soloish

The Vibecoder’s Serenity Prayer God, grant me the serenity to find the bugs I did not write, the courage to delete the code that “should work,” and the wisdom to know when Claude Code is gaslighting me.

I see… yeah, two gpu setups pretty much just like a normal setup… fits in a normal case. More than that… I’m not experienced either. I even suspect with my 2x 3090 setup I still may need to deal with heat a little better… maybe a gpu riser + pcie extended… just so the gpus aren’t packed to closely.. I’m about to find out. Honestly I see most people just using a simple open case style setup… probably because of cooling needs. I haven’t gone there yet.

@soloish90 Thanks, yeah searching on Facebook marketplace seems to be a good bet for finding deals. I'm more curious about steps for building the multi-GPU setups that wouldn't fit in any traditional desktop tower or rack server

I wish the “get an RTX 3090” recommendations were usually followed up with “and the rest of the components, RAM, etc will cost an additional <abc> & can be bought at <xyz>” because I have no earthly idea where to start & how to build those jungle gym looking setups

@keennay * US dollars

I bought on Facebook marketplace… just had to take the time for the right deals to show up: 2x 3090 - one in a full system - ($1100, which i used everything but motherboard.. use case, cooling, cpu i7-12700F, 2TB m2 drive) - one for $700 Bought a motherboard that supports 2 gpus at pcie 8x/8x speeds - most consumer boards don’t do this.. second gpu pcie speed will only be 4x bandwidth. In my case for the i7 processor - asus pro art z790-creator Wi-Fi. Because of the AI market - hard to find for less than $500 but I found one on Facebook for $250. Bought 64gb ddr5 ram for $450 (again facebook because brand new that would be more like $850) Had to bump up the PSU (support 1200 watts). Can’t remember price but not expensive

It’s true. Takes a little patience… there will be many listed at high prices ($1300+) but I just kept looking daily - found one for $700 and one for $900… all very close to me. Also found 64gb ddr5 at half the new prices, plus the perfect motherboard (mostly $500+ everywhere else - bought for $200) Everything working and completely tested… perfect

many ask me where to find a rtx 3090 for $900 or $1,200 in 2026. the one place i suggest you look is facebook marketplace. nothing else comes close, ebay sits 30% above and retail is fantasy.

I think you might want a motherboard that does a x8/x8 split… more specialized. I have an older one for a older cpu I’m building around: ASUS ProArt Z790-Creator WiFi So, not a standard “gaming” or consumer board - specifically made for dual gpu setup Mine board is arriving tomorrow…

Got asked to test if PCIe is bottlenecking my dual 3090 setup First card on PCIe gen4 x16 (direct CPU), second on gen3 x4 (through B760 chipset) Ran nccl tests: - PCIe gen3 x4 theoretical: 3.94 GB/s - all_reduce bandwidth: 0.52 GB/s 😂 13% of theoretical lol Second card routes through the chipset, this is why TP=2 gives me the same speed as single GPU I'm ordering NVLink today

Vibe coding tips for jobs #1: Whenever I’m asked how long a code task will take at work, I run it by Claude first, and handoff the estimate. 1 hour of actual time staring at my computer while Claude works, 3 weeks of extra time for other things.

@Plinz I just started Revelation space yesterday… synchronicity

Alastair Reynold's "Revelation Space" universe is certainly not the hardest SF out there, but it's hard to beat his scope. He takes the largest ideas in SF and zooms so far out that they become tiny details in his stories.

@trq212 @om_patel5

@om_patel5 ugh sorry this was a bug with the 3rd party harness detection and how we pull git status into the system prompt we're reaching out to affected users and giving them a refund + another month of credits (in this case another $200)

THIS GUY LOST $200 IN ONE DAY BECAUSE THE STRING "HERMES.md" WAS IN HIS GIT COMMITS HERMES.md is a real convention used in AI agent projects. it's a system prompt specification file. not some obscure edge case he's on claude max 20x at $200 a month. yesterday claude code hit him with "you're out of extra usage" out of nowhere his dashboard showed 13% weekly usage. 0% current session. 86% of his plan was sitting there untouched but $200.98 in extra usage already burned through what should have been covered by his subscription he tried logout & login, different models, fresh installs and nothing worked anthropic support sent the ai bot (four rounds of the same scripted response). eventually they just gave up on him so he started binary searching repos and commits manually on his own time until he found the trigger the string "HERMES.md" in a recent git commit message uppercase, with the .md extension, anywhere in your commit history that's it claude code includes recent commits in its system prompt and something server side flags HERMES.md and quietly routes you off your max plan onto API rate billing > AGENTS.md? fine > README.md? fine > HERMES without .md? fine > lowercase hermes.md? fine > uppercase HERMES.md? you're getting charged API rates he reported it. anthropic support acknowledged the bug three times, called it an "authentication routing issue", thanked him for finding it then refused to refund the $200 so the man pays $200 a month for max, lost another $200 to a billing bug they confirmed, did anthropic's QA work for free on his weekend, and got a "thank you for your patience" in return check your commit history before claude code quietly drains your account too

@FrameworkPuter According to my profile, I’m an influencer of 4. It’s still influence Dell, don’t argue.

Every time we engage with an influencer on X dot com, Dell sends them an XPS. Anyone want a free Dell XPS?

@outsource_ @grok Right 👍🏻

@soloish90 @grok Need compute

Hey @grok rate this build I found / the price

@outsource_ @grok I’m actually in the middle of pulling together a comparable 2x 3090 machine…not sure what the total will be - close to or just under. But it would have saved me a lot of driving and searching just to get in all one stop. Worth it.

@soloish90 @grok 100%

I want to try vllm-studio soon. I’m putting together a dual 3090 system and figure this might be a fun try. Any recommendations for the system? Much to think about - in the middle of ordering missing hardware for a full system Also, I’m thinking vllm + tensor parallelism for the 2 cards - but completely new to this kind of setup. Best way to go with software? (Linux)

I need designer friends to tell me what I'm doing wrong.

@thangel3535 Is this the new ChatGPT Images 2.0? So good… 😂

~2hr sketch while listening to nerd snipe great podcast

@Truthcoin Idiot

BREAKING: New Bitcoin Fork I am helping create a **new Bitcoin Hardfork** -- dropping this August, called "eCash". - Your coins will split. For example, if you have 4.19 BTC, then you will get 4.19 eCash. - You may sell your eCash -- or keep it. Or ignore it! Vegas: - Yes, I will be in Vegas next week. - No, I won't mention this, on stage -- (that would be rude). Our L1 Node... - is a near-copy of Bitcoin Core. - is Sha256d mined. - forks via a one-time difficulty-reset -- to its minimum value. (So, mining will be crazy at first.) - Yes, we will change the seed nodes, the name, the network magic, etc. Codewise, the L1 will remain compatible with Bitcoin Core: - We will continue to merge their changes (even the bad ones). - The L1 will activate Bip300/301 via CUSF -- the core untouched soft fork. So, no lines of code will be changed, on the L1. - The activation client will be published periodically (link below). - We will do several bug bounty contests this summer. - The client will be frozen 30 days prior to the fork. Yes, there will be Drivechains: - We have 7 in developement right now. - Users can also submit their own. - Drivechain is a vision of "competing L2s" -- this avoids the "dev capture" problem. - These L2s are all Merged Mined. Miners automatically get free $. - Our L2s are already capable of planetary scale, and onboarding 8 billion users. - We also have a zCash-like L2, with strong privacy. - Other L2s: Truthcoin (Prediction Markets), CoinShift (Decentralized Exchange), BitAssets (NFT etc), BitNames (Identity), Photon (Quantum Resistant). Unlike BCH (the 2017 fork): - There is no "Bitcoin" in the name. New name, new brand. - You are getting advanced warning (4 months). - We are replaying all txns (at first). We will release a coin-splitter tool. - This is a permanent & sustainable fix to Bitcoin's problems (instead of a 1 MB to 8 MB temporary fix). - Back in 2017, the BTC tech stack was strong, and expectations for Lightning were strong. Today, it is the reverse. Video to follow.

@0xSero Need it

What do you guys think? I'm trying to make something to make all our benchmarks, speeds, recipes, hardware, etc.. data available. It's a really hard problem because there's so many details involved and variables to consider. Built on HuggingFace & Ideally Hermes.

@0xSero Llamacpp rpc?

On bad WIFI I split Qwen3.5-122B between a single 3090, and a Framework desktop. I got 17 tokens/s decode and 95 tokens/s I also didn't optimise the split, could probably 2x the speed of the Strix Halo for MoEs. BRR

Ran Qwen3.6-35B-A3B Strix GGUF benchmarks on Framework Desktop / Strix Halo. Build: llama.cpp b8838 Settings: ngl=99, fa=1, ctk=q8_0, ctv=q8_0, ub=2048, b=2048, p=4096, n=128 ROCm Q4_K_M: pp4096 = 1380.52 ± 16.54 t/s tg128 = 58.41 ± 0.09 t/s ROCm DYNAMIC: pp4096 = 1440.36 ± 12.78 t/s tg128 = 53.00 ± 0.17 t/s Vulkan Q4_K_M: pp4096 = 1053.75 ± 4.17 t/s tg128 = 71.15 ± 0.14 t/s Vulkan DYNAMIC: pp4096 = 931.22 ± 3.07 t/s tg128 = 64.68 ± 0.24 t/s Takeaway on my box: ROCm wins prefill Vulkan wins decode DYNAMIC helps prefill on ROCm Q4_K_M is the better Vulkan variant here

Would love to have someone test this. I tried to find the best mix for GGUF, could be awful tho. huggingface.co/0xSero/Qwen3.6…

@0xSero i was literally just about to spin up a different Qwen3.6-35B-A3B on my Framework. i will try this instead...