تغريدة مثبتة
"only a small number of users care about this" is a bad argument when those users are the people everyone else asks before spending money
English
neggles (Andi H)
56.7K posts
neggles (Andi H)
@neggles
expect nothing and appreciate everything. opinions are solely my own and do not reflect the views or opinions of my employer. @neggl.es on bsky
Melbourne, AU انضم Haziran 2007
1K يتبع811 المتابعون
@IanCutress but ian don't you understand??? the TAM is the entire global PC market, duh!!!!!!
😑
English
Please tell me how a 50W+ chip (or even 30W) in a $2k+ Windows Surface addresses a $300b TAM. Please tell me.
Athena@Athena_Slays
We decoded it. The coordinates (25.0528, 121.5990) are Taipei. Computex keynote is Sunday. "A new era of PC" = the N1/N1X. NVIDIA's first ARM-based laptop chip, built with MediaTek. Competing with Apple Silicon and Qualcomm. We wrote about this two days ago - the "surprise product" Jensen teased. Bullish for $NVDA (new $300B TAM), $ARM (CPU license), $TSM (3nm fab).
English
i wonder if i'm on the same flight to taipei as @HardwareUnboxed again
English
<obligatory hermes-agent shillpost>
no but fr hermes is based and you can use your ChatGPT account with it too! (the nous tool gateway continues to be killer, though)
Tibo@thsottiaux
A little secret. About 5% of our production traffic is on the Pi harness, about another 5% is on OpenCode. Reminder you can use your ChatGPT account in a flourishing set of other tools. We’ll continue to make Codex awesome, but you have options.
English
love is a distributed system (everyone is lying, briefly) suno.com/s/Wx3JkFcTiMTv…
English
@lzchen_ut need to have a word with you, please. you should be able to DM me or i'm same username on discord.
English
neggles (Andi H) أُعيد تغريده
Rumors about the new Gemini Flash coming in. And holy, if true then big:
92% of GPT-5.5’s coding and reasoning performance, reportedly at 15–20x lower inference cost. And the latency? Sub-200ms for most queries.
That would be nuts. no joke.
Bindu Reddy@bindureddy
Gemini 3.2 Flash - Capitalizing on DeepMind's clever distillation techniques... Rumors are that benchmarks show it's hitting 92% of GPT 5.5's performance on coding and reasoning tasks while being 15-20x cheaper on inference costs. The latency improvements are insane - sub-200ms for most queries. Google's distillation + sparsity techniques are paying off massively. They've essentially compressed a frontier model into a flash variant without the usual quality cliff.
English
neggles (Andi H) أُعيد تغريده
Hyperthread is the Intel branding from the Intel region of France.
Otherwise it's just sparking Simultanous Multi-Threading.
Qubitium@qubitium
320 cores of AMD Turin, 640 hyperthreads.
English
@FelixCLC_ unfortunately the smartfusion2 SoC-FPGA they use for secure boot enforcement and inline flash encryption/decryption on most vaguely recent meraki gear is a much tougher nut to crack
English
i have to wonder if a *lot* of older enterprise gear with license lockout might be at risk of being cracked in the next months
English
i miss when security breaches were caused by actually skilled threat actors rather than fuckwits with a copy of claude code
English
@viccpoes why is this even a question? of course you should, everyone should, open source ALL THE THINGS
but is there an actual chance you will or are you just baiting ;_;
English
English
Apple’s most critical supplier Foxconn has allegedly been breached
>Foxconn Wisconsin data center plant is currently suffering a cyber attack
>8TB of data compromised of 11m files
>Nitrogen Ransomware samples include Google and Intel sample topology specs
English
@FelixCLC_ brb reverse engineering RTL from a CnC microarchitecture block diagram wcgw
English
@FelixCLC_ oh hey neat, proof they were out of inference compute capacity
English
tl;dr: Qualcomm do not have an excuse here.
they have the resources, engineers, and upstream experience. other non-x86 vendors with far less leverage have already shown it can be done.
they could ship a sane client platform compatibility story.
they are choosing not to.
English
@FelixCLC_ is there a reason that this couldn't be implemented with a per-cluster accelerator like SME2 or Apple's AMX is? feels like it could be
English
I have major concerns of highly coupled requirements between AMX and AVX10 registers caused by ACE meaning you'd want/have to slow scalar to AVX10 code paths down
English
@sahajsarup @qubitium @JDorbal1989 @Qualcomm how is Phytium any different to Qualcomm here?
aarch64, custom cores, custom SoCs, multiple OEMs making motherboards and laptops and the like, reference firmware package provided that Byosoft et. al. apply mods to for those OEMs
it's literally the same setup as qualcomm???
English
@neggles @qubitium @JDorbal1989 @Qualcomm All the vendors you mentioned are single or low count. Not many players in Arm server, easy to standardize and it works well.
Phytium, loongarch whatever else, single vendor the entire ecosystem is under their control, easy to enforce and implement standards.
(1/2)
English
ASUS PR: "Upon confirmation with our chip supplier, @Qualcomm, there are currently no plans to support Linux systems on the Snapdragon X2 Elite platform."
QC assured us that SDX2E would have day-0 linux support, multiple times.
so that was a %#&$ing lie
English