ericd

7 ms·

31.
by ericd 29d ago
Nah, Taalas was putting the weights into silicon as a mask ROM. Their demo chip was hardwired to serve Llama 3.1 8B, and could never be updated. New models, even new versions without any architectural/size changes meant new tape outs.
32.
by ericd 1mo ago
The sites that do that won't be getting money from my and others' agents. Guessing that's going to become more and more of a problem for those sites.
33.
by ericd 1mo ago
Ahh yeah, I've seen people using those. I was thinking you'd ideally want to connect them full speed, but it looks like they're 200gbps ports anyway, and I guess even if they were 400gpbs, you don't really need that much
34.
by ericd 1mo ago
No, the asic could only ever run one model/set of weights, no updates possible, ever. These are general purpose processors that can have their models updated. But the chips are enormous, with a substantial amount of on-die memory along
35.
by ericd 1mo ago
Well, this one only has 4 ports, would you daisy chain them somehow to get to 8? Or bridge three of these switches?
36.
by ericd 1mo ago
Yeah, my cousin had one, was like a computer from a parallel evolution chain, kind of weird, but kind of awesome.
37.
by ericd 1mo ago
Yeah, someone reverse engineered Red Alert 2 recently. Took Fable cranking for almost a month, but it's pretty amazing to see.
38.
by ericd 1mo ago
Wow, that's gorgeous for something written in assembly, running in 512k, nice work!
39.
by ericd 1mo ago
Well, is there a clear route to being a >$10B business? If not, it's probably not a good match for VC. But that's OK.
40.
by ericd 1mo ago
Essentially, yes. The DGX Sparks, at 128 gigs for ~$4700 are one of the cheapest ways to get enough high-ish speed memory cobbled together to run one of the more capable open weight models to run at home/for a small biz. This switch le
41.
by ericd 1mo ago
Oh, nice haha, way better. And no worries, I'm actually waiting on a quote/timeline from them anyway.
42.
by ericd 1mo ago
Ah yeah, one of mine is from Central. And yeah, crazy how much they've gone up. But I can see why, they scream.
43.
by ericd 1mo ago
Ah thanks for the solid info, too bad. I'd seen them come up as a pretty good price for 6000 RTX's in the past, which seem generally pretty available, good source for those?
44.
by ericd 1mo ago
Ah gotcha. Have you tried to order something like this in the past?
45.
by ericd 1mo ago
That's not apples to apples on almost any dimension.
46.
by ericd 1mo ago
What're you using that monster for?
47.
by ericd 1mo ago
Ha fair, I'd definitely confirm with a salesperson before wiring them $300k. But most of the signs on the configurator seem to point to it including the GPUs? Not going to make 30k BTUs/hr of heat without the 8kw of GPUs.
48.
by ericd 1mo ago
>You couldn’t buy one of these if you wanted to right now. You can: https://www.exxactcorp.com/Exxact-TS4-149591758-E149591758 . You can get thousands of tps of GLM 5.3 output out of this thing, which grades around Opus
49.
by ericd 1mo ago
Looks like GLM 5.2 is coming in at <2x the tokens of Opus 4.8 (and 1/10 the cost)? Great showing from Sol, though. But also, it's Baba Is You :-D
50.
by ericd 1mo ago
Not sure, I haven't run it, I've just been running DS V4 Flash non-stop since it came out, and that's replaced a lot of my Claude Code usage. People seem very impressed, though, it seems like it trades vram/world knowled
51.
by ericd 1mo ago
I honestly wouldn’t bother with local models right now unless I either had a 5090 and was happy with running Qwen 3.8 27B, or a pair of DGX Sparks running DSv4 flash, or better, 2x6000 RTX Blackwells. Those are the kinds of rigs that the lo
52.
by ericd 1mo ago
I have something like this persistent in my waybar, with session, weekly, and fable usage, with little color coded percentage bars for each. Super helpful, highly recommend it. Great idea on the hook alert. I feel like the advent of LLMs ha
53.
by ericd 1mo ago
Well I never expected it to be treated exactly the same as their first party customers, did you? Especially if I’m paying a much lower rate. Just have to evaluate the actual results, and hop if you’re unhappy. I’ve been extremely happy with
54.
by ericd 1mo ago
Assets, mind share, etc. All the things that get sold off in bankruptcy proceedings. A railroad company also has a huge amount of steel roads with cross country right of ways. You could fire everyone, and it'd still be extremely valuab
55.
by ericd 1mo ago
It’s a core part of the open model infrastructure. Their hardware business benefits greatly if this works really well and open models proliferate to every corner of the economy, with everyone buying GPUs with a lower capacity factor/du
56.
by ericd 1mo ago
I really, really doubt that, and I'm a bit surprised (not too surprised, HN has gotten really cynical) that this is the top comment. I think the future nvidia most doesn't want is a small number of closed labs that run away with i
57.
by ericd 1mo ago
Except in this case, the gun is the one directly improving the body armor.
58.
by ericd 1mo ago
That's kind of a broad brush, Google Fi is great, especially the fact that they work mostly seamlessly for the same price in like every country, and they're an MVNO.
59.
by ericd 1mo ago
Great followup interview, too: https://www.youtube.com/watch?v=AJCWsoozg0Q
60.
by ericd 1mo ago
Haha right, well, usable speed. And I guess there are some parallels with internet, the internet is technically usable with HughesNet, but people used to fiber would probably consider it unusable.

More ›