Search
45 results for “AMDmi3”
-
@AMDmi3 I don't know. I am not using #Nix on a daily basis. Nix is only a package manager, so once it builds and installs, it is good to go (more or less). #nixpkgs is going to be involve a lot of actual work but that is going to happen independently from the #FreeBSD ports.
I was wondering if Nix could use FreeBSD ports as a source of information on how to build software on FreeBSD (I've heard that it has such possibilities for #Hackage but I have to read more about it).
-
@AMDmi3 Great! I'll add it to the workshop todo list.
-
LLM Inference Takes Aim at Production Realities
New disaggregated LLM serving is faster and cheaper than old aggregated methods for businesses using AI. Tests show better performance.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/
-
New tests show a disaggregated LLM serving method is 2x faster than older methods using fewer resources. This means AI services will work better.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/ -
LLM Inference Takes Aim at Production Realities
New disaggregated LLM serving is faster and cheaper than old aggregated methods for businesses using AI. Tests show better performance.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/
-
LLM Inference Takes Aim at Production Realities
New disaggregated LLM serving is faster and cheaper than old aggregated methods for businesses using AI. Tests show better performance.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/
-
LLM Inference Takes Aim at Production Realities
New disaggregated LLM serving is faster and cheaper than old aggregated methods for businesses using AI. Tests show better performance.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/
-
LLM Inference Takes Aim at Production Realities
New disaggregated LLM serving is faster and cheaper than old aggregated methods for businesses using AI. Tests show better performance.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/
-
New tests show a disaggregated LLM serving method is 2x faster than older methods using fewer resources. This means AI services will work better.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/ -
New tests show a disaggregated LLM serving method is 2x faster than older methods using fewer resources. This means AI services will work better.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/ -
New tests show a disaggregated LLM serving method is 2x faster than older methods using fewer resources. This means AI services will work better.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/ -
New tests show a disaggregated LLM serving method is 2x faster than older methods using fewer resources. This means AI services will work better.
#LLMServing, #AIefficiency, #OracleCloud, #AMDMI300X, #TechNews
https://newsletter.tf/disaggregated-llm-serving-faster-than-aggregated/ -
Trump Criticizes Taiwan Semiconductor Industry After Court Ruling
US President Donald Trump criticized Taiwan’s semiconductor industry after the US Supreme Court ruled that …
#Conflict #Conflicts #War #AMDMI325X #AsiaBusinessOutlook #China #ChinaTaiwanConflicts #InternationalEmergencyEconomicPowersAct #NVIDIAH200 #Semiconductortariffs #Taiwan #Taiwanexports #Taiwansemiconductorindustry #trump #TSMCArizona #uschinatrade #USSupremeCourtruling #USTaiwanrelations
https://www.europesays.com/2799222/ -
✅ Azure introducing Nvidia ND H200 v5 Tensor Core GPUs for larger AI models & faster inferences (aka.ms/nvidiah200)
✅ Azure NCCv5 Confidential VMs using Nvidia H200 Tensor Core GPUs (aka.ms/accblog) for highly sensitive end-to-end data protection
✅ AMD MI300X accelerates Azure VMs for large gen AI workloads (aka.ms/amdmi300x) 192GB of HS mem + 5.2TB/sec of bandwidth delivers the fastest memory speed/capacity
Satya Nadella's Keynote speech at #Microsoft #Ignite2023:
https://youtu.be/FZhbJZEgKQ4?feature=shared -
Sometimes a great feature is only a few hours of relaxed work away.
-
It looks like #rust #axum release 0.9 is cooking. I'm really looking forward to it as it will feature support for routes with static suffixes prefixes like `/images/{id}.png`, IMO an absolute must have (or, rather, router without it would just be crippled). Many thanks for Ibraheem Ahmed for implementing it in matchit crate and David Mládek for adding support for it on axum side, while I just stood up to update it to the state acceptable for merging.
-
Take related to #rust ai #ethics case.
Projects without no-AI policy: AI may be involved in writing the code.
Projects with no-AI policy: AI still may be involved in writing the code. AI is also involved on contributors side to ensure that written code does not look like AI written code to get changes accepted. AI is involved on upstream side (dare to ask me why) to verify that contributed code is not AI written.
-
Take related to #rust ai #ethics case.
Bram Moolenaar, creator of vim text editor, had supported AIDS victims in Uganda (for which he has my deepest respect). As far as I know, no other text editor author does the same. Are they ignoring ethics probem? Should we have policies of now allowing these editors?
-
So I've finally polished and published https://crates.io/crates/libversion crate, a pure #rust implementation of version comparison algorithm #Repology depends on.
You may or may not be a fan of rewriting anything in rust, but when someone submits you rust bindings of your own C library, it's absolutely 100% the valid case.
In fact though, it was the first thing I've rewritten in rust when I started learning it 2 years ago, though it took more time to gain enough XP to implement and publish it cleanly.
-
It looks like #axum is finally going to get proper route matching with prefixes and suffixes support (e.g. `/images/{foo}.jpg`, while before only `/images/{foo}` was supported).
https://github.com/tokio-rs/axum/pull/3702 -
It looks like #axum is finally going to get proper route matching with prefixes and suffixes support (e.g. `/images/{foo}.jpg`, while before only `/images/{foo}` was supported).
https://github.com/tokio-rs/axum/pull/3702 -
#repology service may be intermittent due to DDoS attack on the hoster. Status and plans: https://github.com/repology/repology-rs/issues/278
-
#repology's linkchecker rewritten in rust and now back live (wasn't working since Aug 2024). Keep an eye for new broken upstream and download URLs in your repositories.
-
On the downside, I'll have to maintain ruleset (merging related projects, splitting unrelated ones, and blacklisting incorrect versions) in whole another ecosystem, and one which lacks good package naming (names are long unicode strings which need normalization) and versioning (`v` prefix and random suffixes all over the place) traditions.
So for now I'm hesitant about making Android repos first class citizens in #repology.
-
#Repology only has marginal support for F-Droid, only seeing a few handpicked packages for software which is also available in Linux:
https://repology.org/projects/?inrepo=fdroid
I've recently had a few PRs which improve F-Droid support and add #IzzyOnDroid, allowing full-fledged version comparison within Android ecosystem.
I wonder if any #fdroid maintainers or @IzzyOnDroid would be interested in that.
-
So after doing a few small projects in #rust, I've finally started rewriting #repology backend (currently in #python) in Rust.
First discovery - what I naively though of as a database-bound workload, e.g. where database RTT (for small queries) or response generation time (for complex queries) dominates, are still 5-10x faster (in terms of both max RPS and latency) in Rust!
Also rust process takes 3-10x less memory compared to single uwsgi worker, which there are 10-20 of.
-
If you work with #NVD, note that CPEs in NVD are subject to change. I've just discovered that a lot of CPE bindings in #Repology are outdated due to changes in CPE vendors and products (aiohttp_project:aiohttp -> aiohttp:aiohttp, stlport:stlport → stlport_project:stlport, soundexchange:sound_exchange → sox_project:sox etc). If you maintain CPE bindings for any purpose, you should revisit and update them.
-
New repositories support added recently to #repology:
- #chromebrew (package manager for Chrome OS)
- #opam (#OCaml package manager; this should be of great help to highlight outdated packages of ocaml modules in all supported repositories)
- @serpentosIn other news, #guix is blocking access from Russia again, so it's not updated in Repology.
-
New repositories support added recently to #repology:
- #chromebrew (package manager for Chrome OS)
- #opam (#OCaml package manager; this should be of great help to highlight outdated packages of ocaml modules in all supported repositories)
- @serpentosIn other news, #guix is blocking access from Russia again, so it's not updated in Repology.
-
New repositories support added recently to #repology:
- #chromebrew (package manager for Chrome OS)
- #opam (#OCaml package manager; this should be of great help to highlight outdated packages of ocaml modules in all supported repositories)
- @serpentosIn other news, #guix is blocking access from Russia again, so it's not updated in Repology.