dual or quad-core CPU for FreeBSD 7? Message-ID: <46EC4F59.7070104@FreeBSD.org> In-Reply-To: <8cb6106e0709151421h7bfdeb6fo7dc671820294e9c7@mail.gmail.com> References: <26ddd1750709141351i3646e9bdg8d8b7e93461167f9@mail.gmail.com> <bef9a7920709141441r5c228a8bu1fcf2ea15868c3c@mail.gmail.com> <26ddd1750709151014x2112b022r9bcb999fbf1e7e49@mail.gmail.com> <46EC270A.3020100@FreeBSD.org> <8cb6106e0709151421h7bfdeb6fo7dc671820294e9c7@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Josh Carroll wrote: >> In general, if you are running a multi-process or multi-threaded >> workload, FreeBSD 7 will be able to make good use of 8 CPU cores. >> >> Over the past 2 years we have done extensive benchmarking and >> optimizations that have resulted in *huge* performance improvements on >> many common workloads on 8-core systems. FreeBSD 7 is now regularly >> outperforming Linux on the workloads we have compared. In the near >> future we will be widening our scope to 16 core systems as well as >> investigating more benchmarks as we find them. > > Isn't the default scheduler still 4BSD on -CURRENT? Is ULE considered > stable on SMP systems now, and does it really outperform 4BSD? If so, > will it be set as the default scheduler once 7.0 is released? Yes, 4BSD is still the default, although you definitely want to use ULE for performance reasons (NB: only on 7, dont use ULE on 6). I don't know whether the release engineers plan to change that default, but I will check. > The P965 chipset boards will support the Q6600 and many of them will > support Penryn when it comes out (the 45nm based true quad core Intel > CPU). I have an Asus P5B and a Q6600 running at 3.4 GHz on 6.2-RELEASE > and it screams (8:20 to build world with make -j8, for example). So > even 6.2 will take good advantage of the 4 cores, and I imagine it'll > only get better when 7.0 is released. I'd just avoid the bleeding edge > motherboards/chipsets. JFYI, buildworld is a really bad benchmark for testing SMP performance in general (on 4 cpus it is not too bad), because the makefiles are not written to efficiently parallelize builds on many CPUs, so large parts end up running with only a single make job at a time. Kris
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46EC4F59.7070104>