For You got quite a bit faster and more efficient in October - the average latency dropped from 100ms to 27ms.
October 30, 2026 at 1:00 PM
For You got quite a bit faster and more efficient in October - the average latency dropped from 100ms to 27ms.
November 2, 2026 at 11:50 AM
I'm done hosting forgejo publicly man
someone was hammering it and driving the 95th percentile of request latency to 10s for all my services T.T
someone was hammering it and driving the 95th percentile of request latency to 10s for all my services T.T
November 3, 2026 at 9:35 PM
I'm done hosting forgejo publicly man
someone was hammering it and driving the 95th percentile of request latency to 10s for all my services T.T
someone was hammering it and driving the 95th percentile of request latency to 10s for all my services T.T
Best full duplex low latency voice ai tts/stt/local model combos I’ve found after testing a bunch are github.com/speakrail/sp... running on 3090 (uses tuned Gemma 12b + call to deepseek) or unmute.sh (swap gptoss for deepseek 4.1 flash)
GitHub - speakrail/speakrail: Speakrail - a low-latency fully-local voice assistant that runs on a single RTX 4090
Speakrail - a low-latency fully-local voice assistant that runs on a single RTX 4090 - speakrail/speakrail
github.com
November 2, 2026 at 1:45 PM
Best full duplex low latency voice ai tts/stt/local model combos I’ve found after testing a bunch are github.com/speakrail/sp... running on 3090 (uses tuned Gemma 12b + call to deepseek) or unmute.sh (swap gptoss for deepseek 4.1 flash)
With the way the FGC has input latency as their top priority and streamers wanting big visuals:
You’d think one would have more self-preservation on your hobbies as due to AI, the former will reach an event horizon of nevering been salvaged with all corps pushing towards streaming/bad DRM (cont.)
You’d think one would have more self-preservation on your hobbies as due to AI, the former will reach an event horizon of nevering been salvaged with all corps pushing towards streaming/bad DRM (cont.)
November 2, 2026 at 1:07 PM
With the way the FGC has input latency as their top priority and streamers wanting big visuals:
You’d think one would have more self-preservation on your hobbies as due to AI, the former will reach an event horizon of nevering been salvaged with all corps pushing towards streaming/bad DRM (cont.)
You’d think one would have more self-preservation on your hobbies as due to AI, the former will reach an event horizon of nevering been salvaged with all corps pushing towards streaming/bad DRM (cont.)
yeah lmao
it's so bad on anything that isn't just a simple gtk app the latency is horrible
it's so bad on anything that isn't just a simple gtk app the latency is horrible
November 2, 2026 at 5:27 AM
yeah lmao
it's so bad on anything that isn't just a simple gtk app the latency is horrible
it's so bad on anything that isn't just a simple gtk app the latency is horrible
Oh spindle also runs off sqlite. It does very minimal record keeping and CI is once again, very latency lenient.
November 2, 2026 at 1:09 PM
Oh spindle also runs off sqlite. It does very minimal record keeping and CI is once again, very latency lenient.
October 31, 2026 at 8:34 AM
And cosmic ray protection, and network latency, and…
It’s not gonna happen. Ever.
It’s not gonna happen. Ever.
November 2, 2026 at 12:36 PM
And cosmic ray protection, and network latency, and…
It’s not gonna happen. Ever.
It’s not gonna happen. Ever.
latency is in top right hand corner
(excuse the fucked hdr, just wanted to record this quickly)
(excuse the fucked hdr, just wanted to record this quickly)
October 31, 2026 at 12:25 AM
latency is in top right hand corner
(excuse the fucked hdr, just wanted to record this quickly)
(excuse the fucked hdr, just wanted to record this quickly)
It's election day! That means once again watching Steve Kornacki deal with this horrible touch screen latency.
November 4, 2026 at 12:13 AM
It's election day! That means once again watching Steve Kornacki deal with this horrible touch screen latency.
overall performance of a system is dictated by the latency between its parts
November 3, 2026 at 9:18 PM
overall performance of a system is dictated by the latency between its parts
luckily, you can just wish* latency away
November 3, 2026 at 9:28 PM
luckily, you can just wish* latency away
It's the Memory Latency, Stupid https://pivotlake.io/blog/its-the-memory-latency-stupid/ (https://news.ycombinator.com/item?id=50318720)
It's the Memory Latency, Stupid
pivotlake.io
October 31, 2026 at 9:24 PM
It's the Memory Latency, Stupid https://pivotlake.io/blog/its-the-memory-latency-stupid/ (https://news.ycombinator.com/item?id=50318720)
Negative Latency is back on the Menu Boys!
November 3, 2026 at 5:46 PM
Negative Latency is back on the Menu Boys!
Its been like 6 months to get to this point. This is so much more complicated than it seems. From latency calculation, sound isolation, event streams, to who wins and how to decide in a tempo dispute etc.
November 2, 2026 at 6:46 PM
Its been like 6 months to get to this point. This is so much more complicated than it seems. From latency calculation, sound isolation, event streams, to who wins and how to decide in a tempo dispute etc.
how does write latency feel, subjectively? is it noticeable in apps? I assume R2 has ~S3 latency
October 29, 2026 at 3:06 PM
how does write latency feel, subjectively? is it noticeable in apps? I assume R2 has ~S3 latency
p99 latency drop in one of our agentic systems after replacing LLM-as-a-judge with Jev + deterministic evaluations.
Fantastic!
Fantastic!
November 3, 2026 at 3:37 PM
p99 latency drop in one of our agentic systems after replacing LLM-as-a-judge with Jev + deterministic evaluations.
Fantastic!
Fantastic!
iirc the official Game Boy Player is typically avoided for speedruns because of pretty significant input delay. Some unofficial ones have a "speedrun mode" that goes for zero latency
November 1, 2026 at 6:44 PM
iirc the official Game Boy Player is typically avoided for speedruns because of pretty significant input delay. Some unofficial ones have a "speedrun mode" that goes for zero latency
the latency on Kornacki’s giant monitor input is awful
November 4, 2026 at 12:11 AM
the latency on Kornacki’s giant monitor input is awful
Questions for other developers:
How important is optimisation for you? When do you start optimising?
How do choose your latency requirements?
I'm building a multi-platform DAW so the most important latency for me is the DSP latency.
I choose 48kHz at 1024 frames which gives 21.33ms of room.
How important is optimisation for you? When do you start optimising?
How do choose your latency requirements?
I'm building a multi-platform DAW so the most important latency for me is the DSP latency.
I choose 48kHz at 1024 frames which gives 21.33ms of room.
November 3, 2026 at 1:02 PM
Questions for other developers:
How important is optimisation for you? When do you start optimising?
How do choose your latency requirements?
I'm building a multi-platform DAW so the most important latency for me is the DSP latency.
I choose 48kHz at 1024 frames which gives 21.33ms of room.
How important is optimisation for you? When do you start optimising?
How do choose your latency requirements?
I'm building a multi-platform DAW so the most important latency for me is the DSP latency.
I choose 48kHz at 1024 frames which gives 21.33ms of room.
It's the Memory Latency, Stupid | Discussion
It's the memory latency, stupid · Pivot blog
When your database CPU is at 100%, what is it actually doing? Memory stalls, CPU caches, and how radix partitioning makes GROUP BY faster.
pivotlake.io
October 31, 2026 at 8:20 PM
It's the Memory Latency, Stupid | Discussion
No scale? No problem.
But off the top of my head I've run into: terrible concurrency w/per process connections, garbage and VACUUM overhead/performance, crappy heuristics query planner, WAL replication but DIY failover, no horizontal write scaling, write contention, jittery query latency, etc.
But off the top of my head I've run into: terrible concurrency w/per process connections, garbage and VACUUM overhead/performance, crappy heuristics query planner, WAL replication but DIY failover, no horizontal write scaling, write contention, jittery query latency, etc.
there is no reason to make a new database. postgres can do it. if you think postgres can't do it, you just haven't found the right way to enable it in pg_settings yet
November 2, 2026 at 7:39 PM
No scale? No problem.
But off the top of my head I've run into: terrible concurrency w/per process connections, garbage and VACUUM overhead/performance, crappy heuristics query planner, WAL replication but DIY failover, no horizontal write scaling, write contention, jittery query latency, etc.
But off the top of my head I've run into: terrible concurrency w/per process connections, garbage and VACUUM overhead/performance, crappy heuristics query planner, WAL replication but DIY failover, no horizontal write scaling, write contention, jittery query latency, etc.
my biggest concern was commit latency but it sounds like it's not a big deal. writes from the client are generally eating a bunch of network latency regardless
apps should be local-first anyway
apps should be local-first anyway
October 29, 2026 at 4:15 PM
my biggest concern was commit latency but it sounds like it's not a big deal. writes from the client are generally eating a bunch of network latency regardless
apps should be local-first anyway
apps should be local-first anyway
A little bummed, but tried multistreaming tonight with @altstream.online and my latency was terrible on there vs things being fine with the Twitch stream. Still going to try - maybe my network was just being butts tonight? No idea.
November 3, 2026 at 4:16 AM
A little bummed, but tried multistreaming tonight with @altstream.online and my latency was terrible on there vs things being fine with the Twitch stream. Still going to try - maybe my network was just being butts tonight? No idea.