The most prestigious law school admissions discussion board in the world.
|
 |
 
|
NSAM's built the AI from hell with some Nvidia GPUs, go ahead and doubt it
Poast new message in this thread
Date: February 26th, 2026 11:52 PM Author: Jet-lagged stead (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698772) |
Date: February 27th, 2026 12:20 AM Author: green mind-boggling step-uncle's house specs and config (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698808) |
 |
Date: February 27th, 2026 12:21 AM Author: Jet-lagged stead 2x 3090s, a 5090, and a 5060 ti + i5-14400 with 128gb DDR5. llama.cpp in a Debian 12 container (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698810) |
|
Date: February 27th, 2026 12:22 AM Author: green mind-boggling step-uncle's house what mb. how so many pcie lanes? (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698811) |
|
Date: February 27th, 2026 12:27 AM Author: Jet-lagged stead Gigabyte z790 with PCIe bifurcation on the top PCIe 5.0 slot, plus two x16-size PCIe 4.0x4 slots on the mobo. I think all the Gigabyte z790 motherboards give you lanes out the ass on the slots, even the budget series (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698814) |
|
Date: February 27th, 2026 12:32 AM Author: green mind-boggling step-uncle's house damn. what is total gpu vram? does llama see it all? how many parameter u can run?
edit, wait u run 5090 in 4.0 pcie slot? |
|
Date: February 27th, 2026 12:43 AM Author: Jet-lagged stead The PCIe 5.0 slot is split into x8x8, and the 5090 only uses 5.0x8. But it doesn't matter because you're never going to be limited by bandwidth until you drop below PCIe 3.0. It's just not an issue because the GPUs aren't sending/receiving that much data to begin with. I rarely see any GPU spike over 900 MiB/s in nvtop.
By far the biggest difference is Blackwell vs non-Blackwell, but it doesn't matter to me because I have multiple Debian containers with different GPU passthrough configs. So if I want to load big 70b models on the 3090s, and I just need another 8-12gb of VRAM, I can put the 5060ti in that container and give it the extra 16gb. Right now that's what I'm doing because the 5090 runs so well by itself. But I can also move the 5060 ti to that container if I need more than 32gb and I want to keep Blackwell features. And of course I can put all four in one container for 96gb, though I've seen no need to do that so far. Deepseek 4 is a wildcard, I have no idea what to expect |
|
Date: February 27th, 2026 12:50 AM Author: green mind-boggling step-uncle's house so 5090 and 5060 split 5.0 8x8? and the 3090s running on 4.0 lanes? does llama see aggregate vram or you running containers that can only see portion of total vram? i am confused.
edit, saw your last 2 sentences got it. damn just slam everything into one container and see what you can do. |
|
Date: February 27th, 2026 1:07 AM Author: Jet-lagged stead The way Blackwell does KV offloading is black magic. The 5090 by itself will run 48gb models no problem. It just populates the VRAM and then it only populates <1gb of system RAM. I have no idea how to account for the missing 16gb. How can a 48gb model only use 32gb of VRAM and no system RAM?
|
|
Date: February 27th, 2026 1:13 AM Author: green mind-boggling step-uncle's house ok fuck it. im buying a 5090 tmr. only running 4090 right now.
edit, ive been trying to snipe a 5090 FE, no luck. i will just get whatever like an asus. |
|
Date: February 27th, 2026 1:28 AM Author: Jet-lagged stead the MSIs are really good. I have the Gaming X Trio, which is in between the Ventus and the Suprim. The FEs are considered inferior and more likely to overheat or malfunction. Some Asus cards have fit issues with the power connectors too, need to research that (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698855) |
|
Date: February 27th, 2026 1:47 AM Author: green mind-boggling step-uncle's house (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698869) |
|
Date: February 27th, 2026 1:33 AM Author: Jet-lagged stead also this magic KV offloading requires llama.cpp, and only works with gguf files. vLLM and SGLang won't do it. Ollama will do it but it's literally 1/10 the speed of llama.cpp with Blackwell (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698862) |
|
Date: February 27th, 2026 10:11 AM Author: Stirring dark whorehouse puppy Buy an RTX Pro 6000. It's far better than dual 5090s. (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49699206) |
|
Date: February 27th, 2026 1:06 PM Author: Jet-lagged stead Oh wow my stalker can use Google! (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49699699) |
|
Date: February 27th, 2026 4:04 PM Author: green mind-boggling step-uncle's house i would and agree, but i dont want to dump that much $ into my llm, which is only a hobby. i am willing to wait out the inevitable price drops on nvidias enterprise stuff during ai data center upgrade cycles. we are just at the very beginning of this. (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49700462) |
|
Date: March 8th, 2026 6:16 PM Author: Jared Baumeister Having 96gb on one GPU would suck for me because I can't divvy it up and run separate models in separate containers. Right now I can put 96gb in one container, but I can also use one GPU to do vision/OCR in Ollama while the others run Qwen3.5. If all my VRAM was on one GPU I could maybe load two models at once, but that's glitchy and it requires me to run everything in one container. (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49726643) |
|
Date: February 27th, 2026 2:54 AM Author: tantric church building how important are the extra GPUs? am i gonna be happy with the performance of a local model if i set up a 5090? (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49698903) |
|
Date: February 27th, 2026 7:31 AM Author: Jet-lagged stead No idea. I also haven't probed the limits of the 5090 yet. All I will say that I was initially disappointed with its performance in Ollama, and I didn't see big gains until I started using llama.cpp. You HAVE to compile llama.cpp and let it build whatever features it thinks it needs into the binaries (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49699009) |
|
Date: February 27th, 2026 9:28 AM Author: Jet-lagged stead PS if you're using llama.cpp you can ask Claude how to tune the parameters to your particular situation.
You should also phave all drivers installed before you compile llama.cpp, so that it detects and installs the right modules
Also, if you have a mix of GPUs with different amounts of VRAM, you have tell llama.cpp how many layers to offload to each one and how many layers (if any) go to the CPU. It's such a grind that I'm making a spreadsheet of scripts for launching different models in different configurations
This is the PCIe bifurcation card I use. Even though it says it's only 4.0, it shows up as 5.0x8 in nvtop
Finally, all of these GPUs can be drastically power limited so that they run off one PSU. The 3080s can be run at 200W (400Wx combined), the 5060 uses 170W so you're only at 370W, then power limit the 5090 to 380W. So total draw is only 750W (and my CPU can only pull 65W, so it's not a problem to run them on one PSU. I'm using a 1600W PSU but I could probably get by with 1000W. Performance isn't an issue unless you're gaming, and cooler temps extend longevity |
|
Date: February 27th, 2026 4:07 PM Author: green mind-boggling step-uncle's house that card is 1/3 the price if you buy from straight from china/aliexpress. (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49700478) |
|
Date: February 27th, 2026 6:15 PM Author: tantric church building I'm going to be building a local model soon so all of these tips are helpful. (http://www.autoadmit.com/thread.php?thread_id=5838844&forum_id=2most#49700842) |
|
Date: March 8th, 2026 5:49 PM Author: Jared Baumeister cot damn, so here's what I got going:
1. Qwen3.5 122b with the context window blown out. I mean just totally raped to death. If it gets too big it goes into system RAM, but it never gets slow.
2. Gemma3:27b on a separate server for image recognition and OCR
3. some shit called "whisper" running on an Intel ARC a380 with only 6gb of VRAM. Somehow this fuckin thing can TRANSLATE VIDEOS FROM ANY LANGUAGE and generate a transcript.
All of these run simultaneously and can talk to each other no problem. Each model is running on a different server. I can combine the Gemma and Whisper servers onto one server but what's the point? |
|
Date: March 8th, 2026 6:06 PM Author: Jared Baumeister Hello dear viewers, when people say that I only build military vehicles, this is the north-eastern part of Tehran and the eastern part of Tehran. When they build a petrol station, they want people to be annoyed, they don't want military work, they want people to build petrol stations so that your cars and ambulances can easily run in the city. This place has a very close distance to the offices of the authorities It is a very crowded place for people and houses This is not the way I just make a mess This is not the way I tell you in the media This place is for me and you It is the fuel for me and you If we show you the offices of the authorities This is the office of the authorities This is not the place to make a mess There is no military here, but we are all concerned about the fuel and the fire-fighting vehicles I hope we all help each other to get through this storm You just didn't pay attention to the foreign media who say I'm going to the military There is no military here, people, there is no military here This is the city of Tehran, which has been targeted. Let's go.
https://x.com/MarioNawfal/status/2030747330883056004?s=20
https://i.imgur.com/lgIC68V.png |
|
Date: March 8th, 2026 6:12 PM Author: Jared Baumeister 1 00:00:00,000 --> 00:00:03,960 My next guest was used also in worshipping the devil
2 00:00:04,480 --> 00:00:12,040 Participated in human sacrifice rituals rituals and cannibalism. She says her family has been involved in rituals for generations
3 00:00:12,040 --> 00:00:16,280 She is currently in extensive therapy suffers from multiple personality disorder
4 00:00:16,720 --> 00:00:20,240 Meaning she's blocked out many of the terrifying and painful memories of her childhood
5 00:00:20,240 --> 00:00:26,480 He Rachel who is also in disguise to protect her identity you come from generations of ritualistic abuse
6 00:00:26,480 --> 00:00:31,760 Oh, yes. My family has an extensive family tree and they keep track of who's been involved
7 00:00:31,760 --> 00:00:40,000 and who hasn't been involved. And it's gone back to, like, 1700.
8 00:00:40,000 --> 00:00:41,000 And so you were?
9 00:00:41,000 --> 00:00:47,120 Right. I was born into a family that believes in this.
10 00:00:47,120 --> 00:00:51,320 And this is a, this is, does everyone else think it's a nice Jewish family? From the
11 00:00:51,320 --> 00:00:53,840 outside you appear to be a nice Jewish girl?
12 00:00:53,840 --> 00:00:54,840 Definitely.
13 00:00:54,840 --> 00:00:59,500 you all are worshipping the devil inside the home? Right. There are other Jewish families
14 00:00:59,500 --> 00:01:06,980 across the country. It's not just my own family. Really? And so who knows about it? Lots of
15 00:01:06,980 --> 00:01:17,080 people now. I talked to a police detective in the Chicago area and several of my friends
16 00:01:17,080 --> 00:01:24,880 know and I've spoke publicly before and... So when you were brought up in this this
17 00:01:24,880 --> 00:01:30,560 kind of evilness did you just think it was normal? I've blacked out a lot of the
18 00:01:30,560 --> 00:01:36,880 memories I had because of my multiple personality disorder but yes I mean it's
19 00:01:36,880 --> 00:01:41,360 like if you go off with something you think it's normal. I always thought... So
20 00:01:41,360 --> 00:01:44,480 what kinds of things? You don't have to give us the gory details but what kinds
21 00:01:44,480 --> 00:01:47,480 What kinds of things went on in the family?
22 00:01:47,480 --> 00:01:53,480 Well, there would be rituals in which babies would be sacrificed and you would have to, you know...
23 00:01:53,480 --> 00:01:54,480 Who's babies?
24 00:01:54,480 --> 00:01:59,480 There were people who bred babies in our family. No one would know about it.
25 00:01:59,480 --> 00:02:03,480 A lot of people were overweight so you couldn't tell if they were pregnant or not.
26 00:02:03,480 --> 00:02:07,480 Or they would supposedly go away for a while and then come back.
27 00:02:07,480 --> 00:02:11,480 The other thing I want to point out, not all Jewish people sacrifice babies, I mean.
28 00:02:11,480 --> 00:02:12,480 No, no.
29 00:02:12,480 --> 00:02:18,640 This is the first time I've heard of any Jewish people sacrificing babies, but anyway.
30 00:02:18,640 --> 00:02:21,040 So you witnessed the sacrifice?
31 00:02:21,040 --> 00:02:22,040 Right.
32 00:02:22,040 --> 00:02:29,480 When I was very young, I was forced to participate in that, in which I had to sacrifice an infant.
33 00:02:29,480 --> 00:02:31,560 And the purpose of sacrifice is to what?
34 00:02:31,560 --> 00:02:32,560 Is to bring you what?
35 00:02:32,560 --> 00:02:33,960 What are you sacrificing for?
36 00:02:33,960 --> 00:02:36,960 For power.
37 00:02:36,960 --> 00:02:40,560 Power.
38 00:02:40,560 --> 00:02:45,240 And so you were ever used, were you ever used yourself?
39 00:02:45,240 --> 00:02:49,840 I was molested, I was raped several times.
40 00:02:49,840 --> 00:02:54,960 What's your mother doing in all of this?
41 00:02:54,960 --> 00:02:55,960 What's her role in all of this?
42 00:02:55,960 --> 00:03:01,480 I'm not exactly what her role is, I haven't recovered all of my memories, but her family
43 00:03:01,480 --> 00:03:04,800 was extremely involved.
44 00:03:04,800 --> 00:03:08,960 She brought me to it, both of my parents brought me to it.
45 00:03:08,960 --> 00:03:11,360 Where is she now?
46 00:03:11,360 --> 00:03:15,720 She lives in the Chicago metropolitan area.
47 00:03:15,720 --> 00:03:18,100 She's on the Human Relations Commission of the town
48 00:03:18,100 --> 00:03:19,480 that she lives in.
49 00:03:19,480 --> 00:03:20,880 And she's an upstanding citizen.
50 00:03:23,960 --> 00:03:25,800 Nobody would suspect her.
51 00:03:25,800 --> 00:03:28,680 Were you raised with a sense of right and wrong, Rachel?
52 00:03:28,680 --> 00:03:29,400 Yeah.
53 00:03:29,400 --> 00:03:31,400 I mean, it's like I had both.
54 00:03:31,400 --> 00:03:34,480 I mean, to the outside world, everything we did
55 00:03:34,480 --> 00:03:36,120 was proper and right.
56 00:03:36,120 --> 00:03:39,680 And then there were the nights that things changed,
57 00:03:39,680 --> 00:03:41,280 that things just got turned around.
58 00:03:41,280 --> 00:03:46,040 What was wrong was right, and what was right was wrong.
59 00:03:46,040 --> 00:03:51,280 That's what helps to create somebody to develop MPD.
60 00:03:51,280 --> 00:03:53,080 Multiple personalities.
61 00:03:53,080 --> 00:03:55,240 Now, in your family, did you all call it
62 00:03:55,240 --> 00:03:56,840 worshipping the devil?
63 00:03:56,840 --> 00:03:57,400 I don't know.
64 00:03:57,400 --> 00:03:59,440 It was just evil, the things you did.
65 00:03:59,440 --> 00:04:00,040 Right.
66 00:04:00,040 --> 00:04:04,440 Well, I said it was evil, and they said it was good.
67 00:04:04,440 --> 00:04:09,080 There's a book that I had just come across called Lilith Cave, which is a book of Jewish
68 00:04:09,080 --> 00:04:17,080 mysticism and supernatural, and there's a lot in that book that relates to what I, you
69 00:04:17,080 --> 00:04:19,400 know, endured when I was a child.
https://x.com/redpillb0t/status/2022335766710641103?s=20 |
|
Date: March 8th, 2026 7:16 PM Author: Jared Baumeister NSAM's AI translates Farsi:
Do you know that this slogan of yours, your death to America, this cry that the people of Iran make, these are the roots and roots of logic and power. These are the roots and roots of reason. It is also clear that death to America is not death to the people of America. The people of America are like the rest of the people. That is, death to the policies of America, death to the establishment of a new regime. This is what it means. These are the behind the scenes of logic. Our fundamental law is based on this logic. Alhamdulillah, the Islamic Republic has opened its own path and is moving forward. I have no doubt that you, dear youths, have seen the days when many of the high hopes that the Islamic Republic has built up in your country, in your life, have been fulfilled. Thank you very much.
https://x.com/SuppressedNws1/status/2030771902952644735?s=20 |
|
Date: March 8th, 2026 7:20 PM Author: Jared Baumeister Every day, we are coming, we are coming, we are coming to Gaza, we are coming to Lebanon, we are coming to Iran, we are coming to everywhere. Think about how much we are going to kill you and how much we are going to expose you to each of the 1,300 people that you killed, that you killed. You haven't seen numbers like these in the newspapers of the Arab world. I assure you that it's coming. In case you're confused, I assure you that it's coming. Numbers that you didn't even imagine that it's possible. It's possible to get to them. And we're ready to enter international unity, and we're ready to fight with the United States, and we're ready to fight with the whole world and its brothers. How long will it take until all of you, including all your supporters, Go up to meet Allah and we will kill him. Let it be clear. Let it be clear. This is the situation in Israel. This is the sentiment. So log on to social media. Make Palestine free. Make all your misery. And we will come to destroy you. To destroy. To destroy. Translate it. Upload this video. Let all your friends see. What are we standing for?
https://x.com/Jvnior/status/2030722198822891526?s=20 |