Bobert

[Kaguya’s Top Gal @hayasaka_aryan]

When Mitch Mcconnell literally dies but then he hears Trump is slightly cutting aid to Israel

[Oilfield Rando @Oilfield_Rando]

“But what if Elmo is in Black English class at school, can Elmo say it then?”

[Chris Martz @ChrisMartzWX]

If you generate me more fake scary red clown maps, you might be able to convince me that we are facing a climate crisis.

[Chris Martz @ChrisMartzWX]

If droughts are becoming more frequent and severe like you claim, then why does the real-world data beg to differ with that claim, Dr. Overhype?

[Spencer Pratt @spencerpratt]

I will never stop fighting for my community.

[Ken Colbert @KColbertReport]

INDIANA

EVANSVILLE was infiltrated with 10,000 TPS Aka illegal Haitians after the plandemic.

Local employers were hiring the H1-B slave labor, under the approval of Federal legislators without any recourse.

Interestingly a local credit union administrator, that personally owns low-income housing, was providing section 8 housing to the TPS.

Now we’re discovering that the illegals are receiving mortgages to purchase Homes through Federal programs. No proof of income, no credit requirements that are required of US Citizens.

We are digging on the local banks and credit unions. Discernment says we will be uncovering a premeditated agenda.

EVANSVILLE is the epicenter of the socialist and Marxist agenda. The societal collapse is imminent.

@MicahBeckwith @LGMicahBeckwith @GovBraun @Indy_reporter_ @indianagop @EvansvilleINGov

[cutoffs.io @cutoffs_io]

🇺🇸🇮🇳 Xbox Indian CEO Asha Sharma just announced firing 3,200 Americans.

She filed for 5,000 H-1B visa hires this year.

Asha was appointed CEO of Xbox in February by Indian Microsoft CEO Satya Nadella.

[Josh Pigford @Shpigford]

Knockoff is now live!

Filter out the knockoff crap brands on Amazon.

Sorry to brands like WNPETHOME, EHEYCIGA, YXYL, LU&MN, JOYIN, TOMY, GODONLIF, YOOJEE, LINGTENG, LANEIGE, VISCOO, BIODANCE, COOFANDY, BALENNZ, TOSY and LUENX.

https://t.co/9mLk0EAsfG

[Aesthetica @Anc_Aesthetics]

It's national suicide that off-shoring hasn't been addressed yet. Our biggest tech companies are being hollowed out by Indian CEOs moving American jobs to India.

It's an order of magnitude worse than the hollowing out of our manufacturing base which destroyed small town America and it's happening right before our eyes.

[Angry Cops @AngryCops]

A fucking spear?!?! Everyone’s racist uncle is going to forward this screaming ”See! SEE!!!!”

[Big Glenn Network @BigGlennNetwork]

New fantasy football punishment unlocked

[Big Boss @0xBADB01E]

First you have to understand that modern LLM inference already disaggregates weights as models outgrew single chips years ago. You shard either by layer (pipeline parallelism) or by slicing every layer (tensor parallelism), and the two do very different things.

As an example, let’s look at Llama 3.3. It has 70B of weights and at FP8 that’s 70 GB of memory which is enough to fit on a single H100. Now that H100 has 3.35 TB/s of HBM, so the fastest it can ever decode for one user is 70/3.35 ≈ 21 ms/token or ~48 tok/s while using under 1% of its FLOPs. Now if we pipeline it across 8 chips: each chip holds ~8.75 GB, which means it only needs 1/8th the bandwidth and 1/8th the FLOPs to sustain the same aggregate throughput. Now crucially the token/sec a user gets is limited by the amount of data that crosses the link. In current LLMs all that is a small amount of activations for LLama 3.3 it’s ~8 KB per token….

Yes, you read that right it’s 8 KILOBYTES we are sending over a <900 GB/s link. That’s only 9 ns of serialization time but the overhead of 224G PAM4 SerDes adds ~100 ns per link traversal with RS-FEC which is 11x longer than the payload itself. And then you have the NVSwitch adding ~300 ns per hop and you need to pay twice. That’s ~600 ns of just hardware latency wrapped around 9 ns of data making a 98% tax before software even shows up. Then NCCL’s collective stack turns 600 ns into 10-20+ us… all to move 8 kilobytes lol. For comparison 8 KB serializes over 10 Gigabit Ethernet NRZ, in just 6.6 us. Pipeline parallelism however doesn’t make a single user faster as the token still needs to visit every layer in the sequence, so per-user speed is still weights / per-chip bandwidth.

To get more speed per user token you need to use tensor parallelism and have all the chips work on the same layer simultaneously. TP costs you 2 all reduce OPs per layer, 160 per token on llama 3, that’s still kilobytes of traffic but with NVLink overhead it’s a massive tax and why pipeline parallelism on most models still gives more interactivity per user. However, this gives you a huge latency lever to pull that scales tokens per second with interconnect speed instead of memory BW.

The clever amongst you might have also realized that sharding doesn’t just cut memory bandwidth per chip it also cuts FLOPs per chip and is why we have such bad MFU on decode. So once you’ve sized the link for the memory, you need to size the compute for it too. This is called “balancing the pipeline”, and currently no shipping chip does it because they were all designed as standalone monsters. Remember Tokens/sec = ~aggregate memory BW / bytes touched per token. At batch 64 in FP4 you need ~250 FLOPs per byte, and Blackwell ships 1,250. Provisioned 5x more than the narrow pipe of HBM. Nobody saturates shit cause they are all building around HBM.

So now it all comes full circle. Parallelism reduces memory bw pressure and thus FLOPs but increases interconnect latency pressure. Despite having HBM and GigaSERDES we aren’t actually doing more work lol. But if you really wanted to balance the pipeline you need to match the memory bandwidth, the flops, and most importantly the interconnect.

So what does that look like ? Well if you build around LPDDR’s lower bandwidth, lower your interconnect latency, you actually can beat Nvidia on decode with a fraction of the silicon.

[Chris Martz @ChrisMartzWX]

On this date in 1911, nearly 60% of the U.S. was ≥90°F and 15% was ≥100°F. This weekend's heatwave is only unprecedented if you think history began 30 years ago.

[Eric Daugherty @EricLDaugh]

🚨 NOW: In an incredible moment, US Marines are actively carrying out a Central Park NYC run with the AMERICAN FLAG

These guys are going MAXIMUM PATRIOT MODE in communist Mamdani's NYC! 🇺🇸

[🄽osliw Pilf 🦭🚩 @trvecharlie]

Oh my god. It’s him

[John Galt @john_galt_42069]

>be 19yo uni student
>get into guns, think of design fix problems with current designs
>shelve drawings because broke and scammed by uni tuition
>graduate early, spend several years as code monkey
>late 20s, jeets and MBAs destroy SWE job market
>pull out drawings from a decade ago, start making 3d models out of drawings
>design new bullpup rifle from scratch in 3 months
>design barely works, since no exp or creds in MechE
>keep fixing issues and burning through savings
>start YT to document build process, terrible prod quality, barely get any views/subs
>finally get it working, parts keep breaking
>both patents granted b4 30th birthday
>make DFM, weight, cost optimizations, reduce parts count, fix breakages, continue testing
>start open sourcing parts of rifle
>still suck at marketing/social media
>looking for MFG to license

[Spencer Pratt @spencerpratt]

Gavin Newsom's CA State Park employees destroyed the Palisades &amp; Malibu, they mocked residents in leaked texts as their homes were burning. Now, on Independence Day 250, they're hanging the American flag upside down, in full view of the burned out Malibu lots. Unreal.

[Eric Daugherty @EricLDaugh]

🚨 UPDATE: President Trump has infuriated Democrats by posting this truth nuke

"Over 273 Americans have been shot since the war in IRAN began..."

"in Chicago."

CHICAGO = FAILURE