Date: Fri, 12 Mar 2004 11:59:59 -0800 From: Brooks Davis <brooks@one-eyed-alien.net> To: Jem Matzan <valour@thejemreport.com> Cc: freebsd-amd64@freebsd.org Subject: Re: Peer review of AMD64/FreeBSD article Message-ID: <20040312195958.GA32345@Odin.AC.HMC.Edu> In-Reply-To: <4052130E.9060601@thejemreport.com> References: <200403121301.i2CD1oQC076505@lurza.secnetix.de> <4051B7D3.8020404@thejemreport.com> <20040312174736.GD7661@Odin.AC.HMC.Edu> <4052130E.9060601@thejemreport.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--gKMricLos+KVdGMg Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Fri, Mar 12, 2004 at 02:44:14PM -0500, Jem Matzan wrote: > Brooks Davis wrote: >=20 > >On Fri, Mar 12, 2004 at 08:14:59AM -0500, Jem Matzan wrote: > >=20 > > > >>Hyper-Threading seemed to help with processes that didn't require a=20 > >>heavy CPU load. The OpenSSL tests show it being markedly faster in the= =20 > >>smaller algorithms, but lagging way behind the 64-bit Athlon64 when the= =20 > >>serious number crunching comes into play. Intel's press kit shows HT=20 > >>(and SSE3) giving an advantage when multitasking with four desktop=20 > >>programs in Windows XP. It's just too hard to show that reliably though= =2E=20 > >>There's a lot of anecdotal evidence to suggest that AMD64 is faster on= =20 > >>the desktop (in X) in 64-bit mode than the Prescott is in 32-bit, but= =20 > >>I'm having trouble proving it. > >> =20 > >> > > > >I think it would be a mistake to assume the HT is what accounts for > >the performance difference. There are so many other architectural > >differences it's hard to see how you could isolate the effects of > >HT. My suspicition is that better performance on small=20 > >algorithms is due to them being more or less memory bound (and thus > >similar to the pure synthetic benchmarks). > > > By comparing the Pentium4 to the Athlon64 in i386 mode, you can better=20 > see the advantage of HT Technology. This is especially evident in the=20 > OpenSSL tests. I strongly disagree. All you can see is that they differ. The architectural differences between P4 CPUs and current generation amd64 CPUs are a whole lot more then HT. You've got a different memory memory system, vastly different pipeline lengths, etc. For that matter, on paper at least, I wouldn't expect HT to help much if any in this situation since you're not trying to do two things at once. -- Brooks --=20 Any statement of the form "X is the one, true Y" is FALSE. PGP fingerprint 655D 519C 26A7 82E7 2529 9BF0 5D8E 8BE9 F238 1AD4 --gKMricLos+KVdGMg Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.1 (GNU/Linux) iD8DBQFAUha9XY6L6fI4GtQRAokIAKCycB2+Tzi3ArakGZCGqThmJ25MLQCdHpdR jKUiWEUi00s/FRIDbShYIYY= =99Tq -----END PGP SIGNATURE----- --gKMricLos+KVdGMg--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040312195958.GA32345>