Nvidia debuts $4,999 DGX Spark with half the RAM and storage, amid memory crunch

In an effort to provide a more affordable AI system, Nvidia is introducing a cut down version of the DGX Spark with half the memory and storage. The GB10-based systems will be offered exclusively through hardware partners including Acer, Asus, Dell, Gigabyte, HP, and MSI, and are expected to retail for around $4,999. The less- …

[Ian Johnston]

Desperate to milk every last penny out of every last mug before the bubble bursts ...

[cyberdemon]

Waste of chips

They've borked the LPDDR5X supply chain with their HBM madness, and now they're wasting what few LPDDR5x chips that are still manufactured on this shit?

At least these ones can be desoldered after the crash to make something useful.. like Raspberry Pis

[elsergiovolador]

Idiots

They should have released DGX with 2GB of RAM at the same price.

[rgjnk]

More accessible??

"While the 64 GB version’s lower MSRP should make the systems more accessible amid the ongoing memory crunch."

It's not more accessible, it's just maintaining a level of product access at a similar price point. Squeezing the mugs for vast margins in the process.

'More accessible' doesn't usually involve paying significantly more than the original better version cost not very long ago - if price was the issue then this hasn't improved at all.

A less capable version at a lower price traditionally would give broader access, but not when they're cranking the prices up at the rate they are.

[MonkeyJuice]

Re: More accessible??

The entire point of the DGX is it makes training and finetuning possible, which requires plenty of ram. It's DDR5 bandwidth limited, and is truly embarrassingly slow for inference workloads. a 64gb version could be swapped for dual 32gb cards clocked ten times the speed for 2/3rds the price, there is absolutely no reason to pay through the nose for this configuration. I guess nVidia just think people are idiots at this point.

[steviebuk]

Nvidia

needs to fail when the cock known as Jensen claims kids no longer will need to know maths. The fuck whit.

[xyz123]

What FPS does it get in WoW?

[IGotOut]

Zero.

[Sproggit]

How is the RAM Installed?

I’m curious…. Are these machines shipped with all the RAM soldered down, or do they use conventional sockets?

It seems at least plausible that the current insanity with RAM pricing will be followed by a crunch… and when that does people shelling out *SEVEN THOUSAND DOLLARS* today might like to wring a little but more shelf life from these things.

I ask because the cases look so compact I have a hunch that might not be the case and that nVidia decided to explicitly design out upgradeability before the first unit rolled of a production line.

I’m now of the view that companies that fail to sell repairable/upgradeable technology should be charged an e-Waste Tax that pushes up their price even further - 100% would be nice - until they get the message that we’ve had enough of their sh**.

[MonkeyJuice]

Re: How is the RAM Installed?

You guessed it- memory is soldered in (because of course it is!).

Also, very specifically- opening the Spark and replacing things will instantly void the warranty.

The "upgrade path" was always about lashing two of them together, not tinkering with the insides.

[Sproggit]

Re: How is the RAM Installed?

OK, so, that's a "Hard 'No!'" then...

[Sproggit]

We all know this is Insane, but then...

... a quick check on Overscalpers shows that they are now listing all nVidia 5090 GPUs for £6000.

Yes, you read that right. Six grand. For a GPU.

For context, on May 10th this year, the Gigabyte Aurus (water cooled) was £3,200. On 10th June the same card was £4,200 and the Inno3D iChill Frostbite Pro came in at £3500.

On August 3rd the Gigabyte had moved to £4200 and the Inno3D had bumped to £4000.

Today, roughly 60 days later, they are both £6000.

I mean, I know this is Overscalpers we're talking about... but a 50% price increase in 60 days?

[MonkeyJuice]

Re: We all know this is Insane, but then...

That's the nvidia tax for you. if you _really_ must have something that can run local llms, probably the best bang for your buck might be Intel Arc- they're not seeing the spike so much because they're uncool, but you can get 32gb and even 48gb dual workstation cards and there's still a fair bit of stock. They'll give you more oomph than a DGX spark anyway, because nVidia decided to memory bandwidth bottleneck the Spark, just like Apple did- only Apple have an excuse, because they're not routing around a high end grace blackwell acting as an ornamental space heater.

It's all just an expensive hobby of course... the only winning move is not to play. Your best bet is to rent open models that you could run on any potential hardware you're looking at, benchmark your problem sets with them, and conclude you didn't want or need the new shiny anyway.

[Sproggit]

Re: We all know this is Insane, but then...

To be honest... right now I'd much rather have a GPU setup that would enable me to run games at a decent frame rate on my existing monitor setup - which is a trinity of 27" Gigabyte Aurus monitors at 2560x1440 each, for a total desktop of 7680x1440. Give me a GPU that can run 100fps with titles like Horizon : Forbidden West, Cyberpunk 2077, Witcher 3 and the like... and I'll quite happily sit out the AI bubble, if that's OK with everyone... ?

[ecofeco]

Too many digits

The price has too many digits.

But suckers gonna be suckers.

[druck]

Both cognitive and financial surrender in one small box.