Has anyone else noticed a short supply of MSRP or close-to-MSRP GB10’s recently? I guess I am spoiled by in-store MicroCenter pricing, because that was all I checked for the past few months.
Not only is Central Computer running low (completely out of 1TB Asus GX10’s, out of Sparks besides bundles),
but MicroCenter’s 1TB Asus GX10’s seem to be sold out in the stores I checked,
Same here in Europe, cheapest retail option I can find is 5500 usd for 1T Asus. Opposing to sub 5K I paid in May for 4T Gigabyte Gen5. Those are about 7K+ now.
I settled for the CRS812 because the CRS804 was out of stock at the time. Guess I won’t have to worry about needing 8x clusters anytime soon at these prices! 😅
Here in Switzerland GX10 2TB goes for $ 5’500, 1TB ($ 5’250) and 4TB ($ 7’000) seem to be sold out.
On the other hand, DGX Sparks are available for $ 6’200. I got mine a few days ago for 5’600. Prices seem to go up quickly here as well.
yeah the prices have gone up alot! The plus is that it shows there is a market for them, so lets take the positives, continued support and continued newer versions.
I ordered two more GX10s end of April in Norway and they have yet to arrive. Delivery date keeps getting pushed a week or two in the future. Price has increased by almost 2x since then, so I’m keeping the order live and hoping they eventually arrive…
I would speculate, OEMs redirected hardware towards RTX Spark, which wont have connect-x7 and they still can charge $3k+ as alternatives, including 32-48gb gpus, ~doubled original pricing.
I still struggle to see the use case for that machine, but time will tell I guess. I for one wouldn’t want my laptop running inference on battery
IMO it’s just due to capacity being moved to HBM. The memory makers claim that DDR actually has much better margins than HBM so I wonder if they come back at some point if hyperscaler demand drops.
When they first announced RTX Spark, I was intrigued. Portable local inference, cool. Then I started thinking about it. Virtually anywhere I would go where portable compute (i.e., a laptop) would be useful would also have connectivity. Through the magic of the Internet, that means I would have connectivity back to my local inference in the office. Ergo, portable local inference, for me, is not necessary.
On the other hand, I can imagine at least one use case where it might be useful. Sales demo, say, where you don’t want to deal with the vagaries of prospective (or current) clients’ guest wifi. Or, maybe you’re an avant-garde performance artist using it to deliver generative music at a rave using crowd reaction (video+audio inputs?) to drive the gen of the music near-real-time with accompanying video. I’d actually maybe pay to see that…once.
I have no idea why GB10 prices have gone up so much this week. Low-key disappointed in it, because I was kind of working up a justification for a fourth one. Which I don’t really need, but would be cool; so, if they’re now unobtainium…financial hit avoided.
That’s true. It’s likely more desirable to use the memory for other dev tasks if someone has the option. That includes just having a lot of browser tabs open at once. I think they’ll compare to MacBooks with Apple Silicon. So just a good dev laptop that can run AI models as an added benefit.
I watched the ASUS GX10’s go out of stock in real time, while Amazon screwed my account over with suspicious activity locks for trying to buy a pair before everyone ran out of the $4k each stock. Today I drove down to the Microcenter and picked up a couple DGX Sparks. My main reason for looking into these devices now: new Flash models are really really good, Hermes is my main token driver these days, and the community here has done amazing work with well tuned recipes to run these Flash models with solid tok/s and concurrency.
It wouldn’t surprise me if there’s a run on supply. The new Mac’s are nice, I have a Mac for AI use, but it’s mostly just a chat box and not a local agentic inference platform that can run 10 agents at once.
RTX Spark laptops may or may not destroy intel/amd/qualcomm premium aka for creative laptops, but I’m sure it wont affect that much macbooks, as macbook hardware+optimized software, where RTX Spark is… Windows first.
At the same time, RTX Spark as mini pc for $3-4k with DGX Spark for $5-5.5-6k due to addition of c2c for ~$1-1.5k, but everything else is the same (even cases), will be popularized as access to “free tokens” for your main service, like on Perplexity computer, I suspect at least partially demand for sparks grew because of that release last month. Check offers for AI as a service, almost all runs from DGX Spark, as for such service you need to serve model and harness/db/files etc.
