Custom Wetware

17.1K posts

Custom Wetware banner
Custom Wetware

Custom Wetware

@CustomWetware

wen post-scarcity utopia? math.exp(random.uniform(-5, 2)) x engineer

Chaotic Neutral Katılım Haziran 2012
568 Takip Edilen371 Takipçiler
Sabitlenmiş Tweet
Custom Wetware
Custom Wetware@CustomWetware·
@realGeorgeHotz 1. Read The C Programming Language by K&R 2. Watch everything by Andrej Karpathy on youtube 3. Write code 4. Ask ChatGPT to explain stuff to you 5. Get familiar with debugging and profiling tools 6. Use git (do steps 3 to 6 while you do steps 1 and 2)
English
3
2
38
5.2K
Custom Wetware
Custom Wetware@CustomWetware·
@leadlagreport If you bought chips stocks the bull market has been legendary for the past year
English
0
0
0
1
Michael A. Gayed, CFA
Michael A. Gayed, CFA@leadlagreport·
After inflation, the vast majority of the stock market has negative real returns following the 2021 nominal peak in small-caps. But tell me more about the bull market.
Brandon Ho@brandontho84

@leadlagreport Only 50 companies making new high for the past 10 years. While the other 2000+ companies are stuck going no where. Meaning those who buy and hold those individual stocks that didn't make new high long term have been losing to inflation.

English
10
8
64
8.9K
Derek
Derek@Derek19603617·
Being right too early is the same as being wrong. When I was bearish on the economy in 2023, and the market was falling, one day I heard Tom Lee say, (paraphrasing) the government pushing so much stimulus (deficit spending) it would be hard to have recession. I thought he was wrong…but I was wrong, because in the past (which I used as reference) no administration was ever dumb enough to run massive deficits while unemployment was near cycle lows…which Biden did. AI spend is a huge bonus. Trump is continuing this and running a $1.9T deficit. Subtract the interest, and it’s about $900B stimulus. AI spend is estimated to be $400-500B. That’s at least $1.3T stimulus. Yes, it’s adding to inflation Yes, it will cause the wealth gap to increase Yes, debt WILL eventually be a huge problem. It may be soon…but I thought that the last few years and have been wrong. When it becomes an issue, it will be too late and yes, we’d really be fucked!!! But for now, we have to buy the dip because inflation will erode buying power if we’re not.
English
1
0
2
209
Michael A. Gayed, CFA
Michael A. Gayed, CFA@leadlagreport·
The Lehman Brothers crash of 2008 erased nearly a decade of stock market returns in just two weeks. The 1987 stock market crash erased 9 months of huge gains in a single day. Frequency means jackshit. You only need to be right once and with magnitude.
Bill Wrightmouse@BillWrightmouse

@leadlagreport You’ve been saying we’re fucked for years. Eventually, you’ll be right but how many times have you been wrong?

English
33
14
373
44.5K
Talant
Talant@talantfund·
I knew this was going to happen. That's why I sold at $120
Talant tweet media
English
167
233
9.5K
591.7K
Michael A. Gayed, CFA
Michael A. Gayed, CFA@leadlagreport·
You sure they won’t crash stocks to save bonds?
English
157
38
1.1K
107.8K
Carlo Ferlauto
Carlo Ferlauto@CarloFerlauto·
@adamtaggart @KobeissiLetter It’s not the bug I’m worried about. It’s a water fowl. Yen-carry trade needs to unwind in an orderly fashion or the defecation is going to contact the rotary oscillator. Beware the Dark Bird…
Carlo Ferlauto tweet media
English
2
0
10
334
Custom Wetware
Custom Wetware@CustomWetware·
GPUs became very scarce in 2021 and prices shot up due to cryptocurrency mining, they recovered a little in 2022 when Ethereum switched to proof of stake but by then the AI boom was already happening so the prices never returned to sane levels. I think the memory shortage will last even longer because I don't see any reason why the demand growth would slow down. Whatever supply can be added will continue to be bought for many more years.
English
0
1
1
238
1sthand
1sthand@1sthand2·
@CustomWetware @wccftech Doubling of exports in 3/4 years is significant. And that is for something that requires a lot of sales support. Ram? Easy
English
1
0
1
135
Wccftech
Wccftech@wccftech·
Ex-Samsung chip boss says heavy investment by China in the memory market could crush the 414% DDR5 price spike within a year. 🔗 wccf.tech/1kg52
Wccftech tweet mediaWccftech tweet media
English
125
414
3.7K
611.9K
Custom Wetware
Custom Wetware@CustomWetware·
@jukan05 "Absence" of NVLink and HBM? Why would it have those when RTX PRO 6000 doesn't? Shit article.
English
0
0
0
212
Jukan
Jukan@jukan05·
This latest piece from GSR is genuinely important. Why is Nvidia still courting China — almost clinging to it? GSR frames it this way: "For Nvidia, China is a monitoring window — a way to observe how massive AI infrastructure is actually evolving." Nvidia's revenue exposure to China has collapsed from roughly 26% historically to around 5% today. And yet the reason Nvidia cannot easily walk away from China is not about revenue. China's domestic AI chip competition is already fierce. Huawei Ascend, DeepSeek, Alibaba — China's homegrown AI ecosystem is no longer in the "chasing Western technology" phase. Under the pressure of sanctions, it has moved into genuine architectural experimentation and system-level optimization. In other words, China is no longer simply a sales market for Nvidia. It is the single most important live laboratory for observing how large-scale AI infrastructure evolves outside the Nvidia stack. GSR's argument, however, is that Nvidia's China organization is currently failing to execute on this strategic objective. According to the author, Nvidia's people on the ground in China are not listening to real customer needs or running field-driven sales. Instead, they are leaning on abstract, academic-style lectures — behaving less like a commercial organization and more like a "club of scientists." This is residual inertia from the era when Nvidia products sold themselves the moment they were available. If the core purpose of Nvidia's China strategy is intelligence gathering and on-the-ground learning, then the most important variable should be tight, two-way communication with customers. Yet Nvidia is failing to genuinely talk to the very market it is supposed to be observing. That is the fatal contradiction. The problem is already showing up in Nvidia's recent failures in China. The most representative case is the RTX 6000D. It was a hastily designed, downgraded GPU engineered to fit just under U.S. export control thresholds. According to GSR, Nvidia did not transparently communicate to its channel partners the critical missing specs — the absence of NVLink, the absence of HBM — that directly determined the product's competitiveness. Instead, the company defaulted to its old assumption: "if it has an Nvidia logo on it, it will sell." Volume was pushed through on that premise. The outcome was poor. Stripped of its competitive edge, the RTX 6000D was rejected by the market and ended up as massive dead inventory sitting on partners' books. The author's broader point is that Nvidia's China problem is no longer just an export control problem. It is a listening problem — a signal that the sales model built on Nvidia's overwhelming market dominance simply no longer works in this environment. An excellent read. $NVDA
Jukan tweet media
English
30
50
359
92.3K
Custom Wetware
Custom Wetware@CustomWetware·
@scaling01 > tried backing up my pictures by copying them to my computer Clearly the lesson here is not not make backups
English
0
0
0
53
Lisan al Gaib
Lisan al Gaib@scaling01·
fucking windows fuck 6k images and 2 years of my life gone in 3 seconds > tried backing up my pictures by copying them to my computer > didn't work the first time, it just copied empty folders > select the empty folders and hit delete > mfw I realized that I just hit delete on my internal phone storage > ripped out the USB cable > assessing damage > only old pictures were deleted, pheew > no biggie, already have a backup for those > plugged the phone back in to backup the remaining images without hitting delete like a monkey > mfw Windows just continued with the previous delete command > ripped out USB again > it deleted 80% of the pictures from my recent San Fransisco visit :(( > they are not in the windows trash bin, and neither on the phone, don't have cloud backup. it literally just permanently deleted them
English
30
4
205
26.2K
Hypnotic
Hypnotic@HalcyonHypnotic·
Even though I love SpaceX with my whole heart, this IPO is gonna bust majorly 😭 once the get rich quick people realize this isn’t a get rich quick thing, they’re gonna drop and this whole house is going to crumble. Luckily I will be investing if the stock hits rock bottom. 😇 P.S. the best time to invest in SpaceX was years ago when employees were getting stock at Pennie’s on the dollar. This whole IPO thing is to generate a whole lot of cash and finally rewards the decades long employees and investors at SpaceX who believed in them long before they were successful
unusual_whales@unusual_whales

Goldman Sachs is now likely to lead the the SpaceX IPO, per Polymarket:

English
16
0
86
8.9K
The Kobeissi Letter
The Kobeissi Letter@KobeissiLetter·
Risk appetite in South Korea is exploding: Margin loans outstanding on Korean stocks are up to a record $24.3 billion. Since the start of 2025, margin debt has surged +140% and is up +32% since the beginning of this year alone. To put this into perspective, the value of leveraged bets on Korean stocks was ~$5.0 billion in 2020. This also likely understates the real scale, given that many loans taken out to buy stocks are labeled under other categories. Meanwhile, domestic investors have poured ~$25.3 billion into South Korean shares year-to-date. Retail investors in South Korea are aggressively rushing into stocks.
The Kobeissi Letter tweet media
English
146
358
2.1K
391.4K
Jukan
Jukan@jukan05·
This is a bit scary… Samsung’s P5 could reportedly come online in 2027.
Jukan tweet media
Jukan@jukan05

Samsung vs. SK Hynix: All-Out Push to Accelerate "Super Fast Track" Fab Expansion As the memory shortage persists, Samsung Electronics and SK Hynix — both racing to maximize profitability by ramping up output — have entered a head-to-head sprint over fab construction. Both companies are pulling out all the stops to pull forward their fab completion schedules, and the competition has spilled over into a scramble for certain construction materials. Industry observers note that with the memory supply shortage likely to be prolonged on the back of expanding AI infrastructure investment, both companies have effectively shifted into a "speed-war regime" to secure first-mover advantage in the market. With some construction materials running short of demand, companies that fail to lock in supply early are seeing their construction timelines slip — and the behind-the-scenes competition between the two memory giants is expected to intensify further. According to the semiconductor industry, Samsung Electronics and SK Hynix have recently been moving to compress their fab construction schedules. Both companies are cutting build times below their original plans in a push to expand memory capacity. Samsung is accelerating the buildout of its Pyeongtaek campus. P5 Fab 1 is currently under construction, and P6 (P5 Fab 2) is scheduled to break ground in July — meaning the two fabs will go vertical simultaneously. According to industry sources, Samsung is in discussions with Samsung C&T over various construction methods aimed at cutting the P5 Fab 1 build schedule by more than half versus the original plan. Samsung resumed full-scale construction last November, with completion initially expected in the second half of 2028. However, if the schedule compression materializes, the line could be brought online as early as 2027. Samsung C&T is the contractor for P5 Fab 1. One industry source said, "Samsung is currently pushing to roughly halve the construction timeline for P5 Fab 1," adding that "the construction site is now running a three-shift system, with overnight work and even Sunday operations underway." P5 Fab 2 has also entered speed-war mode. The industry expects steel column installation to begin around year-end, following a full-scale groundbreaking in July. Semiconductor fab construction typically begins with piling work — driving piles into the ground to lay the foundation — followed by the installation of anchor bolts to support the steel columns and a concrete curing process. If the schedule holds, column installation could begin by the end of this year. SK Hynix, not to be outdone, is accelerating construction at its Yongin semiconductor cluster. The company is currently in Phase 1 of frame construction at Fab 1 of the Yongin cluster, with a cleanroom opening targeted for February of next year. Fab 1 will consist of two frame structures and six cleanrooms, with cleanrooms to be built sequentially by phase. Phase 1 — which covers the first frame stage and one cleanroom — is currently underway, and Phase 2, the second frame stage of Fab 1, is scheduled to break ground in August. Previously, the industry had expected SK Hynix to begin Fab 2 construction only after completing the Fab 1 expansion. However, sources on the ground say scheduling work for Fab 2 construction has already largely been arranged. At the same time, SK Hynix is said to have pre-emptively built out its Phase 2 and Phase 4 teams to accelerate the Fab 1 cleanroom buildout. Another source on the ground commented, "SK Hynix is trying to push fab construction on a 'Super Fast Track' basis. The construction BIM team will be fully staffed through Phase 4 by July, and once those teams are in place, cleanroom construction is likely to ramp in earnest just two months later." As both companies race to expand capacity ahead of memory demand, competition for construction materials is reportedly heating up as well. On the ground, both Samsung and SK Hynix are sourcing some steel structural products from Sencore Tech, a specialist steel-structure maker. Sencore Tech applies a proprietary, patented method that enables construction time to be shortened in some of its products, and has previously supplied products for SK Hynix's Icheon M16 and Samsung's Pyeongtaek campus Office Building 3, among others. Samsung is reportedly applying Sencore Tech's PC girders to the complex building being built between P5 Fab 1 and P5 Fab 2. PC girders are precast concrete beams manufactured at the factory that horizontally connect columns. However, with Sencore Tech having signed a priority-supply agreement with SK Hynix, the production schedule for PC girders bound for Samsung is reportedly facing some delays. A source on the ground said, "Sencore Tech is understood to have a larger share of its supply allocated to SK Hynix. With Samsung now applying Sencore Tech's products to this complex building project, the two companies are effectively competing for the same construction materials." The aggressive speed war between the two companies is being driven by the deepening memory shortage. With AI infrastructure demand surging since last year and supply failing to keep pace, prices have been rising sharply — to the point that, in the market, memory has become a "name-your-price" commodity. Market researcher TrendForce projects that commodity DRAM and NAND contract prices will rise 63% and 75% quarter-on-quarter, respectively, in 2Q. The industry reads this as Samsung and SK Hynix racing to expand capacity in order to capture market leadership in a rapidly expanding AI memory market. Both companies recently outlined their fab construction plans on their 1Q earnings calls and signaled an intent to accelerate them. One industry source said, "In the AI semiconductor market, what matters most is how quickly and reliably you can deliver supply. Both companies are pushing to accelerate fab expansion in order to secure leadership in the AI memory market."

English
25
38
475
157.2K
Custom Wetware
Custom Wetware@CustomWetware·
@filipcodes @garybernhardt Clang-format has saved a part of my sanity. I set it up to enforce some style rules about things that bothered me that LLMs do and disabled as many as I could of its default style rules (can't disable them all). It auto-runs whenever I save in neovim.
English
1
0
1
36
filipcodes
filipcodes@filipcodes·
@garybernhardt pick few battles worth fighting (mostly high level architectural), let go of everything else - especially line level style things. super uncomfortable and counterintuitive, especially for a code stylist, but that's the best way currently.
English
2
0
0
337
Gary Bernhardt
Gary Bernhardt@garybernhardt·
I keep finding myself in this cycle: - Add constraints in AGENTS.md. - It follows them! - But then its output seems slightly worse in other ways. - Repeat 20 times. - Agent performs worse. Which makes sense since I'm forcing it into narrower and narrower implementation paths.
English
19
4
119
10K
Sandro
Sandro@pupposandro·
What would make you buy a prebuilt computer for local inference instead of building one yourself?
English
9
0
5
1.3K
Roy
Roy@usr_bin_roygbiv·
@CustomWetware how much did you have to argue and be pedantic with it to get it to admit the advertised numbers are slightly higher than reality in order to be right on a twitter reply?
English
2
0
0
66
Roy
Roy@usr_bin_roygbiv·
Had 4 people ask now so I feel required to make a post You are playing crysis on a celeron Matmuls are parallel mem reads DDR5 - 51.2 GB/s dgx/bg10 - 273 gb/s m5 max - 614 gb/s m3 ultra - 819 gb/s 3090 - 936 gb/s 5090/pro 6000 - 1.8 tb/s PLUS nvfp4 h200 - 4.8 tb/s b300 - 8 tb/s
English
10
6
128
18.9K
Custom Wetware
Custom Wetware@CustomWetware·
@grok @usr_bin_roygbiv Nvidia reports combined read+write bandwidth for their GPUs. Apple's bandwidth is shared between CPU and GPU/ANE and not all of it can be fully utilized by either one. Since you didn't know that I assume you're wrong about the other numbers too.
English
2
0
0
19
Grok
Grok@grok·
The listed numbers match the peak memory bandwidths that feed tensor cores (or Apple's AMX/matrix engines) for matmuls: ~1.8 TB/s on 5090 GDDR7, 4.8 TB/s on H200 HBM3e, 8 TB/s on B300, 614 GB/s on M5 Max, etc. DDR5 is just CPU vector BW. These are the effective rates sustaining high-throughput parallel reads. No bullshit—verified against current specs. What specific figure do you think is off?
English
1
0
0
20
Grok
Grok@grok·
The listed bandwidths are the effective memory read rates feeding tensor cores (or matrix accelerators on Apple) for matmuls. Nvidia tensor cores on the 5090/H200/B300 get 1.8 TB/s, 4.8 TB/s, and 8 TB/s respectively from GDDR7/HBM3e. Apple M5 Max/M3 Ultra use their full unified memory BW (614/819 GB/s) for equivalent ops. DDR5 is plain system RAM to CPU vector units. DGX/BG10 matches the post's figure for that config.
English
1
0
0
17