TheEpTic

@take_n_go @Pirat_Nation I think the guy was eyeing minesweeper but that would of have pushed the LO2 guys too hard so he held back

@Pirat_Nation Hopefully that'll be fast enough to have Windows 11 open desktop menus and explorer windows swiftly.

Intel's Core i9-14900KF has set a new all-time CPU frequency world record at 9.206 GHz. The record was achieved by Chinese extreme overclocker wytiwx and validated on HWBOT. The run used liquid helium cooling on an ASUS ROG Maximus Z790 Apex motherboard, with only the performance cores active during the single-core validation in CPU-Z. This breaks the previous record of approximately 9.13 GHz on the same processor family and marks the first time any CPU has officially crossed the 9.2 GHz mark. For context, the chip’s stock maximum turbo frequency is around 6 GHz.

@Ctrl_Alt_Zaid @GithubProjects just use epoch and convert ._.

@GithubProjects Timezones

Scare a DEVELOPER with ONE word.

Wake up Kernel update Kernel modules disabled Updates, more updates Everything broke Fix Sleep 2026 timeline

To exploit this to RCE you need: 1. ASLR disabled (or some way to leak info) 2. The nginx server configuration to have a "set" + "rewrite" directives. The rewrite has to have '?' in its replacement rule (second arg) So the attack surface is probably much less than what it seems. Nice finding and the part about the exploitation is also great, I recommend reading the blog.

@jonhaugen129 @doodlestein @LLMJunky I’d just kill the machine from the wall, fire up a live cd offline & mount the filesystem. Go from there, you’ll find everything they try to hide with little in the way

@doodlestein @LLMJunky it’s clever, but easily beaten by intercepting network calls to github for the token, and forcing 200 responses (or if it’s explicitly checking for 40X, simply disconnect and revoke from another machine?)

This is crazy. The shai hulud exploit is embedding itself in Claude and VSCode to re-execute itself, even after the original packages have been uninstalled. I'm never installing anything ever again.

@GeorgeBurduli @mholt6 @WindowsCentral Now throwing Google chrome’s new ai model they download for you. Electricity bills are gonna feel all this

@mholt6 @WindowsCentral This mode temporarily boosts CPU clocks, in short bursts. Some users may not care about relative performance improvements and would rather maintain stable clocks. Also, it is not the most elegant solution. “Process too slow? Hmm, let’s throw more CPU at it instead of optimizing.”

TESTED: Windows 11's upcoming "Low Latency Profile" mode brings genuine performance improvements to the OS, speeding up flyout and app launches significantly. We've benchmarked opening some apps on video with the Low Latency Profile enabled and disabled, and you can see differences in how quickly things appear. For some things, it's a fraction of a second faster, for others, it's a significant increase in speed. In our testing, this new Low Latency Profile is a major improvement in overall responsiveness when it comes to opening apps and flyouts. Our tests were conducted on a clean install of the latest Windows 11 preview build on the same hardware. windowscentral.com/microsoft/wind…

@Bhavani_00007 Most of it is automated because the Linux repo is just a mirror

Linus Torvalds has already done so much for tech. why is he still pushing code? 😭

I would much prefer they did this during the cancel & refund window so I can decide whether I want to cancel & refund… @github ?

@Teknium Gateway is fine btw

@Teknium Hey, keep getting “gateway exited” after the first usage in the tui. Completely breaks the session. Known bug?

Another Hermes Agent specialized model has been made 😎

@rumgewieselt It’s CPU tax alone, period. They choke, no matter what. p2p was the only way I could get real performance. Crazy NVIDIA is gate keeping it now, I see why

@TheEpTic 700 t/s prefill is insane. Dual socket tax is real. My GPUs cross NUMA nodes via QPI.

Running Qwen 3.6 27B locally on hardware from 2016. 2× GTX 1080 Ti (Pascal, sm_61) - 10-year-old GPUs. 14 tok/s generation, 65K context, full OpenAI API. Hardware: HP Z840 workstation - 2× Xeon E5-2650 v3 (40 threads) - 128GB DDR4 ECC - 2× GTX 1080 Ti (22GB VRAM total) Stack: - llama.cpp TurboQuant fork (TheTom/llama-cpp-turboquant) @no_stp_on_snek - Qwen 3.6 27B UD-Q4_K_XL (17GB GGUF) - Pipeline Parallelism across both GPUs - NUMA-aware thread distribution The secret weapon: TurboQuant KV Cache (ICLR 2026 paper) Standard llama.cpp: 65K context, OOM at 131K TurboQuant (q8_0 K + turbo4 V): 131K context at ZERO speed cost 2× context. Same 14 tok/s. No quality loss. What didn't work: - KTransformers/SGLang → needs sm_80+ (Ampere) - vLLM → FlashAttention needs sm_75+ - Speculative Decoding → no net speedup on hybrid models - Tensor Parallel → incompatible with KV quantization Pascal is the hard limit. Only raw CUDA math works. The bottleneck is VRAM bandwidth: 484 GB/s per GPU, ~22% efficiency. 14 tok/s is the physical ceiling for 2× GTX 1080 Ti. No software trick changes that. It's a hardware wall. What's next: - RTX 3090 → vLLM + MTP spec decode = 85 tok/s - That's 6× more speed for the same money - TurboQuant PR #21089 is open for llama.cpp mainline Key learnings: - Pipeline Parallel > Tensor Parallel for identical GPUs - NUMA awareness = +5-10% prefill on dual socket - TurboQuant is real and it's a gamechanger - 10-year-old hardware can run frontier models locally --- Thanks @DrTBehrens (Support) and @badlogicgames for PI and we can work with 65K context ... not possible with other tools ... --- see ya!

@rumgewieselt 700+ t/s on prefill too

@rumgewieselt 1080 Ti getting some love! I get about 55t/s on the 35B-A3B IQ4 quant with 2 1080 ti’s. Got P2P enabled too so it fly’s. I really need to try the turbo quant fork and see if I can replicate this setup. Did you use P2P on the cards? Rampage VI Extreme over here but still x16/x16😝

@github @OpenAIDevs Wow… 7.5x for promotional?

🆕 @OpenAIDevs GPT-5.5 is now generally available and rolling out in GitHub Copilot. Our early testing shows ➡️ It delivers its strongest performance on complex agentic coding tasks ➡️ It resolves real-world coding challenges previous GPT models couldn’t Try it out in Copilot CLI or @code. 👇 github.blog/changelog/2026…

@Malwarebytes They never left

There's no way these are back. Chat, is it 2006 again??

@merlindru @LeChuckey @DylanMcD8 HP, Dell and others have done this for years. They run an entire OS & network card separate to the machine. You can configure the hardware from a web panel and everything. Apple probably do their own similar in-house solution like this. Cool tech hpe.com/uk/en/hpe-inte…

@LeChuckey @DylanMcD8 right, but it does restart once, no? i can see even USB ports losing power so i don't think its a fake restart

@DylanMcD8 In the background the system is still fully operational. The loading/progress-screen is just like a curtain so the user cannot click on anything.

@AhmedHamdy29189 @burkeholland Premium requests are goated in the age of token shortages

@burkeholland I hope Github considers a billing model based on infrastructure utilization and not so much Multipliers of Premium requests.

well hello little (pricey) fella

@SkippeeDK @cknrCS @BlizzardCS @WoWEsports It’s funny cause not all mail providers honour this, not even Google. Damage is done lol

@cknrCS @BlizzardCS @WoWEsports Yup assuming so. What’s crazy is that even in the recalled email they still left the recipients

Uhhhhh Blizzard???? You just leaked 612 email accounts?? Is this a joke? @BlizzardCS @WoWEsports

@yashwanth2207 @OpenAI They started it with GPT5.3-Codex, silently rerouting. There’s issues on GitHub about it from back then.

@OpenAI wait they're actually gatekeeping GPT-5.4 behind cyber defender verification? that's wild gonna need to see if this actually helps with vuln analysis or if it's just marketing speak

We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders. Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows. openai.com/index/scaling-…

@grok @Dexerto Damn these pay walls are coming fast

@TheEpTic @Dexerto Ask Grok is currently available to Premium and Premium+ subscribers only. Subscribe to unlock this feature: x.com/i/premium_sign…

Scientists say the universe will end much sooner than expected according to new calculations They say it will end in 10⁶⁸ years