SCROLL NEWS / DISCOVERY
Search headlines
Find the stories shaping the conversation.
Results for " ocm " 70 found 🔀 AI-powered Shuffle
Sign in to save shuffle history.
Blue Turf Thinkers: Biomedical Research Institute investigates harmful diseases
At each Boise State home football game, the “Blue Turf Thinkers” series honors outstanding groups...
Voices: We see antibiotics save lives. But unnecessary prescriptions are hurting more than our health.
“The consequences of unnecessary antibiotics don’t stop with the patient,” two infectious diseases physicians write in an op-ed.
Chicago’s gunshot detection pilot plan gets a sudden overhaul as aldermen erupt
Chicago officials are revamping a gunshot detection pilot just weeks after announcing it, frustrating aldermen as pressure mounts to replace ShotSpotter.
AP Decision Notes: What to expect in Georgia’s state primary runoff
Georgia Republicans return to the polls Tuesday to pick nominees for U.S. Senate and governor in a state primary runoff election
New York state launches center for cannabis care and health
The state Office of Cannabis Management on Thursday announced the launch of its Center of Excellence for Cannabis Care and Health Equity.
WTAMU to host business startup seminar Monday
West Texas A&M University is set to host a business startup seminar Monday.
Apple's Beta Website Says 'We'll Be Back Soon' Just Ahead of WWDC
Apple's Beta Software Program website has gone down ahead of the company's WWDC 2026 keynote next hour. Apple's Beta Software Program website has gone down ahead of the company's WWDC 2026 keynote next hour.
Sports make up 96 of top 100 telecasts in 2025, tying an all-time record
Sports will account for 96 of the 100 most-watched shows of 2025, which is tied with 2023 for a record number and up from 87 during an Olympic and election year in 2024.
Henderson welcomes third cohort of small business assistance program
The city celebrated the 2026 cohort with a kick-off event at Henderson City Hall.
Spirit Airlines shuts down after White House rescue deal falls through
The airline said early Saturday it had “no choice” but to wind down operations immediately, dashing hopes of a last-minute financial lifeline.
Trump cancels hearing for intelligence chief pick, says Pulte will remain
The president linked the move to his efforts to pass voting restrictions and said Jay Clayton’s hearing would be delayed until his replacement as U.S. attorney is confirmed.
Blue Turf Thinkers: College of Engineering builds Idaho's energy workforce
At each Boise State home football game, the “Blue Turf Thinkers” series honors outstanding groups...
Can $100 million overcome a Trump endorsement? What to watch in Tuesday's elections
An endorsement from President Donald Trump is worth a lot in Republican primaries
Prominent Cambodian opposition politician seeks Supreme Court reversal of incitement conviction
A prominent Cambodian opposition politician has appealed his conviction on incitement charges to the Supreme Court
Fantasy Football Start 'Em, Sit 'Em Week 16: TE Edition
Which tight end is going to take you to a Christmas fantasy championship?
A far-right backlash is surging in Latin America as crime fears fuel Bukele-style crackdowns
At the start of this decade, Latin America leaned left as progressive politicians gained power amid public outrage over inequities worsened by the pandemic
Judge wins fight to open a Miami mental-health center as an alternative to jail
Steve Leifman has been fighting for decades to open the center. On Tuesday, Miami-Dade commissioners gave the green light.
Trump deepens the dustup with Italy's Meloni over a disputed photo from the G7 summit
President Donald Trump is lashing out at Italian Prime Minister Giorgia Meloni once more
OCM shuts down 2 illegal Rochester cannabis shops
Officials say more than $1.3 million in cannabis products were seized.
'We are getting very, very close': NASA makes final Artemis II preparations as expectation grows for Wednesday launch
NASA's Artemis II crew is ready to fly around the moon, with promising weather forecasts for a Wednesday launch.
Ten Exits SM Entertainment—Here's What It Means For NCT and WayV Full Statement Drama
The K-pop star's departure was announced April 6.
Washington, DC, voters cast ballots in crucial primaries as Trump reshapes the capital
Voters in Washington, D.C. are heading to the polls to select party candidates for mayor and congressional delegate
[Q&A] "Actually, It Is Rocket Science," with Tim Durtschi
Tim Durtschi talks Haines, big lines, snowmobiles and the unexpected variables in filming a ski movie that make us question, is it rocket science?
MacBook Ultra's New Design Won't Stay Exclusive for Long
The touchscreen and the OLED panel have dominated discussion about Apple's next high-end laptop, but the redesign wrapped around them may prove to be an equally consequential change, largely because it is not expected to stay unique to the machine for long. Bloomberg's Mark Gurman last week laid out Apple's plan to refresh every Mac it sells over the next two years.
The war on Iran’s energy infrastructure creates rising political risk
When Israel targeted Tehran’s oil storage facilities in the initial days of the war launched on Feb. 28, black smoke blanketed the capital and plunged parts of the city into darkness. Roughly ten days later, strikes hit natural gas facilities at the giant South Pars field, key oil refineries, major steel plants, and petrochemical complexes, impacting thousands of workers. These attacks on vital energy and industrial assets revealed a critical dimension of the conflict: compounding an already acute domestic energy...
Why Trump’s payout fund was too much for Republicans
Of all the president’s moves, the fund is the first to prompt real GOP resistance, signaling a sharply shifting political environment.
Trump endorses Collins in Georgia Senate runoff. It's his latest 'MAGA' pick in Republican primaries
Days before the U.S. Senate runoff in Georgia, President Donald Trump has made an endorsement
Athlon Sports Golf Is Building a Bigger Daily Home for Golf Fans
Athlon Sports is expanding its golf footprint, and the next step is built for the way fans follow the game now. The new social channels will extend the growing Athlon Sports golf platform beyond written stories, creating a more immediate home for news, analysis, reactions, player storylines, tournament coverage, instruction, equipment talk and the moments that make golf feel bigger than a scorecard. Athlon Sports Golf is building a bigger daily home for golf fans.
Newly discovered iron minerals might hold primordial water from when Earth formed
Researchers may have found a missing link in how Earth's mantle gets wet: newly discovered iron minerals that can store water at the extremely high temperatures and pressures present near our planet's core.
8 more Central Florida restaurants shut down over health violations
Last week, eight Central Florida eateries were forced to close after a visit from a state health inspector, according to state records.
Versatile Farrugia raring to bring 'unpredictability' to Tannadice
Versatile new United signing Neil Faruggia admits he's "happiest" where he's creating chances, having played a "multitude of positions" this season. The 26-year-old has joined from League One Barnsley, having previously featured for frequent League of Ireland winners Shamrock Rovers. "Having the option to have both feet helps you be more unpredictable I found, especially as a forward player," Farrugia told DUTV.
Engrams Embedding Entendre: Codesign for Efficient DRAM/SSD Offloading
Engram extends standard token embeddings with learned multi-token lookups. Recurring local patterns retrieve vectors directly, reducing the need to reconstruct them through attention and feed-forward layers. With Engram model architecture optimization, it allows for lower HBM capacity to be needed for models at the same quality. This does not mean there won’t be an insane demand for HBM but it just means that model architecture will continue to innovate around constraints. This model architecture design is naturally codesigned for parameter offloading: each token accesses a few embedding rows whose addresses depend on token IDs, not hidden states. The runtime can prefetch those rows from host DRAM while earlier layers compute, keeping the table outside HBM without transferring entire weight matrices. Our Memory model contains our latest estimates of quarter by quarter HBM, DRAM, & NAND supply and demand. Offloading frees HBM for model weights and KV cache, potentially supporting larger batches or more concurrent sessions. When DRAM becomes the next constraint, NVMe offers another tier. Recommendation systems already cache frequently or recently accessed embedding rows in faster memory while backing colder rows with SSDs. After NVIDIA roadmap had to change due to massively despec’ing Rubin Ultra from 1024GB to now ~200GB of HBM per chip, model architecture optimizations like emgram maybe helpful. Our DeepSeek-V4.1-Flash configuration uses roughly 189 GiB of memory for Engram. We replace it with a memory-mapped (mmap) file and measure serving performance with offload to SSD. Later on in our report, we will show our Engram offloading experiments along with the official InferenceX agentic inference serving results on Engram models like DeepSeekv4.1 Flash across all 6 NVIDIA GPU SKUs along with MI355X. Unsurprisingly, the CUDA Moat is still mogging MI355X on the ultra popular DeepSeekV4.1 Flash model. We also show how even on high capacity HBM SKUs, offloading emgrams to DRAM could result in even better performance for most of the pareto than keeping the emgram in HBM. Our benchmark has been widely reproduced, validated and/or supported by almost every major buyer of compute from Google Cloud to Microsoft Azure to Oracle, to Meta and many more. Furthermore, it has the support of the ML community including from vLLM, LMCache, SGLang, PyTorch, Huggingface and the support of major labs like OpenAI, MiniMax, ZAI, Qwen, Moonshot Kimi, etc. Star the InferenceX GitHub repository if you find the open-source benchmark and data useful!. InferenceX is the only inference benchmark in the world to have TPUv7, Jalapeño, Nvidia Rubin NVL72, AMD, and soon, SambaNova and Trainium. Due to how realistic AgentX scenario is to real world agentic inference workloads, AMD has committed to collaborating on MI455X UALoE72 too. DeepSeek did not release the original paper’s two trained Engram models. We replicated its setup on fineweb-edu using the released code and training hyperparameters, at an estimated 6E18 FLOPs per run. We observed the same U-shape scaling: Engram improved performance over pure MoE baselines. We also reproduced DeepSeek’s results, where earlier-layer representations with Engram resembled those of later layers. Like the original Engram paper, it is possible to probe Engram’s gate scores to see what n-grams DeepSeek-V4.1-Flash makes the most use of. Our gate scan finds names, code fragments, relational phrasing, and boilerplate. These examples prioritize interesting-ness over gate strength. One unexpected result was Wright : Ace Attorney. These examples suggest learned memory optimizes the training objective, not a judgment of which facts deserve storage. Licenses, bibliography fragments, API scaffolding, and website furniture can provide prediction shortcuts, so the value of additional Engram capacity may depend on what survives data preparation. This does not show that table capacity is “wasted”: the evaluation-corpus scan establishes neither training exposure nor the capacity occupied by each category. For offloading, strong gates do not identify cache-hot rows. Low gates do not automatically save reads either: computing the gate requires the retrieved key and undoes the performance gain of a fused kernel. Skipping reads would require a separate usefulness predictor before retrieval. In the original paper’s inference-time ablation, factual-knowledge benchmarks retained just 29–44% of their original performance, while reading comprehension retained 81–93%. This is due to the training–inference mismatch. The resulting degradation therefore measures this trained model’s dependence on Engram, not the performance difference between models trained with and without it. In our ablations, suppressing Engram worsens token likelihood across all evaluated domains, especially encyclopedia text and several code corpora. Surprisingly, GSM8K accuracy stays within measured run-to-run variation and removing Engram has no effect. Engram is not a detachable dictionary beside an unchanged MoE. Removing it changes downstream features and expert selection. We tested whether rerouting hurts or compensates by holding tokens fixed in a teacher-forced experiment on CRUXEval, a code-reasoning benchmark of small Python functions where the model predicts a function’s output from its code and an input, and scoring the reference answer. Removing Engram raised answer loss from 0.2848 to 0.3093 bits/token. Forcing the ablated model to use the original Engram-on expert choices made it worse still, at 0.3375 bits/token. Rerouting partially compensates for the missing memory. Memory features and expert selection work together, rather than following a clean “memory stores facts; experts reason” division. On the same CRUXeval, removing Engram during either phase reduced accuracy and increased generated tokens; removing it throughout produced the largest changes. Keeping Engram for prefill leads to more correct answers than keeping for decode likely due to semantically richer KV cache transferred to decode workers, allowing it to mitigate some of the performance loss. Engram’s table is large, but each lookup is small. DeepSeek-V4.1-Flash requests 24 rows at each of two Engram layers, about 12.4 KiB per processed token position across the model, or 3.1 KiB per GPU when split across four GPUs. Currently as of Day 7 since Model Release, MI355X is still 2-4x worse performance per dollar compared to B200 even when normalized by Mi355X’s lower TCO. Our full total cost of ownership breakdown comes from our AI Cloud TCO Model along with monthly market surveys of over 100+ gpu clouds & gpu cloud customers. On the Day 0 release of DeepSeekv4.1 Flash, NVIDIA vLLM works out of the box with zero issues across all 6 SKUs: H100, H200, B200, B300, GB200, GB300! This was thanks to the amazing work by the NVIDIA & Interact teams! In comparison, AMD vLLM did not work on day 0 for DeepSeekv4.1 Flash.Source: SemiAnalysis InferenceX AMD’s vLLM documentation points to using vllm/vllm-openai-rocm:deepseekv41-flash-0909, but from hour 0 of the model release to hour 23, AMD has not publicly released the image. AMD claims, “SPEED IS THE MOAT,” yet it has still not released it by the 23rd hour. We wish that, going forward, the AMD team has a better process for hour 0 model releases. Eventually when they did publicly release for “day 0” image support, performance-wise, it is currently up to 14.8x worse perf per dollar than H200 and up to 42x worse perf per dollar than B200/B300. The power of the CUDA MOAT is NVIDIA’s collaboration with its massive 6 million-developer community ecosystem including most of the vLLM & SGLang & Tokenspeed maintainers which means that CUDA is optimized on day 0. Overall, AMD did make significant improvements but the performance per dollar is still currently 2-4x worse than B200. AgentX Engram DRAM Offloading Improving Performance The HBM and DRAM offload use the same GPU kernel to select and dequantize rows. With HBM, it reads GPU memory; with Unified Virtual Addressing (UVA), it reads pinned host memory directly. Both support full decode graphs. Moving the table into HBM accelerates only the sparse lookup, leaving decoder computation and communication unchanged, resulting in little overall benefit while consuming memory otherwise available to KV cache. Another benefit of Engram offloaded to DRAM which means you can reduce the communication overhead by using less HBM GPUs per replica. For example, when enabling Engram offloading on B300, we are able to switch from TP4 to now TP2 which improves the pareto curve by up to 1.6x. When iso-model quality, less HBM is required as the engrams could be offloaded to host DRAM. Thus HBM bandwidth matters way more than HBM capacity. For inference workloads where memory bandwidth matters the most, 4-hi HBM provides the best $/bandwidth and therefore lowest cost per token. If China continues to make more and more revolutionary model architecture innovations, soon it could potentially 0Hi HBM stacks. [ ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) [ Long Live the Short King: Why 4-hi HBM Wins ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) Myron Xie, Bryan Shan, and 3 others · Sep 13 [ Read full story ](https://newsletter.semianalysis.com/p/long-live-the-short-king-why-4-hi) Moreover, on B300 and week-0 stack, moving the Engram table back to HBM did not improve results and stay within run-to-run variance. This is a result of the work optimizing DRAM offload, such as async, and overlap. SSD Offloading On B200, we created a unoptimized vLLM fork and stored the Engram tables in memory-mapped files on local SSDs. File backing lets the OS reclaim table pages when other applications need RAM. Pages already cached in memory can be served without reading the SSD again. Furthermore, note that we were unable to turn on GDS The unoptimized SSD implementation changes how rows reach the GPU. It copies row IDs to the CPU, deduplicates them, gathers the requested rows into pinned buffers, copies those rows back to the GPU and dequantizes them. This work runs between segments of the GPU execution graph. Native UVA performs row selection and dequantization directly on the GPU, avoiding the CPU round trip. A warm filesystem cache removes physical SSD reads, but leaves the coordination, row gathering and transfers. This is why a file already cached in RAM can still perform worse than a pinned DRAM table. The comparison measures the whole serving path; it does not separate the time spent on each of these operations. B200 DRAM dominates both measured SSD serving curves in total tokens per dollar and P90 interactivity. Near 125 tokens/s/user, DRAM delivers 121 million total tokens per dollar versus 52 million for SSD. For production serving, SSD offloading is likely not worth the tradeoff. On the B200 configurations we measured, SSD offloading loses on both measures: every observed SSD point has a DRAM alternative that delivers higher P90 interactivity and more total tokens per dollar. The only points where Cheaper storage does not automatically produce a cheaper inference service. Moving Engram to SSD leaves the same four expensive GPUs and the rest of the server in place. Reclaiming RAM only creates an economic benefit if it enables a cheaper server configuration or additional useful capacity. The current unoptimized path provides neither benefit, and the filesystem cache still consumes RAM when table pages are resident. Next we will look at the mechanisms and specific implementation of ngrams from DeepSeekv4.1 Flash, LongCat, Qwen3.8 Flash Next.
INTERVIEW: Sorority hosting ‘Fall into Wellness’ health fair and blood drive
As the seasons change, Midlands residents are encouraged to prioritize preventive health and give back to local hospitals during an upcoming community wellness event.
Windows Remote Access Connection Manager Vulnerability Enables Arbitrary Code Execution
A critical security issue involving the Windows Remote Access Connection Manager (RasMan) that allows local attackers to execute arbitrary code with System privileges.
49ers land Steelers’ 16-sack edge rusher in trade proposal
The 49ers could use their draft capital to upgrade positions of need with veterans instead of rookies.
Heather MacLeod | ChicagoCMO ORBIE® Awards
Heather MacLeod is a ChicagoCMO ORBIE® Awards finalist.
Bob Skinner, former Pirates outfielder and coach who won 2 World Series with club, dies at 94
Skinner was a three-time All-Star and won two World Series as a player.
Japan's tech business SoftBank rolls out OpenAI 'patches' against cyberattacks
Japanese technology giant SoftBank Group Corp. is launching a service using OpenAI technology to protect against the looming threat of cyberattacks
In Georgia’s Capitol, Republicans' redistricting session to begin without maps
Georgia is the next Southern state where Republicans are convening to redraw political districts in ways that could diminish the political power of Black and other nonwhite voters
Bills seek to promote telemedicine and regulate AI in New Hampshire’s health insurance industry
A Nashua lawmaker has filed two bills related to New Hampshire’s health insurance industry: one would prohibit health insurers from using artificial intelligence to overrule doctors while the other would prevent insurers from reimbursing providers less money for equivalent services provided via telemedicine.
Medicare Fee Schedule for 2027 Favors Psych Collab Care Models
Behavioral health services, including the psychiatric collaborative care model (CoCM), received rate increases from the Centers for Medicare &
Board Approves Changes to Winter Sports Season Regulations
STEVENS POINT, Wis. – The Wisconsin Interscholastic Athletic Association Board of Control approved limited use of the shot clock in basketball in 2026-27 and acted on other recommendations...
Roaches Galore: 12 more Central Florida eateries shut down over health issues
Last week, over a dozen Central Florida eateries were forced to shut down (at least temporarily) after a visit from a health inspector, according to state records.
Italy’s Meloni faces a far-right dilemma as 'Il Generale' Vannacci rises
Roberto Vannacci, a former Italian army general, is shaking up Italy’s political scene with his new party, Futuro Nazionale
Ebola cases in eastern Congo climb to 782 and deaths reach 181, authorities say
The Congolese Ministry of Health reports that confirmed Ebola cases in Congo have increased to 782, with 181 deaths
Mamdani tests his political clout in New York's primary as he looks to reshape the Democratic Party
New York City Mayor Zohran Mamdani is testing the limits of his newfound political muscle in a bid to reshape the Democratic Party, even if it means challenging his own party’s leadership
Old Pro Golf upgrade in OCMD, Berlin Giant in big 2026 business news
Exciting business news is all around for 2026, with new locations for Chick-Fil-A, Giant and more coming to the Delmarva area.
Flyers-Hurricanes Game 4: Lines, Notes, & How to Watch
The Philadelphia Flyers (0-3, 4-5) face elimination on Saturday when they host the Carolina Hurricanes (3-0, 7-0) at Xfinity Mobile […] The post Flyers-Hurricanes Game 4: Lines, Notes, & How to Watch appeared first on Philly Hockey Now.
California Gov. Gavin Newsom says Trump's Justice Department is investigating him and his wife
Democratic California Gov. Gavin Newsom said Monday that the Justice Department is investigating him and his wife, accusing the Trump administration of targeting him for political purposes as he weighs a presidential bid
Mitch McConnell shares health update, says he's been discharged
Sen. Mitch McConnell gave an update on his health nearly two months after he was rushed to the hospital, saying he'd continue to recover at home.
Promoter of election conspiracy theories wins GOP primary for Nevada secretary of state
Former state lawmaker Jim Marchant has won the Republican nomination for Nevada secretary of state
Trump taps Keith Sonderling, a longtime ally of business, for Labor Secretary
President Trump nominated acting secretary Keith Sonderling to permanently lead the Labor Department after months of turmoil. Business leaders who backed his candidacy say he has already been steering policy and personnel decisions.
'MacBook Ultra' Coming Soon With These 12 New Features
Apple's first high-end MacBook redesign in five years is expected to arrive in the near future, bringing a run of firsts to the Mac. Bloomberg's Mark Gurman expects the laptop to arrive between late 2026 and early 2027, with the later window now looking more likely owing to the global memory chip shortage. Analyst Ming-Chi Kuo expects mass production to begin by late 2026.
Some East Grand Forks businesses must register with city to sell low-potency hemp products
Changes in requirements from the state Office of Cannabis Management mean businesses selling these products must now be licensed with the OCM and registered with their local cities.
A more personal digital health experience for people in Europe
Google and DocMorris have announced a partnership to create a more intuitive and supportive digital health experience.
‘United with GRIT’ to bring adaptive sports to Minot high schools
Prairie Grit is launching a series called United with GRIT where they will pair high school athletics with adaptive sports. The series will kick off at the Minot United hockey game on January 29.
AP Decision Notes: What to expect in Oklahoma’s state primary
Oklahoma voters will select nominees for federal and state offices including U.S. senator and representative, governor, lieutenant governor, attorney general and both houses of the state Legislature in a state primary on Tuesday
'MacBook Ultra' Coming Soon With Touchscreen and 5 Other Changes
Reports and rumors suggest the next high-end MacBook Pro that Apple will release might not be a MacBook Pro at all. It could actually be something altogether new and more exciting – a "MacBook Ultra" – marketed as the company's top-tier laptop offering. The device is likely to sit above the current M5 Pro and M5 Max MacBook Pros, rather than replace them, suggesting that they will remain on sale.
Oklahoma begins choosing a new US senator and governor in crowded primary
U.S. Rep. Kevin Hern of Oklahoma is hoping to win the Republican nomination for the Senate seat once held by U.S. Homeland Security Secretary Markwayne Mullin
RFK Jr. overrules experts to keep hantavirus cruise ship passenger in quarantine
U.S. Health Secretary Robert F. Kennedy Jr. has refused to release a cruise ship passenger exposed to hantavirus from a quarantine facility in Nebraska
Congress wonders as the Iran war draws to a close: Was it worth it?
The question hangs in the halls at the Capitol: Was the war with Iran worth it
Two sides of a political chasm share one fear in Colombia’s presidential race: A return to the past
More than 10 million people are victims of the harshest acts of violence in Colombia’s six decades of armed conflict
DC voters face a new political era without Eleanor Holmes Norton, after her 18 terms in Congress
For the first time in a generation, Washingtonians are waking up to a general election lineup that doesn’t include Eleanor Holmes Norton as delegate
Rep. Kean to return to Congress after mysterious absence, spokesman says
The New Jersey Republican, who has not been present in the chamber since March because of a "personal medical issue", plans to be back on the job later this month.
Man accused of robbing multiple people at gunpoint in Macomb County business
A Macomb County man is facing multiple felony charges after allegedly robbing several people at gunpoint.
FROM ROBOTS TO GAME SHOWS: MSC CRUISES UNVEILS EXCITING NEW ENTERTAINMENT FOR 2026
Media Room