Joe

=E2=80=94

We = have learned a lot about each other this century and it's not = pretty.

<= /div>

= --Apple-Mail=_358CDE3E-3E05-4AB7-BEA4-1563CC82A592-- From owner-chemistry@ccl.net Thu Dec 3 19:05:01 2020 From: "=?UTF-8?B?VGlib3IgR3nFkXJp?= tiborgyri- -gmail.com" To: CCL Subject: CCL: M1 Macs/double-precision calculations Message-Id: <-54231-201203184645-6390-dHGIIWkNRpFFEQ4u78l6bg:+:server.ccl.net> X-Original-From: =?UTF-8?B?VGlib3IgR3nFkXJp?= Content-Type: multipart/alternative; boundary="000000000000ab97c405b597f8bf" Date: Fri, 4 Dec 2020 00:46:19 +0100 MIME-Version: 1.0 Sent to CCL by: =?UTF-8?B?VGlib3IgR3nFkXJp?= [tiborgyri_+_gmail.com] --000000000000ab97c405b597f8bf Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Dear Joe, while I cannot comment on the software situtation too much (other than the fact that the lack of high-performance BLAS/LAPACK libraries might be an issue, especially for packages that currently only work with Intel MKL), I do not think any of the current M1-based machines are well suited for calculations requiring serious performance. If the goal is to have a inexpensive, but very capable desktop computer for development and running substantial QM/MM calculations, I still think AMD/Intel systems are a better value, and will continue to be useful for much longer than Apple computers, due to their ease of upgrade and repair. It is fairly limited what you can do with an M1 system at the moment. You cannot hook up an Nvidia or AMD GPU (extremely useful for MD), max. RAM capacity is I believe 16GB (with no possibility of expansion, as all RAM is soldered to the motherboard) and the SSD is also soldered to the motherboard (QM calcualtions will likely wear it out quickly). In general, at the moment I would recomend avoiding M1-based systems for QM/MD, not because the CPU itself has anything wrong with it (it seems to be a fairly capable little beast, for the amount of power it uses), but the entire hardware platform around it is severely constraining, when compared to modern AMD (and even Intel) systems. Best regards, Tibor Joe Leonard jleonard42#%#gmail.com ezt =C3=ADrta (id=C5=91pont: 2020. dec. 3., Cs, 23:38): > Folks, I know this is a brand new (well-reviewed) platform, but I am > curious whether anybody has built a larger-scale/double-precision > applications like an MD or QM package on them. While raw performance for > everything is great, I live in a business that uses higher-precision > doubles all the time=E2=80=A6 Does anybody have any preliminary(?) info = as to how > these compare to modern Intel/AMD chips? > > Thanks in advance! > > Joe > =E2=80=94 > We have learned a lot about each other this century and it's not pretty. > > --000000000000ab97c405b597f8bf Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable

Dear Joe,

while I cannot comment on the=20 software situtation too much (other than the fact that the lack of=20 high-performance BLAS/LAPACK libraries might be an issue, especially for pa= ckages that currently only work with Intel MKL), I do not think any of the = current M1-based machines are well suited for calculations=20 requiring serious performance.

If the goal is=20 to have a inexpensive, but very capable desktop computer for development and running substantial QM/MM calculations, I still think AMD/Intel=20 systems are a better value, and will continue to be useful for much=20 longer than Apple computers, due to their ease of upgrade and repair.

It is fairly limited what you can do with an M1 sys= tem at the=20 moment. You cannot hook up an Nvidia or AMD GPU (extremely useful for=20 MD), max. RAM capacity is I believe 16GB (with no possibility of=20 expansion, as all RAM is soldered to the motherboard) and the SSD is also s= oldered to the motherboard (QM calcualtions=20 will likely wear it out quickly).

In general,=20 at the moment I would recomend avoiding M1-based systems for QM/MD, not=20 because the CPU itself has anything wrong with it (it seems to be a=20 fairly capable little beast, for the amount of power it uses), but the enti= re hardware platform around it is severely constraining, when compared to modern AMD (and even Intel)=20 systems.

Best regards,

Tibor

= Joe Leonard jleonard42#%#gmail.com <owner-chemistry : ccl.net> ezt= =C3=ADrta (id=C5=91pont: 2020. dec. 3., Cs, 23:38):

Folks, I know this is a brand new (well-reviewed) platform, but I am cu= rious whether anybody has built a larger-scale/double-precision application= s like an MD or QM package on them.=C2=A0 While raw performance for everyth= ing is great, I live in a business that uses higher-precision doubles all t= he time=E2=80=A6 =C2=A0Does anybody have any preliminary(?) info as to how = these compare to modern Intel/AMD chips?

Thanks in advan= ce!

Joe

=
<= div style=3D"text-align:start;text-indent:0px">
<= div style=3D"text-align:start;text-indent:0px">
=E2=80=94
We have learned a lot about each other=C2=A0this cent= ury and it's not pretty.
<= /div>
<= /div>
<= /div>
<= /div>
<= /div>

--000000000000ab97c405b597f8bf--