winwang

30 ms·

271.
by winwang 2y ago
Thanks!
272.
by winwang 2y ago
I definitely resonate with #1 and #3, but I got bills to pay. And some of the things I want to see in the world, well, they would require a lot of money.
273.
by winwang 2y ago
Currently a technical solo founder. Got any video recs?
274.
by winwang 2y ago
I think the vast majority of users just want a good experience. UI fluidity, timeliness of notifications, DMs, image/video quality, etc. Decentralization is near the last of their concerns.
275.
by winwang 2y ago
I'd believe more in a heterogenous chip (e.g. MI300X, Apple M series, or even APUs) than in completely new chip tech.
276.
by winwang 2y ago
This is not true. Take the NVidia 4090. 128 SMs = 4x128=512 SMSPs. This is the number of warps which can execute independently of each other. In contrast, a warp is a 32-width vector, i.e. 32 "same operations", and up to 512 diffe
277.
by winwang 2y ago
Yeah, not just OpenCL, but even "newer" standards like WebGPU. I considered making a blog post where I just put the two hello worlds side-by-side and say nothing else. I was severely disappointed after seeing people praise WebGPU
278.
by winwang 2y ago
Isn't the point of OpenCL to be... open? Not only did Intel and AMD not provide enough value, but neither did the community. CUDA... is kind of annoying. And yet, it's the best experience (for GPGPU), as far as I can tell. I feel
279.
by winwang 2y ago
I was thinking more like "Samsung charges $50/year after the typical 3-4 years of updates they normally give."
280.
by winwang 2y ago
Not a user of StarRocks, but I'm curious about your experience with BigQuery and its price-to-performance (as well as its alternatives). I'm building a product with the gist of "BigQuery but cheaper" (~30%-50% expected c
281.
by winwang 2y ago
What if we turned hardware support into a subscription (kind of like JetBrains model I think?) and stopped yearly releases in favor of more interesting releases? I wonder how many resources are used just to make the next iteration a bit shi
282.
by winwang 2y ago
I count the IDE and stuff like LSP as natural extensions of the compiler. For sure I grep (or equivalent) for stuff, but I highly prefer statically typed languages/ecosystems. At the end of the day, I'm here to solve problems, and
283.
by winwang 2y ago
Because the value the item brought you was worth more than the money you exchanged for it. My Bose QC35s were one of the best investments I've ever made, lasting many years and very many loud plane rides. Or would you consider grocerie
284.
by winwang 2y ago
in fact, e = 2, made rather abundantly clear in "finite difference" calculus, and also that in computer science, the "natural" log base 2.
285.
by winwang 2y ago
Check out this subreddit for a decent "source of truth": reddit.com/r/localllama
286.
by winwang 2y ago
Does AMD lack NPUs on mobile still? It should be coming quite soon, apparently 40 TOPs.
287.
by winwang 2y ago
You're describing games of incomplete information (i.e. as opposed to a perfect game like chess). Interestingly, when I look this up, the first thing that pops up is "Bayesian Game".
288.
by winwang 2y ago
I have a hard time grokking your wording, but I'd like to point out that Bayes' Rule is typically "typed" to events (i.e. B is of type "event"). That being said, I don't think this needs to be the case.
289.
by winwang 2y ago
He also proposed using quantum computers to solve "hard" physical systems. https://arxiv.org/pdf/2106.10522 Though, I'm unsure if he were the first to do so in a semi-public manner.
290.
by winwang 2y ago
I mean, if we're talking about "accelerating by modifying the metric tensor" then yeah, that would be pretty sci-fi :) https://en.wikipedia.org/wiki/Metric_tensor_(general_relativ...
291.
by winwang 2y ago
Dang. +20% is still a nice difference, but not sure how they did that. Here's hoping M4 Max can include more tech, but that's copium. 960 GB/s is 3090 level so that's pretty good. I'm curious if the Macbooks right n
292.
by winwang 2y ago
I would want to know if the LPDDR6 rumors are substantiated, i.e. memory bus details. https://forums.macrumors.com/threads/lpddr6-new-beginning-of... If M4 Max could finally break the 400GBps limit of the past few year
293.
by winwang 2y ago
I agree, but isn't that partially because the ads work? If they didn't, ad revenue would stagnate.
294.
by winwang 2y ago
Sure, everything I said likely goes out the window if we're talking on-prem. No idea how I/O works in that setting, and in fact I am completely unfamiliar with workloads which saturate 800 GB/s network, especially since that
295.
by winwang 2y ago
Let's assume a cloud datacenter context, using GCP as an example for real, verifiable numbers. An L4 GPU node (with 4 CPU cores) is approximately the cost of a 16-core VM. So for the price of 64 cores, you could have ~4 L4 nodes (or pe
296.
by winwang 2y ago
If they aren't network/disk bound, then GPUs have significantly higher memory bandwidth (per $). If you're so network bound that your nodes must be <8 cores, then sure. Otherwise, adding a cheap GPU to your nodes would lik
297.
by winwang 2y ago
That's right, and that means most transforms in big data. The fact that the dataset can be distributed at all typically implies that the task is parallel.
298.
by winwang 2y ago
Although "wide" data is touted as a optimization guideline for Nimble, how well does it fare against "normal"(?) data, i.e. with just a few to tens of columns? Also, are there any preliminary benchmarks?
299.
by winwang 2y ago
I'm surprised people don't talk more about sentiment analysis -- or is that mostly solved? Would also be interesting to see more treatises on tranformer(-like) forecasting. Some discussion here: https://www.reddit.com&#
300.
by winwang 2y ago
(just doing roll call here with initals WW)

More ›