Just a matter of time before these guys are supplying ML card manufacturers with slower memory that works well enough for ML use cases at a fraction of the current prices.
Import and trade restrictions exist in some places but if the rest of the world buys the cheaper Chinese stuff that works about the same then the rug will come out from under the current guys and they will literally never recover.
Guessing the short term profits are all worth it to them, like the typical VC.
The tech overlords are desperately trying to buy up and restrict the supply of all the memory in the hope that judgment day comes and it works out in their favor or some other dumb drug induced scifi fever dream. China will offer slightly shittier chips for a 10th of the price and if we aren’t allowed to buy them we will rent them (and they will of course take our data). These bubbles tend to last about 5 years and we are in year 2. So buckle up.
What makes you think we are in year 2 of a bubble? Genuine question because I think this thing has been going on for at least three years. But five years is an average, right? So averages can be very sneaky, meaning some bubbles may have vastly shorter or longer durations. As a matter of fact, we don’t know. And that’s the worst part, isn’t it? Because a round-tripping scam needs external money to keep going, and question is how long the US government is willing or able to fund it, considering they have already tried appeasing Japan for them to keep their US bonds.
If you are willing to risk uncertainty, a modded 4090 might be better than a 5090 if you need VRAM. I think the Nvidia firmware signature was leaked for the 4090, so modders have been able to add twice the memory to individual cards and have it still work. 48 gigs apiece.
VRAM for VRAM, probably cheaper than a 5090 - if only because a 4090 is less powerful in all other respects.
Anyhow, if you decide to get more RAM, better do it soonish if you are American. Between the Canadian tariffs and the emergency oil stockpile running dry around half a year, prices are set to climb further. This whole situation sucks.
I thought 32GiB would be enough coming from 16, but nope, I found a new use case where it’s too little.
Originally, I wanted to be able to run at least 2 VMs with 8GiB assigned at once.
But the new use case is a 4GiB RAM VM with 20GiB virtual disk stored in RAM (TMPFS). And that’s not all that much.
I’ve also just been using it for random temporary things, as I am still paranoid about SSD wear. If I need to download an ISO just to burn it, there’s no need to take, say, 3GiB from the SSD’s lifespan, no matter how small that may seem.
But even if not used explicitly, it’s still pretty useful for disk I/O cache. Especially with slow things like memory cards. At least for reading, for writing it’s just annoying because I have no idea how much has already transferred, and neither can I safely stop it.
Just a matter of time before these guys are supplying ML card manufacturers with slower memory that works well enough for ML use cases at a fraction of the current prices.
Import and trade restrictions exist in some places but if the rest of the world buys the cheaper Chinese stuff that works about the same then the rug will come out from under the current guys and they will literally never recover.
Guessing the short term profits are all worth it to them, like the typical VC.
The tech overlords are desperately trying to buy up and restrict the supply of all the memory in the hope that judgment day comes and it works out in their favor or some other dumb drug induced scifi fever dream. China will offer slightly shittier chips for a 10th of the price and if we aren’t allowed to buy them we will rent them (and they will of course take our data). These bubbles tend to last about 5 years and we are in year 2. So buckle up.
What makes you think we are in year 2 of a bubble? Genuine question because I think this thing has been going on for at least three years. But five years is an average, right? So averages can be very sneaky, meaning some bubbles may have vastly shorter or longer durations. As a matter of fact, we don’t know. And that’s the worst part, isn’t it? Because a round-tripping scam needs external money to keep going, and question is how long the US government is willing or able to fund it, considering they have already tried appeasing Japan for them to keep their US bonds.
I have 64gb of DDR5 RAM. But given the situation I would buy more even if I do not need it. 128gb won’t hurt.
And if they can make a clone of the Nvidia 5090 for half its cost I’d get that, too.
If you are willing to risk uncertainty, a modded 4090 might be better than a 5090 if you need VRAM. I think the Nvidia firmware signature was leaked for the 4090, so modders have been able to add twice the memory to individual cards and have it still work. 48 gigs apiece.
VRAM for VRAM, probably cheaper than a 5090 - if only because a 4090 is less powerful in all other respects.
Anyhow, if you decide to get more RAM, better do it soonish if you are American. Between the Canadian tariffs and the emergency oil stockpile running dry around half a year, prices are set to climb further. This whole situation sucks.
I thought 32GiB would be enough coming from 16, but nope, I found a new use case where it’s too little.
Originally, I wanted to be able to run at least 2 VMs with 8GiB assigned at once.
But the new use case is a 4GiB RAM VM with 20GiB virtual disk stored in RAM (TMPFS). And that’s not all that much.
I’ve also just been using it for random temporary things, as I am still paranoid about SSD wear. If I need to download an ISO just to burn it, there’s no need to take, say, 3GiB from the SSD’s lifespan, no matter how small that may seem.
But even if not used explicitly, it’s still pretty useful for disk I/O cache. Especially with slow things like memory cards. At least for reading, for writing it’s just annoying because I have no idea how much has already transferred, and neither can I safely stop it.