Ce que TurboQuant de Google peut et ne peut pas faire face à la flambée des coûts de l'IA - ZDNET La quantification en temps réel de Google pourrait jouer un rôle important dans l'exécution de l'IA locale. Voici pourquoi.

Ce que #TurboQuant de #Google peut et ne peut pas faire face à la flambée des coûts de l' #IA
👉Réduire considérablement la consommation de mémoire de l'IA
👉Rendre l'IA plus accessible en réduisant les coûts d'inférence
www.zdnet.fr/actualites/c...

[Shoper Gamer] Google Research เปิดตัว TurboQuant อัลกอริทึมบีบอัดเวกเตอร์ ลดขนาดข้อมูล โดย โดย

Google Research เปิดตัว TurboQuant อัลกอริทึมบีบอัดเวกเตอร์ ลดขนาดข้อมูล

#ShoperGamer #Google #GoogleResearch #Algorithm #TurboQuant #VectorDB #Feed

Again, to prove the only #AmericanAI company that stands a chance against #ChineseAI, #Google is constantly releasing #LLM advancements like #TurboQuant & realizes the market shift to #LocalAI accelerating quickly with #OpenSource smaller models.
#Gemma4 is all that.
youtu.be/Kaq5Ual2ij8?...

#AI2026 Google's TurboQuant cuts LLM memory use.
#TurboQuant IBM, AMD power efficient AI hardware.
#SmallAI World Bank backs lightweight AI for health and education.
#AI2026 #TurboQuant #SmallAI
View in Timelines

¡La compresión de IA ha dado un salto gigante! Google Research presenta #TurboQuant, reduciendo la caché de los LLMs a 3 bits y acelerando el rendimiento hasta 8x sin perder precisión. Descubre esta revolución aquí: youtu.be/i4Q2KUJYSbQ #IA #MachineLearning #LLM #AI #Deepmind

That's why I tagged it #edgeLLM / #edgeAI.

I expect to see #RISC-V #vectorISA get more sophisticated with #TurboQuant or at least #keyvaluecache, #takum & #1bit, maybe much more sophisticated lookups for new #quintic geode & #adelic approaches relying on #tribonnaci etc.

M could just add that.

youtube.com/watch?v=0fWF...
#1bit #LLM #noGPU #noNPU #justCPU

The prospect of functional #edgeLLM in a #RISC-V instruction extension (or M chip) already may have popped #AIbubble. The #TurboQuant #KVC optimization was less disruptive than this approach to small AIs useful for #privacy or #latency.

RE: https://mastodon.social/@arstechnica/116335949908385054

No TurboQuant?
Would have been the perfect showcase …
Hhm..
#AI #TurboQuant
#Google

Google quietly dropped one of the biggest AI efficiency upgrades of the year.

TurboQuant compresses KV cache memory from 16 bits down to 3 bits.

Same accuracy.

Up to 8x faster inference.

Works on existing models instantly.

Even Cloudflare’s CEO called it Google’s DeepSeek moment.

#TurboQuant

TurboQuant: Ο επαναστατικός αλγόριθμος της Google που μειώνει τις απαιτήσεις μνήμης AI κατά 6 φορές Με τις νέες μεθόδους PolarQuant και QJL, η Google υπόσχεται ταχύτερη απόκριση, λιγότερα «hallucinations» και λύση στη διεθνή έλλειψη μνήμης RAM μέχρι το τέλος της δεκαετίας.

Η Google παρουσιάζει το TurboQuant, έναν αλγόριθμο που συμπιέζει τα AI μοντέλα κατά 6 φορές. Μάθετε πώς το PolarQuant και το QJL μειώνουν το latency και την ανάγκη για RAM. #TurboQuant #RAMShortage #RAMageddon

Hannu Varjoranta (@varjoranta) A 60 GB model in 17 GB. I wrote up what I learned compressing both sides of LLM memory with TurboQuant: same math, completely different failure modes. Best part was watching independent researchers co...

A 60 GB model in 17 GB. I wrote up what I learned compressing both sides of #LLM memory with #TurboQuant: same math, completely different failure modes. Best part was watching independent researchers converge on the same fixes.

substack.com/home/post/p-...

TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate Vector quantization, a problem rooted in Shannon's source coding theory, aims to quantize high-dimensional Euclidean vectors while minimizing distortion in their geometric structure. We propose TurboQ...

#TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate
Amir Zandieh, Majid Daliri, Majid Hadian, Vahab Mirrokni

#Google #AI #Quantization #LLM
arxiv.org/abs/2504.19874

Like myself, #TwoMinutePapers waited to talk about #Google 's #TurboQuant. Anyone having worked w/ compression algorithms like Reed-Solomon or others used for data compression knows there are trade-offs. Especially on vectorized matrices vs serialized data formatting. #AI
youtu.be/7YVrb3-ABYE?...

A 30B Model in 17 GB of GPU Memory — Varjosoft Zero-calibration weight compression for any model on HuggingFace. Load a BF16 checkpoint, compress in 4 seconds, run. Qwen3-30B fits on a consumer GPU.

Compressed Qwen3-30B from 60 GB to 17 GB. No calibration, 4 seconds at load time. Quality holds, speed doesn't (yet).

Same #TurboQuant math as my KV cache work, now extended to weights. Honest writeup: varjosoft.com/weight-compression

Nvidia shifted its Vera Rubin architecture from a complex 4-die package to a 2-die setup, prompting an initial 4.5% to 10% market drop for memory makers.

#Micron #HBM #DRAM #Semiconductors #TurboQuant

RAM Fiyatlarında Şok Düşüş: Google Duyurdu, DDR5 Fiyatları Çakıldı! - oyunda.org RAM Fiyatlarında Şok Düşüş: Google Duyurdu, DDR5 Fiyatları Çakıldı! Oyun Haberleri - oyunda.org

RAM Fiyatlarında Şok Düşüş: Google Duyurdu, DDR5 Fiyatları Çakıldı!
#Google 'ın bellek kullanımını azaltan #TurboQuant algoritması sonrası Çin'de DDR5 #RAM fiyatları hızla düşmeye başladı. Yapay zeka kaynaklı stok krizinin ardından gelen bu düşüş,
oyunda.org/43c3f60

Whoa AIs Nelly !
#AIRevolution releases a month ender one and a half hour show!
Everything from the latest AI level ups: China’s 'Human AI' #Robots, #Google3.1 Update #TurboQuant, #OpenClaw Robot & More yet #AINews.

youtu.be/uV2qLhnc85k?...

KOSPI Slides Over 4% as Middle East Tensions Escalate The KOSPI plunged more than 1000 points in a month, approaching the 5000 level as foreign investors dumped over 34 trillion won and nearly 1000 trillion won in

The KOSPI plunged more than 1000 points in a month, approaching the 5000 level as foreign investors dumped over 34 trillion won and nearly 1000 trillion won in market value evaporated.
#KOSPI #MiddleEast #Samsung #SK #Google #TurboQuant #AI #itchosun #chosun #SouthKorea #Korea #NewsUpdate

LLM inference costs got you down? Google's TurboQuant slashes KV cache memory from 300KB to 69KB per token, boosting performance up to 8x with zero accuracy loss!

thepixelspulse.com/posts/llm-kv-cache-probl...

#llm #kvcache #turboquant

#TurboQuant : Comment #Google réduit fortement la taille des modèles #IA sans aucune perte de précision grâce à un algorithme de compression
👉Une méthode qui combine deux algorithmes en un
www.usine-digitale.fr/big-tech/goo...

Google TurboQuant: The Algorithm That Could Crash Nvidia's Memory Business Mar 30, 2026

Google #TurboQuant: The Algorithm That Could Crash Nvidia's Memory Business with a compression algorithm that shrinks AI memory usage by 6x with zero accuracy loss. | Natural 20 buff.ly/vY4HOlI

Google unveils TurboQuant, a new AI memory compression algorithm — and yes, the internet is calling it 'Pied Piper' | TechCrunch Google’s TurboQuant has the internet joking about Pied Piper from HBO's "Silicon Valley." The compression algorithm promises to shrink AI’s “working memory” by up to 6x, but it’s still just a lab expe...

It's never a good thing when the satirical cautionary tale literally merges with stock market press releases.

techcrunch.com/2026/03/25/g...

#TurboQuant
#AIbubble

Remember when #Zelensky TV comedy became a political party? How did that work out?

#TotalCollapseOfTheWest
#LifeImitatesArt
#yrugae

TurboQuant Visual Explainer As AI models get smarter, they are becoming massive memory hogs.

TurboQuant Visual Explainer

whyaiman.substack.com/p/turboquant...

#TurboQuant #KVCache #LLMs #AITraining #VisualExplainer #LocalAI

Glavni okidač za ovu promenu je nova tehnologija #TurboQuant koju je predstavio Google 🚀

benchmark.rs/vesti/biznis...

Signed Cast, Hidden Dagger: Google Smiles at SK While Cutting Memory Demand In the global artificial intelligence (AI) ecosystem, there are no eternal allies. This reality has been proven once again as Google unveiled new technology th

In the global artificial intelligence (AI) ecosystem, there are no eternal allies.
#SK #SKHynix #Google #TurboQuant #AI #HBM #Samsung #SamsungElectronics #TPU #software #hardware #technology #it #itchosun #chosun #SouthKorea #Korea #NewsUpdate

VaultSage Apps & Extensions - All Install Links Discover VaultSage apps for Slack, Gmail, NotebookLM, NAVER Works, DocSend, YouTube, iOS, macOS, and MCP. Install the right app for your workflow.

Available via MCP server, API, extensions, apps, Slack bot, and website.
nurie.ai/apps

Images by #MediaSage - media.nurie.ai

#VaultSage #TurboQuant #AI #Efficiency #NURIEAI #NURIE

Western Digital Slides 8% After Google TurboQuant Debut Western Digital shares fell ~8% on Mar 29, 2026 after Google's TurboQuant debut; hyperscaler procurement timing and HDD vs SSD mix now key to near-term revenue.

Western Digital Slides 8% After Google TurboQuant Debut: Western Digital shares fell ~8% on Mar 29, 2026 after Google's TurboQuant debut; hyperscaler procurement timing and HDD vs SSD mix now key to near-term revenue. 👈 Read full analysis #WesternDigital #TurboQuant #Google #HDD #SSD

GitHub - peva3/turboquant-h2o-streamingllm Contribute to peva3/turboquant-h2o-streamingllm development by creating an account on GitHub.

Holy shit! I just got Turboquant working. They weren't fucking kidding. 4x context space with the same memory usage!
I used: github.com/peva3/turboq...

#llm #ai #turboquant

Google's #TurboQuant is a compression mechanism for Key-Value Cache — the 'memory' of an #LLM.

It is derived from #PolarQuant which converts KV vectors from Cartesian to polar co-ordinates.

Early tests (admittedly by Google) show 6x reduction in memory usage.

This seems quite important for #AI.