jedbrown

8 ms·

31.
by jedbrown 5y ago
This is honestly quite disappointing compared to prior art. https://metacpan.org/pod/Acme::EyeDrops
32.
by jedbrown 5y ago
It sounds like you're pitching https://hackmd.io/ Also, https://overleaf.com for LaTeX projects.
33.
by jedbrown 5y ago
MyST-Parser/MyST-NB provides a great environment for extensible markdown with Sphinx (fully supplanting any need to work with rST). Combined with sphinx-pydata-theme, you get a clean look with little configuration. We use this for some
34.
by jedbrown 5y ago
Agreed that it was EOL'd because it was priced too low, though the Pro also lifted the double precision throttle (3.5 TF to 6.5 TF) and added Infinity Fabric.
35.
by jedbrown 5y ago
Radeon VII has comparable memory bandwidth to an RTX 3090 (or V100) and 7x the double precision performance of RTX 3090 for a quarter the cost. It has almost double the memory bandwidth of RX 6800 and 3x the double precision performance at
36.
by jedbrown 5y ago
It's informative to think about what is missed in the listed dependencies ( https://docs.github.com/en/github/setting-up-and-managing-yo... ), such as C and Fortran libraries vendored within NumPy and SciPy, pa
37.
by jedbrown 6y ago
Notably, it's easy to make a review site that integrates with a merge request so you can see it before deploying and get visual reviews any time you have multiple stakeholders. https://docs.gitlab.com/ee/ci/re
38.
by jedbrown 6y ago
16 channels of DDR4-3200, versus 8 in single socket EPYC or 4 for Threadripper.
39.
by jedbrown 6y ago
How does one get the Slack sidebar inside a browser tab? I loathe the Electron app, but don't have a different viable way to manage 30-odd workspaces.
40.
by jedbrown 6y ago
Point 2 here is huge. If Apple cared about open source or cross-platform developers, they would pay at least one full-time Homebrew developer and upgrades would be smooth. It speaks volumes that they are swimming in money and can't be
41.
by jedbrown 6y ago
This is backward -- this sort of serialization is overwhelmingly bottlenecked on bandwidth (not CPU). (Multi-core) compute improvements have been outpacing bandwidth improvements for decades and have not stopped. Serialization is a bottlene
42.
by jedbrown 6y ago
Condor and the like are for independent jobs "throughput computing" but the authors here are using MPI for tightly-coupled jobs. SLURM and Flux are actively-developed schedulers for these kind of jobs. https://slurm.sch
43.
by jedbrown 6y ago
Cloud pricing is competitive and thus a good proxy for relative on-prem costs if you keep the on-prem hardware busy. If you want to buy the capability to run the job on-prem, but the hardware sits idle most of the time, then NVMe might be
44.
by jedbrown 6y ago
Unless your data is at rest, you're paying for bandwidth, not capacity. SRAM > HBM > DRAM > NVMe when you're buying bandwidth. Seriously, when we have a low-latency network, the most cost-effective way to run some simulat
45.
by jedbrown 6y ago
The index type for CSR need only index nodes. It's the offsets (one per row) that need to index edges. The blog's portrayal of NVMe vs distributed memory is a bit off: > With modern NVMe drives random seeks aren’t slow anymore,
46.
by jedbrown 6y ago
I recently switched from xmonad to sway and have been using this fork [1] of Emacs. See [2] for install instructions: > We (@masm11 and @fejfighter) are going to push @fejfighter's fork to the fsf's repository, instead of my fo
47.
by jedbrown 6y ago
I have a 13" HP Spectre from mid-2017 with a 4k display. It's noticeably sharper (and brighter) than the 1080p alternatives. Prior to that, I had a QHD laptop from circa 2013. No font size issues on Linux that affected my work.
48.
by jedbrown 6y ago
I think it's between a test written inside a comment (where smart IDE features don't apply) versus `#[test]` in normal code, where you have a rich editing environment and instant feedback (including `cargo build` failing immediate
49.
by jedbrown 6y ago
If you have one (or a small number of) right-hand sides, I would try to make LSQR work. It can find a minimum norm solution even if A is rank-deficient, and you can use preconditioning. Also, if your problem is a good fit for a method like
50.
by jedbrown 6y ago
PETSc developer here. You're correct that we don't have a sparse QR. I'm curious about the shapes in your problem and how you use the rank-revealed factors. If you're a heavy user of SuiteSparse and upset about the licen
51.
by jedbrown 6y ago
One must use SIMD and significant on-node parallelism to saturate a CPU or GPU machine. We have no disagreement there. The question is whether GPUs have a durable architectural advantage for versatile SIMD workloads. Their inverted cache hi
52.
by jedbrown 6y ago
I agree in general, but I'm not confident of this: > GPUs can perform 1000 binary searches faster than a CPU can perform 1000 binary searches. With several vector lanes and ILP on a 64+ core node, the 1000 parallel searches on the C
53.
by jedbrown 6y ago
You're arguing at a particular granularity and even that doesn't hold uniformly. A64FX is HBM2 at equivalent bandwidth to GPUs (with lower power). CCX is much finer granularity than an entire GPU, so not a direct comparison. L3 ba
54.
by jedbrown 6y ago
_mm256_rsqrt_ps is 12-bit accuracy in AVX. https://software.intel.com/sites/landingpage/IntrinsicsGuide...
55.
by jedbrown 6y ago
It may be useful to mention that modern architectures often have vectorized instructions like vrsqrt14ps (accessible via the _mm512_rsqrt14_ps intrinsic) that provides a 14-bit approximation (there are more accurate variants too) in every l
56.
by jedbrown 6y ago
What about a less ambitious goal: fuzzy keyword search that displays the dependency graph and cumulative build time of each node in the package graph (activity stats on hover)? Users can assess for themselves whether a simpler library meets
57.
by jedbrown 6y ago
CodeRunner, group scheduling features for interview grading, batch capabilities with mastery grading, and customizable grade aggregation, to name a few. Canvas core is AGPLv3, but many important plugins are proprietary so I'm not aware
58.
by jedbrown 6y ago
Clearly there are toll roads and many other forms of commercially viable "free market" transportation, so your argument is that some forms of transportation demand public funding. The same applies to software: like cheap and ver
59.
by jedbrown 6y ago
Requesting federal funding for interstate highways is really an admission of failure. It is an admission that the transportation infrastructure concept cannot survive on its own merit in the marketplace.
60.
by jedbrown 6y ago
I'm a professor. Our department has run its own Moodle for years, despite the university switching to Canvas. The university has exerted lots of pressure and this is the first semester in which all classes are primarily on Canvas. Ther

More ›