shaklee3

8 ms·

61.
by shaklee3 2y ago
That's not true. It does slow things down because it has an atomic access. How slow depends on the platform. unique_ptr does not slow things down.
62.
by shaklee3 2y ago
You don't need to shut down processes on the server. All you have to do is isolate CPU cores and move your workloads onto those cores. That's been a common practice in low latency networking for decades.
63.
by shaklee3 2y ago
You could have used RoCE for reliability over UDP
64.
by shaklee3 2y ago
Sure, but as soon as you put out source you're going to have suggestions/comments/criticisms/PRs/etc.
65.
by shaklee3 2y ago
That was supposed to say he's not watching.
66.
by shaklee3 2y ago
Sean has said he would consider open sourcing it later, but doing so now would defeat the purpose of the project. He makes a lot of progress simply because he's but watching all the issues and prs in GitHub.
67.
by shaklee3 3y ago
Not true. Read Matt Levine's article from 3 days ago
68.
by shaklee3 3y ago
Tesla and Twitter are not doing fine. He's trying to get Tesla to agree to his original compensation bonus that was rejected by the Delaware courts. But that bonus was for bringing the valuation over $650 billion. The valuation has now
69.
by shaklee3 3y ago
A100 (> 5yo GPU) has a 1-bit tensor core engine
70.
by shaklee3 3y ago
Drivers for the H100 are available right on their website
71.
by shaklee3 3y ago
It's extremely uncommon in the US to have induction hobs over 4kW. Even if the outlet could supply it, they're very rare.
72.
by shaklee3 3y ago
And works on any phone and not just iPhone
73.
Early C++ Reflection Feature Lands in EDG(old.reddit.com)
1 points by shaklee3 3y ago|0 comments
74.
by shaklee3 3y ago
Thanks for the excellent summary. I said this in previous stories (not as eloquently) and got an angry mob for not praising spacex. There's a difference between respecting what they're doing for launches and what they're doin
75.
AMD did not use latest Nvidia software on MI300X keynote comparison(forbes.com)
5 points by shaklee3 3y ago|0 comments
76.
FCC affirms its earlier denial of SpaceX's application for $885M in funding(fcc.gov)
3 points by shaklee3 3y ago|2 comments
77.
by shaklee3 3y ago
Google fi works fine on messages and uses rcs.
78.
by shaklee3 3y ago
Their own slides show a 20-30% speedup on attention tasks.
79.
by shaklee3 3y ago
It's transparent, but it has a performance cost. There will be low-level ways to improve locality for certain libraries, but it's nearly impossible for common users to claw that performance loss back.
80.
by shaklee3 3y ago
It also implies that it's harder to program efficiently and is slower on the interconnect.
81.
by shaklee3 3y ago
OP's point is that the hardware specs are 2-3x higher in many places, but all their benchmarks are 20-30% higher. The article mentions this as well. It means AMD couldn't even utilize their own hardware very well at this point.
82.
by shaklee3 3y ago
You can do 100G per core with DPDK with bigger frame sizes
83.
by shaklee3 3y ago
BI is windows only. Frigate is mostly used on Linux.
84.
by shaklee3 3y ago
Not really. Blue Iris has too many features that aren't really needed, and the lack of Linux support makes it a non starter in many cases. Also, the AI features on BI are far worse.
85.
by shaklee3 3y ago
I've been using Frigate for about 6 months. It's significantly better than Blue Iris and zoneminder.
86.
NVPL: High-Performance Mathematical Libraries for ARM(developer.nvidia.com)
4 points by shaklee3 3y ago|0 comments
87.
by shaklee3 3y ago
You can do device graph launch in cuda https://developer.nvidia.com/blog/enabling-dynamic-control-f...
88.
by shaklee3 3y ago
We had a problem like this too at a satellite company. A customer claimed their service was cutting out every day between 3-4pm. There were no service outages, and it only seemed to affect them, but we did see their modem drop consistently
89.
by shaklee3 3y ago
I think they mean idempotent
90.
CMake Raytracer(64.github.io)
19 points by shaklee3 3y ago|3 comments

More ›