winwang

19 ms·

241.
by winwang 2y ago
Idea of Flipper Zero was to lower our guard and trust their company with security, while actually shipping hardware with backdoors -- which the Flipper Zero will, for some reason, not be able to detect.
242.
by winwang 2y ago
#1 does feel a bit much but I don't mind getting a kick out of titles/situations which could be interpreted immaturely.
243.
by winwang 2y ago
Why would that be true? Certainly there are algorithms (or portions of them) in which warps can just read whichever values exist in shared mem at the time, no need to sync. And I think we were mostly talking about global memory?
244.
by winwang 2y ago
Agreed. Kagi even works well with Reddit. Great that I can just fall back to google easily too.
245.
by winwang 2y ago
Typically WFH would expect you (or your boss) to have availability similar to being in person, though?
246.
by winwang 2y ago
Is this really true in general? I'd expect it to be true for highly homogenous blocks, but I'd also expect that kernels where the warps are "desynced" in memory operations to do just fine without having 3-4 blocks per SM
247.
Can GenAI Improve Developer Productivity?(resources.uplevelteam.com)
1 points by winwang 2y ago|1 comments
248.
by winwang 2y ago
Here's a blog post for those who don't want to grab the PDF: https://shenisha.substack.com/p/are-ai-coding-assistants-rea... Unfortunately, the PDF itself doesn't say much, so I'd take it with half
249.
by winwang 2y ago
Do you have other examples of interesting workloads? Benchmarking is difficult, lol.
250.
by winwang 2y ago
Are there "batch"-esque workloads where we want extreme throughput, but can tolerate a large latency (10-100 micros)?
251.
by winwang 2y ago
Would be really interesting to see a similar study but with skiplists! I'd imagine it would be slower than hashmaps for many of the exact same reasons outlined in the article, but numbers would be cool.
252.
by winwang 2y ago
Yep -- and more generally, sequential access > random access, just a bit less so in memory than on SSD, not even mentioning spinning disk.
253.
by winwang 2y ago
Consider an unchanging singly-linked list as a universe. Time (or perhaps "the passage of time") in the intuitive sense does not exist for this universe, but it does exist in the sense that there is a global ordering of values. Bu
254.
by winwang 2y ago
But 12:30am is 30 past twelve!
255.
by winwang 2y ago
Not just guts -- the will to gather real feedback and constantly improve. Beautiful!
256.
by winwang 2y ago
Intuitively (subjectively) why this argument doesn't work: if you switch to the path in the lower bunk, you can switch to the upper bunk "at any moment", regardless of if there is a path which exists on the lower bunk. In thi
257.
by winwang 2y ago
Is the contrapositive of this "philsophers (are those who) do stuff which don't work"? :P
258.
by winwang 2y ago
"Computer" in (simplified) Chinese is still "电脑" -- literally "electric/electronic brain".
259.
by winwang 2y ago
Surprising. DRAM (and more importantly high-bandwidth DRAM) seems to be scaling significantly better than SRAM -- and I'm not sure if that could be seriously expected to shift.
260.
by winwang 2y ago
imo, the way you first do something + your first responses to serious criticism are the most telling regarding character. because afterwards, you can just play the act of "we're sorry because the backlash was really big and we hav
261.
by winwang 2y ago
I disagree, here's a line which rubs me the wrong way: "We tried to be transparent about what we’d done as much as possible since the beginning of our journey, but that wasn’t good enough." Couple this to "talking about
262.
by winwang 2y ago
Isn't this confounding legality and morality? But I do agree with the idea of trying to establish a norm somehow, since legality is a decent proxy.
263.
by winwang 2y ago
Mostly tangential, but in mathematics, there are sometimes "canonical" objects, typically a 'natural' viewpoint which is proven to be unique and sometimes also 'universal' in a sense. So, in many cases, you can
264.
by winwang 2y ago
Hey. It could count to 1, which is something.
265.
by winwang 2y ago
As far as I understand, the main issue for LLM inference is memory bandwidth and capacity. Tensor cores are already an ASIC for matmul, and they idle half the time waiting on memory.
266.
Unified Communication X(github.com)
3 points by winwang 2y ago|1 comments
267.
by winwang 2y ago
Saw the post about Tesla Transport Protocol ( https://news.ycombinator.com/item?id=41621680 ) and thought I'd share this great project (seems to be BSD3 license). Those interested in a design diagram can find one here:
268.
by winwang 2y ago
I mean, you can definitely do CPU-like things on the GPU. Just... why would you make a bandwidth-optimized processor your primary processor, when most programs rely on the low-latency single-thread perf? In the case of parallelism, they are
269.
by winwang 2y ago
I mean, you can definitely do CPU-like things on the GPU. Just... why would you make a bandwidth-optimized processor your primary processor, when most programs rely on the low-latency single-thread perf? In the case of parallelism, they are
270.
by winwang 2y ago
I have a hard time believing that dealing with a singular ISA for two different compute cases would be an "obviously better" solution. I don't doubt that it's plausible/possible, especially if RISC-V + extensions ar

More ›