Hacker News Reader

Comment by cwizou | Hacker News Reader

> I don’t see much evidence that TSMC ever really slowed production, at least for their most advanced processes that are in high demand.

One thing I heard is that the drought (worst one in 50 years in Taiwan) has really affected them, and various sources seem to confirm it, here's one : https://fortune.com/2021/06/12/chip-shortage-taiwan-drought-...

And while I wholeheartedly agree with the crypto/GPU issue, all in all, the GPU market is overall fairly tiny compared to TSMC's total output.

Most of the GPU issue is linked to massive demand and conservative allocations that were bought by GPU makers a while back. Adding extra allocation because of demand is really hard specifically on constrained high end nodes like GPU uses.

Car makers and others in the industry don't use those nodes, they use older ones and those are equally constrained.

The GPU/crypto issue is the cherry on top of an already very constrained TSMC, it seems.

View on HN

> And while I wholeheartedly agree with the crypto/GPU issue, all in all, the GPU market is overall fairly tiny compared to TSMC's total output.

From an economics perspective, it doesn’t really matter. The price of a market-rate good is the marginal price of producing one more unit. As long as cryptocurrency is financially incentivizing people to consume 100% of the marginal production, prices will continue to go up.

This also cascades down their lineup: Vendors who would like to use 7nm for their chip will instead choose 10nm, which will bump some of the customers who wanted to use 10nm back to the 12nm or 16nm nodes and so on.

GPU's are market-rate goods?

Crypto mining is a market rate good.

Nvidia already added in a hardware crypto mining block to it's latest models and they are still having massive supply issues. It's not about crypto even though everyone likes to make a big deal out of it to feel better.

The latter does not follow from the former (even if it might be true). The mining block only affects the latest batches of a few models, but profitability of mining also affects demand for the other models, for AMD GPUs, and in the second-hand market. Unless the supply of mining-blocked GPUs is sufficient to cover all displaced demand from these other segments, it doesn't necessarily help much.

They don't block all mining, just specific algorithms. It's easy enough to mine some other coin and convert to the one you want. In fact, that's often more profitable.

Yes. We've seen chipmaker MSRPs inflated, especially in the most recent releases like the 6600XT. Chipmakers are constrained by contractual obligations and PR though. AIBs have also inflated their MSRPs for cards. They seem to have contractual agreements with the chipmakers to sell some portion of their cards at Chipmaker MSRP, at least around the launch. However, to their best ability they are putting their die allocation into their highest end card variants that have the highest MSRPs. Then distributors, retailers, and PC builders mark up the cards over the AIB MSRP. Again, these are often limited to some extent by agreements with their suppliers. If all of these markups don't reach the market rate the cards are bought up by scalpers with bots and sold at the true market rate on eBay.

I think you may be massively underestimating the lead times on those decisions. We're talking years.

Also, you can't simply choose another node last minute, your design is intimately linked to the PDK of the node. Porting to another node is far from trivial.

A recent Typhoon (one of the half dozen that hammered China) refilled all their reservoirs about a month or so ago.

The one thing that complicates economics is that it's not a linear equation like they teach in Macroeconomics. The equations are more or less "true" in a vague sense but each term is actually a differential equation term in both time and other quantities.

If you are familiar with simple one variable ODEs per STEM undergrad courses, imagine the 2nd order forms but now those exist across many variables and time spans. It makes you realize that CNBC is just pulling causes/effects and making predictions out of their asses.

> A recent Typhoon (one of the half dozen that hammered China) refilled all their reservoirs about a month or so ago.

Yes I should have phrased that better, sorry. The thing is, making one wafer of chips takes roughly 3 months, and everything else has even more terrible lead times. So there's a delay in the impact of events.

The drought was over months ago. Taiwan is well into typhoon season (they've had two already). The water reservoirs are full[1].

And TSMC was never really affected. They were not forced to save water, unlike every other company on the island.

[1] https://www.taipeitimes.com/News/biz/archives/2021/07/05/200...

> One thing I heard is that the drought (worst one in 50 years in Taiwan) has really affected them ...

This is the one aspect of manufactuing of chips that many are actually ignorant about - it needs huge amounts of pure water.

This is the main reason why India doesn't yet have any fab, despite having the human resource for it and investors and government showing great interest to setup one.

Two or three decades back Intel desired to setup a fab in India, but the deal fell through because it would mean India would have to divert a huge amount of water from agriculture and potable water for the population to such fabs. TSMC has approached the indian government many times (even this year) seeking permission to setup a fab in India. But the indian goverment has remained non-commital.

The indian government believes it is in their long-term interest to wait for fab tech to evolve to use less water. Indian states already squabble over sharing water resources, and India has to feed a billion people, and such a diversion of resource doesn't make sense in the long-term.

(Perhaps the James Bond movie was right - we'll all be fighting for fresh water in the coming decades. :)

What happens to the water the plants use? Is it polluted and unusable, or just tied up temporarily and can be released [safely] back into wherever it was diverted from?

Evaporation, goes to ground water table, stored in plant matter, goes back to river now enriched with fertilizers from fields. If there is fight over water the last one is likely quite small fraction.

They meant the water used in fabs, not agriculture.

Is the returned water safe though? That's what I wasn't sure about, do the fab plants pollute the water to the point it's unusable, or is it able to be re-used(drank).

Yes, it is polluted and even toxic.

> The manufacturing of semiconductors generates wastewater that contains heavy metals and toxic solvents. Due to the hazardous nature of these liquids, semiconductor companies historically have incurred clean-up costs related to groundwater contamination of up to USD 100 million ... Some companies are proactively meeting these challenges and mitigating related risks by reducing their reliance on chemicals and pro-actively improving the treatment of water before discharge. STMicroelectronics, for example, recycled and reused 45% of the total water it withdrew in 2015 and it partnered with Micron Technology to install on-site fluoride waste treatment facilities.

Source: https://www.sustainalytics.com/esg-research/resource/investo...

The report also gives concrete numbers: Intel fabs consumed 9 billion gallons (around 3.4 trillion liters) of water in 2015.

Apparently a fab roughly consumes around 20,000 tons (around 1.98 million liters) of water per day:

> The chipmaking industry is water-intensive. According to nonprofit China Water Risk (CWR), an average semiconductor factory consumes roughly 20,000 tons of water a day, equivalent to the daily water use of a city of 58,000. Yet despite its high water consumption, the chip industry—in Taiwan, at least—is not normally a noticeable drain on communal resources.

Source: https://fortune.com/2021/06/12/chip-shortage-taiwan-drought-...

This is a very false premise. Water is no reason to not have a fab, just like Japan builds semiconductor & nuclear plants in a seismically active zone. So does Israel in a dessert.

India receives way more rain than it requires even in the drier parts, the problem with soil degradation and deforestation that came with unchecked "modern" agriculture.

The productivity & economic benefits from a FAB and semiconductor ecosystem in India will far outweigh any negative, putting a plant close to a nuclear plant and a massive desalination plant to ensure both power, water as well as security will do the job.

> The productivity & economic benefits from a FAB and semiconductor ecosystem in India will far outweigh any negative, putting a plant close to a nuclear plant and a massive desalination plant to ensure both power, water as well as security will do the job.

Diversion of water resources to fabs (they require around 2 million liters a day; Intel fabs consumed 9 billion gallons / 3.4 trillion liters in 2015) can gravely affect our food security as 60% of our cultivable land are in "drought prone" areas. Moreover, the water released from the fabs is toxic and can heavily pollute our environment even after they are treated and stringent standards are complied with (and expecting that is a joke in India).

Your suggestion of using a nuclear plant and a desalination plant might seem practical, but it ignores the fact that the byproduct of desalination plants are huge quantities of various salts. And dumping them causes pollution (if you dump them back in the sea, it can create dead zones for kms around as most aquatic life cannot survive in such a "salty" concentrated environment).

The run off from agriculture and tanners is several orders of magnitude greater.

If you factor the over all benefit to the pollution and resource consumption, a fab is several orders of magnitude better, why else would everyone else scramble to get one built.

A fab at the expense of food security is not a good deal. When you can't feed your people, you are at the mercy of other nations, and that greatly harms national security. India used to once have frequent famines, and had to beg food from other countries. Today it can feed its billion+ population because it is the world leader in producing fruits, vegetables, grains, diary produce etc. A starving population doesn't care about CPU chips.

> One thing I heard is that the drought (worst one in 50 years in Taiwan) has really affected them,

The power issues earlier this year in Texas also impacted production.

Weather issues are often the cause of the bust cycles in silicon manufacturing. Just people outside of the industry rarely discuss when memory prices or whatever spike due to a typhoon hitting Japan/Korea/Taiwan. This year, we've have major weather events hit chip production across the globe, at basically the same time, plus the pandemic.

Drought has been over for awhile now and it never really slowed chip production.

https://archive.ph/7bGOO article link

Reply on news.ycombinator.com