Sabitlenmiş Tweet
so i lowballed 100s of luxury watch dealers in the USA using a voice AI and managed to negotiate $15k off MSRP for a Daytona
youtube.com/watch?v=ohQrh3…
YouTube
English
Tony Ge
380 posts
this'll be a wakeup call for the rest of big inference
Google for Developers@googledevs
Breaking LLM inference’s autoregressive bottleneck 🛠️ We've teamed up with @haozhangml, @YimingBob, and @aaronzhfeng, among others from UCSD to achieve a massive 3.13X speedup for LLM inference on Google Cloud TPUs using Diffusion-Style Speculative Decoding (DFlash). Read the blog: goo.gle/4naZ8Yv
English
@masterzorgon bro if you're willing to fund a run for your own campaign using this tech i'll do it just for the data
English
so uh i actually built this over the weekend... can someone fund my ad account 🥺
if it works i'll give you the code
Tony Ge@largePrawn
Meta ads CLI + auto researcher + GPT-2 image + $1000 ad spend would probably print
English
@storagereview @DellTech @NVIDIAAI whoa so cool of you guys to be announcing your 4x rtx 6000 pro giveaway next and only guys named tony are eligible to win
English
We now have four RTX Pro 6000 GPUs for the XE7740 server. Kevin walks through the bracket install and a few other behind-the-scenes details.
#servers #datacenter #gpu @DellTech @NVIDIAAI
English
X has this beautiful ability to make me feel like everything i do is utterly useless somedays 🫠
English
@shanneykar @malikwas1f 96gb DDR4 ECC RAM
1TB NVME
threadripper 5XXX (cant remember)
It's an old Lenovo P620 some startup liquidated for $1100 (3090s not included)
English
@largePrawn @malikwas1f Do you mind sharing rest of your AI workstation rig that compliments the RTX3090s
English
Hitting 140 tok/s on Qwen 3.6 27B
running vLLM with 2x 3090s using the following @malikwas1f's repo
github.com/noonghunna/clu…
Literally just pointed claude at it and walked away. Came back to a 2.5x speed bump 🤯🤯🤯
English
English
@MaxKerkula @malikwas1f Earlier i couldn't be fucked to figure out when the next available labcorp appt was bc i needed to get a blood test so i booted up hermes and used qwen to power it
English
@jarekceborski Latency and reliability. No random outages + near instant responses (when models are warm)
English
@keennay iirc there's a way to RPC across any # of machines so you'd buy 2x of these (assuming each box has 4x PCIe 16x slots) and then run link them over network via vLLM
youtube.com/watch?v=1lbBrT…
YouTube
English
@largePrawn what are the options here for >= 8 GPUs, aside from enterprise rack servers by Supermicro / Dell / Gigabyte / ASUS
English
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
Sandro@pupposandro
respect the take but I think this is incorrect. instead, I recommend people to get a 3090: durability: 3090s run for years with basic maintenance. swap the gddr6x thermal pads every couple years and the card is basically immortal. just don't buy one that lived its life in a mining farm and you're fine power: 3090 pulls 350W. 4090 is 450W, 5090 is 575W, 7900 XTX ~355W. the 3090 is the LEAST power hungry card in its bandwidth tier. wrong completely !! heat: same story. every card at this performance level runs similar temps. not a 3090 problem price: used 3090s are $800-1000 today. launch was $1499. that's nowhere near launch price, and it's still the best $/GB-vram and $/GB-bandwidth on the market by a wide margin. 4090 is $2,500. 5090 is $3,500 the real recommendation depends on your use case. need more than 24GB of vram? skip the 3090 and go for a Mac Studio / DGX / Strix Halo. running qwen3 27b at high throughput on a budget? absolute best card on the market, no contest! ideally? you get both
English
if you pointed 1T frontier toks at openclaw with just the prompt "make it better", what are the odds it just recreates hermes
David Ondrej@DavidOndrej1
openclaw is super bloated updating takes forever restarting gateway takes forever this is not what it used to be like...
English
@largePrawn Unfortunately, you’re overqualified. We need a principal slop orchestrator
English
