한국에서 어떻게 중국으로 수출규제품인 HBM이 유출되는지 추적한 기사. 간단하게 요약하면, AI 칩 수출 규제가 RAM 성능이 아니라 GPU 성능으로 정해지는 헛점을 유용해서 '아무도 안 쓸' GPU에 최신 HBM을 붙여서 3국을 걸쳐 빼돌린 뒤, 메모리만 분리해낸다고. www.theinformation.com/articles/sm...
How Smugglers Get Expensive Memory Chips to China from South Korea
For an illustration of how porous U.S. restrictions on the shipment of AI chips to China can be, consider a technique that has begun to catch on among smugglers involving a key type of memory chip. While U.S. export regulations ban the shipment of high-bandwidth memory chips on their own to ...
www.theinformation.com
November 3, 2026 at 3:26 AM
한국에서 어떻게 중국으로 수출규제품인 HBM이 유출되는지 추적한 기사. 간단하게 요약하면, AI 칩 수출 규제가 RAM 성능이 아니라 GPU 성능으로 정해지는 헛점을 유용해서 '아무도 안 쓸' GPU에 최신 HBM을 붙여서 3국을 걸쳐 빼돌린 뒤, 메모리만 분리해낸다고. www.theinformation.com/articles/sm...
its a race to the bottom to find new vulnerabilities in ever broader sectors. it was way harder and way more expensive before AI. and before that, there wasnt really a way to get your ransom. well, thanks to all these fine wastes of GPU cycles out there, the evil work is optimized!
November 3, 2026 at 3:25 AM
its a race to the bottom to find new vulnerabilities in ever broader sectors. it was way harder and way more expensive before AI. and before that, there wasnt really a way to get your ransom. well, thanks to all these fine wastes of GPU cycles out there, the evil work is optimized!
12k出ないのがまともに動かせる性能とケーブル規格を備えたGPUが存在しないから
→多分最低5年は5090超えるパンピーに買えるGPU出ない
絶望
→多分最低5年は5090超えるパンピーに買えるGPU出ない
絶望
November 3, 2026 at 3:18 AM
12k出ないのがまともに動かせる性能とケーブル規格を備えたGPUが存在しないから
→多分最低5年は5090超えるパンピーに買えるGPU出ない
絶望
→多分最低5年は5090超えるパンピーに買えるGPU出ない
絶望
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
xenospectrum.com
November 3, 2026 at 3:17 AM
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
xenospectrum.com
November 3, 2026 at 3:17 AM
NVIDIA製GPUの価格高騰が止まらない、RTX 5090は約148万円相当に メモリ不足だけでは説明できない異常な値上がり
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
https://xenospectrum.com/gpu-prices-memory-cost-geforce/
Rick Stevens returns to GCACC! 🧬 Join our exclusive follow-up briefing on the DOE's Genesis Mission. We’re diving into the 100k-GPU Solstice & Equinox systems, the American Science Cloud, and autonomous labs. Register: https://goo.gle/4yJN45e ☁️🔬
What’s new with Genesis? Rick Stevens returns for an exclusive GCACC briefing
google.smh.re
November 3, 2026 at 3:10 AM
Rick Stevens returns to GCACC! 🧬 Join our exclusive follow-up briefing on the DOE's Genesis Mission. We’re diving into the 100k-GPU Solstice & Equinox systems, the American Science Cloud, and autonomous labs. Register: https://goo.gle/4yJN45e ☁️🔬
At long last I've hit the dreaded point where I need to upgrade some pc parts before I can play any shiny brand-new games.
-Went looking at RAM and GPU prices-
uh. Well, good thing I like playing old games I guess.
-Went looking at RAM and GPU prices-
uh. Well, good thing I like playing old games I guess.
November 3, 2026 at 3:04 AM
At long last I've hit the dreaded point where I need to upgrade some pc parts before I can play any shiny brand-new games.
-Went looking at RAM and GPU prices-
uh. Well, good thing I like playing old games I guess.
-Went looking at RAM and GPU prices-
uh. Well, good thing I like playing old games I guess.
V100 SXM2 pinout and NVLink reference
https://github.com/pigeon-labs-stack/v100-sxm2-reference
[comments] [12 points]
https://github.com/pigeon-labs-stack/v100-sxm2-reference
[comments] [12 points]
GitHub - pigeon-labs-stack/v100-sxm2-reference: Curated V100 SXM2 pin maps, two- and four-GPU NVLink connection plans, and connector-only KiCad resources toward an open-source carrier board.
Curated V100 SXM2 pin maps, two- and four-GPU NVLink connection plans, and connector-only KiCad resources toward an open-source carrier board. - pigeon-labs-stack/v100-sxm2-reference
github.com
November 3, 2026 at 3:03 AM
V100 SXM2 pinout and NVLink reference
https://github.com/pigeon-labs-stack/v100-sxm2-reference
[comments] [12 points]
https://github.com/pigeon-labs-stack/v100-sxm2-reference
[comments] [12 points]
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
국가 슈퍼컴퓨터 6호기 ‘한강’이 12월 14일 공식 서비스를 시작 할 예정임. 10월 26일 베타서비스를 열었으며, 11월 2일 대전 KISTI 본원에서 가동식을 개최 NVIDIA GH200 8,336개와 H200 160개 , 총 GPU 8,496개를 탑재하며 CPU는 9,936개로 구성됨 이론 연산성능 614PF...
국가 슈퍼컴퓨터 6호기 ‘한강’이 12월 14일 공식 서비스를 시작 할 예정임. 10월 26일 베타서비스를 열었으며, 11월 2일 대전 KISTI 본원에서 가동식을 개최 NVIDIA GH200 8,336개와 H200 160개 , 총 GPU 8,496개를 탑재하며 CPU는 9,936개로 구성됨 이론 연산성능 614PF...
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
국가 슈퍼컴퓨터 6호기 ‘한강’이 12월 14일 공식 서비스를 시작 할 예정임. 10월 26일 베타서비스를 열었으며, 11월 2일 대전 KISTI 본원에서 가동식을 개최 NVIDIA GH200 8,336개와 H200 160개 , 총 GPU 8,496개를 탑재하며 CPU는 9,936개로 구성됨 이론 연산성능 614PF...
news.hada.io
November 3, 2026 at 3:00 AM
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
국가 슈퍼컴퓨터 6호기 ‘한강’이 12월 14일 공식 서비스를 시작 할 예정임. 10월 26일 베타서비스를 열었으며, 11월 2일 대전 KISTI 본원에서 가동식을 개최 NVIDIA GH200 8,336개와 H200 160개 , 총 GPU 8,496개를 탑재하며 CPU는 9,936개로 구성됨 이론 연산성능 614PF...
국가 슈퍼컴퓨터 6호기 ‘한강’이 12월 14일 공식 서비스를 시작 할 예정임. 10월 26일 베타서비스를 열었으며, 11월 2일 대전 KISTI 본원에서 가동식을 개최 NVIDIA GH200 8,336개와 H200 160개 , 총 GPU 8,496개를 탑재하며 CPU는 9,936개로 구성됨 이론 연산성능 614PF...
SpaceX just signed a massive $40B deal to lease GPUs, not buy them.
How this off-balance-sheet debt shifts the AI race—and who survives.
Full breakdown: lnkd.in/p/gBmx-kae
#AI #TechNews #Business #Nvidia
How this off-balance-sheet debt shifts the AI race—and who survives.
Full breakdown: lnkd.in/p/gBmx-kae
#AI #TechNews #Business #Nvidia
#spacex #ai #apollo #nvidia #aiinfrastructure #techfinance #technews | Sandeep Raiza
SpaceX Just Signed the Largest GPU Financing Deal Ever: $40 Billion, and the Chips Aren’t Even Bought, They’re Leased
On October 26, SpaceX disclosed a binding commitment that tells you a lot about h...
lnkd.in
November 3, 2026 at 2:59 AM
SpaceX just signed a massive $40B deal to lease GPUs, not buy them.
How this off-balance-sheet debt shifts the AI race—and who survives.
Full breakdown: lnkd.in/p/gBmx-kae
#AI #TechNews #Business #Nvidia
How this off-balance-sheet debt shifts the AI race—and who survives.
Full breakdown: lnkd.in/p/gBmx-kae
#AI #TechNews #Business #Nvidia
The ceiling there is the link, not the GPU. Experts streamed from NVMe arrive at a few GB/s over PCIe 4.0 x4, versus about 360 GB/s from a 3060's own VRAM, so keeping the hot experts resident is the whole game. PCIe and bandwidth math: hardwaretheory.net/lab/bandwidth
November 3, 2026 at 2:58 AM
The ceiling there is the link, not the GPU. Experts streamed from NVMe arrive at a few GB/s over PCIe 4.0 x4, versus about 360 GB/s from a 3060's own VRAM, so keeping the hot experts resident is the whole game. PCIe and bandwidth math: hardwaretheory.net/lab/bandwidth
ちょっと前からMacBook Air M1を仕事用に貸し出されて使っていて使用中。
ワタシのIntel MacBook Pro 2019とあんまりパワー変わらないから、Airに個人アカウントも作って使って自分のMacBook Proは売却しちゃおうかな、って思ったんだけど、やっぱメモリー容量の差と、GPU占有メモリの差は大きいな。
ワタシのIntel MacBook Pro 2019とあんまりパワー変わらないから、Airに個人アカウントも作って使って自分のMacBook Proは売却しちゃおうかな、って思ったんだけど、やっぱメモリー容量の差と、GPU占有メモリの差は大きいな。
November 3, 2026 at 2:58 AM
ちょっと前からMacBook Air M1を仕事用に貸し出されて使っていて使用中。
ワタシのIntel MacBook Pro 2019とあんまりパワー変わらないから、Airに個人アカウントも作って使って自分のMacBook Proは売却しちゃおうかな、って思ったんだけど、やっぱメモリー容量の差と、GPU占有メモリの差は大きいな。
ワタシのIntel MacBook Pro 2019とあんまりパワー変わらないから、Airに個人アカウントも作って使って自分のMacBook Proは売却しちゃおうかな、って思ったんだけど、やっぱメモリー容量の差と、GPU占有メモリの差は大きいな。
abliteration already happens, but i don't think? the majority of gpu use will go to abliterated models? maybe? with this, i am hopeful that the majority of gpu use will go to understood models?
November 3, 2026 at 2:58 AM
abliteration already happens, but i don't think? the majority of gpu use will go to abliterated models? maybe? with this, i am hopeful that the majority of gpu use will go to understood models?
I'm told that there are shops that'll swap the horrible 12VHPWR connector on 4000 and 5000-series Nvidia GPUs with other, less meltdown-prone arrangements and I sort of think I need to track one down to mod my Miku GPU. I don't consider its death a question of "if," only of "when."
November 3, 2026 at 2:41 AM
I'm told that there are shops that'll swap the horrible 12VHPWR connector on 4000 and 5000-series Nvidia GPUs with other, less meltdown-prone arrangements and I sort of think I need to track one down to mod my Miku GPU. I don't consider its death a question of "if," only of "when."
love that there’s games coming out with this kinda concept that also don’t need anything better than a gpu from when i was in high school
Queens Domain bottles the sensation of a PlayStation 1 or 2 game that kinda sucks to play without the specific parts that actually super suck to play.
November 3, 2026 at 2:35 AM
love that there’s games coming out with this kinda concept that also don’t need anything better than a gpu from when i was in high school
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
기존 CPU 기반 계산과학과 시뮬레이션에 더해 GPU를 활용한 AI 기반 과학연구까지 지원할 예정입니다. 이론 연산성능은 614PF로 ‘누리온’의 24배이며, 대학과 연구기관뿐 아니라 기업의 기술개발에도 활용할 계획입니다.
기존 CPU 기반 계산과학과 시뮬레이션에 더해 GPU를 활용한 AI 기반 과학연구까지 지원할 예정입니다. 이론 연산성능은 614PF로 ‘누리온’의 24배이며, 대학과 연구기관뿐 아니라 기업의 기술개발에도 활용할 계획입니다.
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
기존 CPU 기반 계산과학과 시뮬레이션에 더해 GPU를 활용한 AI 기반 과학연구까지 지원할 예정입니다. 이론 연산성능은 614PF로 ‘누리온’의 24배이며, 대학과 연구기관뿐 아니라 기업의 기술개발에도 활용할 계획입니다.
news.hada.io
November 3, 2026 at 2:31 AM
국가 슈퍼컴퓨터 ‘한강’, GPU 8,496개 탑재해 12월 14일 정식 서비스 시작
기존 CPU 기반 계산과학과 시뮬레이션에 더해 GPU를 활용한 AI 기반 과학연구까지 지원할 예정입니다. 이론 연산성능은 614PF로 ‘누리온’의 24배이며, 대학과 연구기관뿐 아니라 기업의 기술개발에도 활용할 계획입니다.
기존 CPU 기반 계산과학과 시뮬레이션에 더해 GPU를 활용한 AI 기반 과학연구까지 지원할 예정입니다. 이론 연산성능은 614PF로 ‘누리온’의 24배이며, 대학과 연구기관뿐 아니라 기업의 기술개발에도 활용할 계획입니다.
We need GPU prices to fall so these freaks have something to do. that's why if elected I will propose the GeForce For All bill on day one.
With Democrats poised to make significant gains in Tuesday’s midterm elections, Trump supporters have been reduced to posting bizarre Star Wars fan fiction.
The Online Right’s Closing Campaign Argument: Fascist Anakin Skywalker Memes
With Democrats poised to make significant gains in Tuesday’s midterm elections, Trump supporters have been reduced to posting bizarre Star Wars fan fiction.
buff.ly
November 3, 2026 at 2:30 AM
We need GPU prices to fall so these freaks have something to do. that's why if elected I will propose the GeForce For All bill on day one.
I used it because I'm used to it and eventually aspire to reverse engineer its GPU but honestly would have preferred Niri
November 3, 2026 at 2:17 AM
I used it because I'm used to it and eventually aspire to reverse engineer its GPU but honestly would have preferred Niri
🔥 オラクルが3.1兆円調達、AI投資の行方と日本企業への波及
#NewsBrief #JPNews
#NewsBrief #JPNews
オラクルが3.1兆円調達、AI投資の行方と日本企業への波及
オラクルが3.1兆円の資金調達を完了し、AIインフラ強化に注力する動きが日本のクラウド市場に影響を及ぼしている。同社はデータセンター容量を3倍に拡大し、GPUリソースを強化することでAIモデルの学習を支援。日本の金融機関を中心に23社と提携し、セキュリティとコンプライアンスに優れたクラウドサービスを提供している。これにより、国内クラウド事業者はAI対応サービスの強化を迫られ、サイバーエージェントや日本IBMも対抗策を進めている。
一方、日本のAI導入は医療分野で医師法の規制に直面している。AI...
news-brief.com
November 3, 2026 at 2:11 AM
🔥 オラクルが3.1兆円調達、AI投資の行方と日本企業への波及
#NewsBrief #JPNews
#NewsBrief #JPNews
“Video Games” but she’s comparing GPU prices and sighing at the checkout screen. The romance is gone.
November 3, 2026 at 2:07 AM
“Video Games” but she’s comparing GPU prices and sighing at the checkout screen. The romance is gone.
If you're building an Immich photo server, buying a GPU might be the most tempting upgrade. But will it really help with AI search, face recognition, and video processing, or just add cost? Find out before you spend. #AI
Does Immich need a GPU? CPU vs Intel Graphics vs NVIDIA for private AI photo search
Does Immich need a GPU for Smart Search? Compare CPU-only servers, Intel OpenVINO, NVIDIA CUDA and remote ML before spending money.
www.popularai.org
November 3, 2026 at 2:01 AM
If you're building an Immich photo server, buying a GPU might be the most tempting upgrade. But will it really help with AI search, face recognition, and video processing, or just add cost? Find out before you spend. #AI
So, the run crashed after 7 hours because I ran out of memory with SMOTE oversampling. Claude Code was all "I told you so." So, I'm making it run the project again. You can't defeat me, Claude (although having only 32 GB of RAM and 13 GB of GPU RAM might).
Claude Code thinks it can deter me. I've got all the time in the world.
November 3, 2026 at 1:45 AM
So, the run crashed after 7 hours because I ran out of memory with SMOTE oversampling. Claude Code was all "I told you so." So, I'm making it run the project again. You can't defeat me, Claude (although having only 32 GB of RAM and 13 GB of GPU RAM might).
You got a much faster GPU. What kind of prefill speed are you getting ? I'm using NVFP4 quant.
November 3, 2026 at 1:40 AM
You got a much faster GPU. What kind of prefill speed are you getting ? I'm using NVFP4 quant.