Samsung's 60% DRAM price hike signals a new phase of global memory tightening

Well, patience as a consumer might pay off in the next year or so when the music stops and hyperscalers are forced to dump their inventories.

There still isn't a clear path to profitability for any of these AI products and the capital expenditure has been enormous.

Its a bit of a shame these AI GPUs don't actually have displayport/hdmi output ports because they would make for nice cheap and powerful gaming GPUs with a lot of VRAM, they would potentially be really good graphics cards.

Will just have to settle for insanely cheap second hand DDR5 and NVMe drives I guess.

> Well, patience as a consumer might pay off in the next year or so when the music stops and hyperscalers are forced to dump their inventories.

Their inventories are not what consumers use.

Consumer DDR5 motherboards normally take UDIMMs. Server DDR5 motherboards normally take RDIMMs. They're mechanically incompatible, and the voltages are different. And the memory for GPUs is normally soldered directly to the board (and of the GDDRn family, instead of the DDRn or LPDDRn families used by most CPUs).

As for GPUs, they're also different. Most consumer GPUs are PCIe x16 cards with DP and HDMI ports; most hyperscaler GPUs are going to have more exotic form factors like OAM, and not have any DP or HDMI ports (since they have no need for graphics output).

So no, unfortunately hyperscalers dumping their inventories would be of little use to consumers. We'll have to wait for the factories to switch their production to consumer-targeted products.

Edit: even their NVMe drives are going to have different form factors like E1.S and different connectors like U.2, making them hard for normal consumers to use.

The problem is that it is not entirely clear that the hyperscalers are buying DDR5, instead it seems that supplies are being diverted so that more HBM/GDDR wafers can be produced.

HBM/GDDR is not necessarily as useful to the average person as DDR4/DDR5

I wouldn't mind my own offline Gemini or ChatGPT 5. But even if the hardware and model were free, I don't know how I'd afford the electricity.

AI GPUs suck for gaming, I have seen a video from a guy playing Red Dead Redemption 2 on a H100 at a whooping 8 FPS! That's after some hacks, because otherwise it wouldn't run at all.

AI GPUs are stripped away of most things display-related to make room for more compute cores. So in theory, they could "work", but there are bottlenecks making that compute power irrelevant for gaming, even if they had a display output.

A single machine for personal inference on models of this size isn't going to idle at some point so high that electricity becomes a problem and for personal use it's not like it would be under load often and if for some reason you are able to keep it under heavy load presumably it's doing something valuable enough to easily justify the electricity.

If you can't afford the electricity to afford to run the model on free hardware, you'd certainly never be able to afford the subscription to the same product as a service!

But anyway, the trick is to run it in the winter and keep your house warm.