coreteks (@coreteks) header
coreteks (@coreteks) profile picture

Youtube: https://teapawt.org/t.co/AGqevEiUXC Patreon: https://teapawt.org/t.co/IkVmWNIBYI

Planet Earth
Joined February 2016
Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)Photo by coreteks (@coreteks)
coreteks retweeted
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1,820
6,414
2,194
41,187
6,673,788
SHOCKING: We have confirmed that multiple Neoclouds have been refused access to GTC Berlin due to them adding AMD GPUs, TPUs, and other chips to their cloud offerings Note that this doesn't apply to hyperscalers, as they have more buying power. But for neoclouds, executives feel NVIDIA has been using high-pressure tactics to keep them NVIDIA-only. As such, neocloud executives are starting to consider offering TPUs or AMD GPUs quietly to avoid getting pressured by NVIDIA and to avoid the feeling that NVIDIA will retaliate. teapawt.org/SemiAnalysis_/status/2…
SHOCKING: Many neocloud executives we spoke with feel that if they have non-NVIDIA networking gear on their cluster, or if their cloud has an AMD GPU or TPU offering, NVIDIA will retaliate. They feel that retaliation includes not giving early allocation or no longer supporting a potential IPO/VC raise.(1/3)🧵
191
166
109
1,795
588,304
coreteks retweeted
After a month of testing, @coreteks brings us this review of #Olares One! The video shows the actual workflow of running #OpenClaw and self-hosted apps directly on Olares One. Watch the full video: 👇 youtube.com/watch?v=XVu9lxgE…
2
5
1
17
1,245
coreteks retweeted
Replying to @Olares_OS @coreteks
My 🫸 exclusive 🫷 launch review of the Olares One, a truly groundbreaking personal computing platform: youtube.com/watch?v=XVu9lxgE…
2
4
480
Dialogue surréaliste entre Adolf Hitler et une gauchiste. C'est du troll de haut niveau. 🤣🤣🤣 ➡️ Telegram : t.me/KimJongUnique
96
1,202
116
3,178
129,362
From “Just buy it!” to the Paywall – How Purch and Future PLC Optimized Tom’s Hardware to Death and Are Still Doing It :( igorslab.de/tux3
3
6
2
15
5,553
Wuchang: Fallen Feathers Dev Seemingly Disabled Native Render Resolution on Certain Hardware To Fix Performance Issues tpu.me/kyhr
3
1
24
2,472
Looking at the list of companies represented in that board, we can't help but laugh our asses off. No wonder they'd be against consumer rights.
Europe's largest video game lobby spoke out against Stop Killing Games, saying it'd make game dev too expensive The lobby's board has members from EA, Ubisoft, Activision and several other publishers who have permanently killed games
56
417
12
2,963
81,016
Nvidia reported sales of $4.6 billion worth of H20 GPUs prior to the ban in April. Looking into the earnings reports of Tencent, Baidu, and Alibaba, and making an approx. calculation for ByteDance, these chinese tech giants purchased in the first quarter of 2025: * Tencent: $1.6 billion worth of GPUs * Alibaba: approximately $1.2–1.4 billion * ByteDance: approximately $1.2–1.4 billion All were billed to their subsidiaries in Singapore. In addition, Baidu purchased $860 million worth of GPUs in the first quarter. These could be divided between Huawei GPUs and Nvidia — possibly H20, but potentially other models as well. Reports haven't mentioned Baidu purchasing H20 back in February/March.
Nvidia H20, Singapore and Billing Invoices Looking closely at the disclosure about H20s in Nvidia's 10-Q, we see that Nvidia reported sales of $4.6 billion worth of H20 units prior to the U.S. government's ban. Now, For the entire quarter, revenue from China (including Hong Kong) totaled $5.5 billion. This means that if all H20 sales were billed to Chinese enterprises, only $900 million in revenue would be left for other products and segments — which is not impossible. Now, where could these H20s have been billed to? Let's consider Singapore, our usual suspect. In Q4, we saw—for the first time—a decline in billing to Singapore, which ended the quarter with just over $6 billion in total billed revenue. Historically, Singapore’s revenue had been increasing by about $1.5 billion on average each quarter, ever since companies suddenly began billing through Singapore. Now looking at the revenue jump from Q4 to Q1, there was an unusual increase of $3 billion — a 50% rise — which is $1.5 billion above the average quarterly growth in Singapore’s brief revenue history. So we can reasonably assume that at least $1 billion worth of H20s were billed to Chinese companies (since only Chinese companies buy H20s) through their subsidiaries in Singapore. My guess would be ByteDance, Tencent, or Alibaba, as they each have reasons to bill through Singapore instead of China. This just shows how Chinese companies use Singapore as a billing location—both for legitimate and less-than-legitimate purposes—to get their GPUs.
3
12
1
51
26,718
SPAIN, PORTUGAL, FRANCE HIT BY MAJOR POWER OUTAGE A widespread power outage struck Spain and neighboring regions in France and Portugal, according to local media sources on Monday. "At around 12:30 pm this Monday, the lights in homes, businesses and stores across the country went out. Gradually, users have also begun to report problems accessing data and Wi-Fi on their cell phones," local paper La Vanguardia wrote. Public transportation has halted, metro services are suspended, businesses are closing their doors and public vehicles are immobilized due to non-functioning traffic signals. Currently, authorities have not identified the reason for the blackout.
174
487
179
2,790
878,331
Confirmed Electricity is down in entire Portugal and Spain peninsula! Crazy!
Power down in whole Portugal and Spain? My power down too
689
996
1,341
12,219
15,493,439
Replying to @coreteks
Don't play the numbers game..it looks weak... just say you are happy for the 136K and move on...I love the channel, but lately you really have been either on the defense or on the attack..I don't know what's going on.
1
1
1
397
AI Playground 2.0.4 beta update is now out. - DeepSeek models now prepopulated to our model list in the Answer tab - Improved language support - New CopyFace workflow - Improved installer - Other fixes Download: lnkd.in/gDT2imuq Release Notes: lnkd.in/grj-PDmP Overview: youtube.com/watch?v=4q-CPZ1R… DeepSeek Demo: teapawt.org/bobduffy/status/188469… Get going with AI Playground, the easiest and most feature rich Gen AI application suite, using the hardware in your Intel-powered AI PC. Supporting Intel® Core™Ultra-H and Intel® Core™ Ultra Series 2 Processors or #IntelArc discrete GPUs
Yep DeepSeek R1 runs on Intel AI PCs using AI Playground 2.0. This is recorded on Intel Core Ultra Series 2 - Lunar Lake showing how to install and run in using AI Playground 2.0 1. Install AI Playground 2.0: intel.com/ai-playground 2. On Answer Tab, click the + icon then paste this: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B 3. Confirm, wait for install, then prompt Have fun!!! #intelarc
2
4
9
649
🎄🐾 Our Christmas gift set just got a whole lot cuter! Thanks for @coreteks sharing this adorable moment. ❤️✨
Minisforum kindly sent me a xmas care package and among other goodies inside was this xmas tree pillow that just so happens to fit perfectly through Oscar's head. I just had to.
2
2
13
2,860
Replying to @coreteks
Did you watch gamers nexus tribute. Turns out there is a go fund me to pay for medical bills...
1
1
3
680
coreteks retweeted
"NextSilicon's Maverick-2, a novel approach to HPC and AI acceleration" Read about it on the coreteks substack: open.substack.com/pub/corete…
3
1
7
1,303
I am really excited to reveal what @GoogleDeepMind's Open Endedness Team has been up to 🚀. We introduce Genie 🧞, a foundation world model trained exclusively from Internet videos that can generate an endless variety of action-controllable 2D worlds given image prompts.
129
531
265
2,388
833,723
🚨 BREAKING: Google DeepMind just revealed Gemini- ChatGPT's biggest competitor. Gemini is the FIRST multimodal AI to outperform human experts on the MMLU, scoring over 90%.
253
1,699
720
8,056
1,860,138
My keyboard wasn't mechanical enough, so I made an upgrade
107
962
161
6,777
750,185
Testing out a ScanStore scan to Metahuman pipeline. Realistic textures and 1 to 1 normal maps correct to the face shape, make a big difference to the look of the character.
53
363
44
3,290
396,389