| ▲ | jazzyjackson a day ago |
| Question: are SoCs with on die memory be effected by this? Looks like the frame.work desktop with Ryzen 128GB is shipping now at same price it was on release, Apple is offering 512GB Mac studios Are snapdragon chips the same way? |
|
| ▲ | nrp 21 hours ago | parent | next [-] |
| We’ve been able to hold the same price we had at launch because we had buffered enough component inventory before prices reached their latest highs. We will need to increase pricing to cover supplier cost increases though, as we recently did on DDR5 modules. Note that the memory is on the board for Ryzen AI Max, not on the package (as it is for Intel’s Lunar Lake and Apple’s M-series processors) or on die (which would be SRAM). As noted in another comment, whether the memory is on the board, on a module, or on the processor package, they are all still coming from the same extremely constrained three memory die suppliers, so costs are going up for all of them. |
| |
| ▲ | mips_avatar 17 hours ago | parent [-] | | How do suppliers communicate these changes? Are they just like yep now it’s 3x higher? Im surprised you don’t have longer contracts | | |
| ▲ | appellations 17 hours ago | parent | next [-] | | Longer contracts are riskier. The benefit of having cheaper RAM when prices spike is not strong enough to outweigh the downside of paying too much for RAM when prices drop or stay the same. If you’re paying a perpetual premium on the spot price to hedge, then your competitors will have pricing power over you and will slowly drive you out of the market. The payoff when the market turns in your favor just won’t be big enough and you might not survive as a business long enough to see it. There’s also counterparty risk, if you hit a big enough jackpot your upside is capped by what would make the supplier insolvent. All your competitors are in the same boat, so consumers won’t have options. It’s much better to minimize the risk of blowing up by sticking as closely to spot at possible. That’s the whole idea of lean. Consumers and governments were mad about supply chains during the pandemic, but companies survived because they were lean. In a sense this is the opposite risk profile of futures contracts in trading/portfolio management, even though they share some superficial similarities. Manufacturing businesses are fundamentally different from trading. They certainly have contracts in place that cover goods already sold. They do a ton of preorders which is great since they get paid before they have to pay their suppliers. Just like airlines trade energy futures because they’ve sold the tickets long before they have to buy the jet fuel. | |
| ▲ | baq 13 hours ago | parent | prev | next [-] | | If you’re Apple, maybe that works, in this case we’re seeing 400% increases in price, instead of your RAM you’ll be delivered a note to pay up or you’ll get your money back with interest and termination fees and the supplier is still net positive. | |
| ▲ | chii 16 hours ago | parent | prev [-] | | > longer contracts the risk is that such longer contracts would then lock you into a higher cost component for longer, if the price drops. Longer contracts only look good in hindsight if ram prices increased (unexpectedly). |
|
|
|
| ▲ | addaon a day ago | parent | prev | next [-] |
| > Question: are SoCs with on die memory be effected by this? SoCs with on-die memory (which is, these days, exclusively SRAM, since I don't think IBM's eDRAM process for mixing DRAM with logic is still in production) will not be effected. SiPs with on-package DRAM, including Apple's A and M series SiPs and Qualcomm's Snapdragon, will be effected -- they use the same DRAM dice as everyone else. |
| |
| ▲ | pixelpoet a day ago | parent | next [-] | | The aforementioned Ryzen AI chip is exactly what you describe, with 128 GB on-package LPDDR5X. I have two of them. To answer the original question: the Framework Desktop is indeed still at the (pretty inflated) price, but for example the Bosgame mini PC with the same chip has gone up in price. | | |
| ▲ | plagiarist 18 hours ago | parent [-] | | Are your two chips in Framework Desktops, or some other package? I'm interested in a unified memory setup and curious about the options. |
| |
| ▲ | zahlman 20 hours ago | parent | prev | next [-] | | https://en.wiktionary.org/wiki/die#Noun "dice" is the plural for the object used as a source of randomness, but "dies" is the plural for other noun uses of "die". | |
| ▲ | layer8 13 hours ago | parent | prev [-] | | *affected |
|
|
| ▲ | piskov a day ago | parent | prev | next [-] |
| Apple secured at least a year-worth supply of memory (not in actual chips but in prices). The bigger the company = longer the contract. However it will eventually catch up even to Apple. It is not prices alone due to demand but the manufacturing redirection from something like lpddr in iphones to hbm and what have you for servers and gpu |
| |
| ▲ | mirsadm 13 hours ago | parent | next [-] | | Apple charges so much for RAM upgrades that they could probably not even increase prices and still be fine. They won't but they probably could. | | |
| ▲ | layer8 13 hours ago | parent [-] | | At the cost of reduced margins, which shareholders may not like. |
| |
| ▲ | greesil 17 hours ago | parent | prev | next [-] | | Apparently Google fucked up https://www.google.com/amp/s/www.indiatoday.in/amp/technolog... | | | |
| ▲ | trollbridge 21 hours ago | parent | prev [-] | | I have a feeling every single supplier of DRAM is going to be far more interested in long-term contracts with Apple than with (for example) OpenAI, since there's basically zero possibility Apple goes kaput and reneges on their contracts to buy RAM. | | |
| ▲ | saagarjha 21 hours ago | parent [-] | | Yes, but OpenAI wants $200 billion in RAM and Apple wants $10. | | |
| ▲ | Applejinx an hour ago | parent [-] | | But you DO have to consider the possibility they Enron themselves and the promised $200 billion never exists. These are not the most honest of people. |
|
|
|
|
| ▲ | dehrmann 17 hours ago | parent | prev [-] |
| I would think so because fab capacity is constrained, and if you make an on-die SoC with less memory, it uses fewer transistors, so you can fit more on a wafer. |
| |
| ▲ | hvb2 13 hours ago | parent [-] | | But bigger chips mean lower yields because there's just more room for errors? |
|
