sol

build

[sol @solkry12]

@ImSh4yy @satyanadella Jev isn't even the best Jev lol

[sol @solkry12]

@0xjohnho @0xSero Only decode is bandwidth bound; prompt processing is still compute bound. And 3090 has a TDP of 350W, also a 3090 doesn't have Blackwell's tensor cores, so you can't run NVFP4 quants correctly like you can with that 4000.

[sol @solkry12]

@CanavanCathal @HeyGenDev @HeyGen Would love to try

[sol @solkry12]

@Rabadakkus @0xSero are YOU really the one building anymore if it's a frontier model doing everything for you?

[sol @solkry12]

@pikodevx @_Systech_ This means you don't know how much your game is worth

[sol @solkry12]

@ProgressSlowly @NoLimitGains Bro you not living off of $300K a month??? Do you eat gold bars for breakfast?

[sol @solkry12]

@0xjohnho @0xSero If you throttle it to 150W you're still 80W above with TOPS and TFLOPS reduced to about 950 and 30, respectively, along with 8GB less non-ECC VRAM

[sol @solkry12]

@vegardev @atomic_chat_hq Yeah but it doesn't look 12.7x better

[sol @solkry12]

@_Oliver_Fish Wow, even Anthropic would’ve paid you more for information about this lol

[sol @solkry12]

@Johannsmitcn4s @Deived3339 @p8stie Everyone knows, that's just how LLMs work, they can't keep track of things like that. OpenAI would have to record what kind of person you are to ChatGPT and give everyone's ChatGPT session access to that data, which would land them in a whole heap of trouble.

[sol @solkry12]

@p8stie ChatGPT can't compare you to other users

[sol @solkry12]

@AiArtFactory @ronaldmannak 3.8 27B at Q4 is great, I've done a lot of work with it, I've also heard the GSQ-RCO quants are even smaller and still very usable too but I haven't bothered with them

[sol @solkry12]

@0xjohnho @0xSero Well it's supposed to be an always on/always running machine, so if you're running a 5070ti at 300w 24/7 that gets expensive quick

[sol @solkry12]

@Aibek_star @cashbychris This is probably just Opus competing with the entirety of the Fern team lol

[sol @solkry12]

@ronaldmannak Why do you say that?

[sol @solkry12]

@yzxlr11 @AGZ_TX @thsottiaux It was their harness, that was uploading code, the model itself is fine

[sol @solkry12]

@AGZ_TX @AI_Bizart @thsottiaux Why are you trying to one-shot things though, every development process is done through iterations

[sol @solkry12]

@TravisV_YT @Randomistic2838 @ClasslessMike No, on PC it's connect a wire and done

[sol @solkry12]

@TaroTheCerelean This is literally a baseless claim with no evidence behind it