gnufx

9 ms·

121.
by gnufx 3y ago
The BLIS parallelism strategy is written up in at least one of the references they provide. (Serial OpenBLAS is typically a bit faster than BLIS, but it has (had?) problems with multi-threading.) See also the distributed equivalents (ScaLA
122.
by gnufx 3y ago
The canonical free-less (or no-op free) solution is conservative GC, isn't it? Specifically the Boehm collector..
123.
by gnufx 3y ago
An interesting variant of Pratt which I haven't seen elsewhere, and should have written up, is to allow application as juxtaposition, i.e. shell command-like syntax rather than FORTRAN-like. I don't remember the code, but it was
124.
by gnufx 3y ago
Of some cultural significance in the UK. At least we still have Whispering Bob (Harris).
125.
Annie Nightingale: Trailblazing BBC Radio 1 DJ Dies at 83(bbc.co.uk)
3 points by gnufx 3y ago|1 comments
126.
Crop spray could lead to mass resistance in new-generation antifungal treatments(manchester.ac.uk)
2 points by gnufx 3y ago|0 comments
127.
by gnufx 3y ago
There's a fair amount of wassailing around. I'll be at one of the two episodes just in the environs of Runcorn next weekend. "We hope that your apple trees blossom and bear so we may have cider when we call next year"
128.
by gnufx 3y ago
Those will not simply give you close to optimal performance (>~90% of peak), even, say, POWER10's 4x4 matmul instruction. You need a structure matching the micro-architecture -- the cache and register structure [1]. That's no
129.
What comes after open source? Bruce Perens is working on it(theregister.com)
137 points by gnufx 3y ago|179 comments
130.
by gnufx 3y ago
I can't tell what that involves but, for instance, the two GUI things I typically use with SSH are Emacs (openssh) and x2go (libssh), and they don't do that. Surely you want the agent anyway.
131.
by gnufx 3y ago
Suid and ambient authority suck. I'd rather a more principalled approach.
132.
by gnufx 3y ago
That doesn't help with 2FA, though, or on, say, EL8.
133.
by gnufx 3y ago
Right (what's the threat model)? The possibilities of restricted passphrase-less keys are under-appreciated for non-interactive use, or even interactive use. I'd rather mint an ephemeral key for an endpoint I control than type c
134.
by gnufx 3y ago
If you want to remove a passphrase, just use ssh -p.
135.
by gnufx 3y ago
MIT Kerberos supports preauth with OTP, or PKINIT (X.509 certifies); I don't know what Heimdal currently has. FreeIPA has been doing good work past that, on integrating FIDO, for instance, and can issue tickets on the basis of externa
136.
by gnufx 3y ago
If you want a session with extra permissions you use an appropriate Kerberos principal, e.g. admin/...
137.
by gnufx 3y ago
I don't see how setting up MIT or Heimdal Kerberos is complex compared with AD (which is more than Kerberos); they seem easy enough to me. SSO seems to me what you want, and Kerberos is the reasonable implementation.
138.
by gnufx 3y ago
Password authentication for SSH is often paired with OTP (by people who think that it's somehow more secure to force me to store two lots of credentials for convenience or non-interactive use than keys, and won't do Kerberos or ce
139.
by gnufx 3y ago
Consider appropriate hardware counters. (Note that multiplexing to measure more then three(?) at once can be misleading.) At least for serial code on x86, there's "top-down performance analysis". Some tools (I know Maqao)
140.
by gnufx 3y ago
Use hwloc[1] to determine the hardware topology of the system, and pin processes/threads appropriately. For a heterogeneous system, you presumably want to avoid low-performance cores. OpenMP has a standard way of assigning threads to
141.
Grasping at Straw(lightbluetouchpaper.org)
1 points by gnufx 3y ago|0 comments
142.
Hate Sites Evade the Censor(lightbluetouchpaper.org)
3 points by gnufx 3y ago|0 comments
143.
by gnufx 3y ago
Related (I think they use Microwatt): https://libre-soc.org/
144.
by gnufx 3y ago
Lacking time to study, is this basically what Phil Wadler has written and talked about, like "Propositions as types"?
145.
by gnufx 3y ago
The legislation is https://www.legislation.gov.uk/ukpga/1988/48/part/I/chapter/... Long ago I had to read the Act on microfiche.
146.
by gnufx 3y ago
I don't do AWS any more, but the adapters apparently used with UltraCluster had a latency of 15µs last I knew. I think the NAS benchmarks are overall less latency sensitive than DFT and would benefit less from GPUs, and I assume Ult
147.
by gnufx 3y ago
Azure has RDMA, though with slightly high quoted latency (I don't know the message rate), and tightly-coupled stuff appears to scale: https://techcommunity.microsoft.com/t5/azure-compute-blog/hp... It seems h
148.
by gnufx 3y ago
Many people ignore various actual or equivalent expenses in running on AWS (like optimization opportunities or bad scaling). It's old, but NASA did a study [1] that rings true and I didn't see rebutted. I saw it after an AWS HPC
149.
by gnufx 3y ago
Fortran does not call by reference, and the term aliasing isn't used by the standard. It calls by what's usually called value-return or copy-in/copy-out, according to the "storage association" rules in the standa
150.
by gnufx 3y ago
Unfortunately I see rather few people doing serious optimizing knowledgeably at all in research computing. (I know how and where it's done well, of course.) That has global effects on productivity of users, though some of the biggest

More ›