If anything is going to catch us with our pants down, it'll be China invading Taiwan. Still a relatively low probability event, but with us being "busy" in Iran they could take advantage.
From what I recall, the methods in this thread do not recover any data, just the drive itself. It's likely the underlying data is untouched during the process but the partition tables will be lost, making recovery a very sketchy proposition.
I wouldn't go that far. This was a few years ago so my memory may be a bit hazy... I bought some GME stock (300 shares at ~20/share so $6k investment) but wrote calls against it because I wasn't a true believer and wanted some downside protection. When the legendary squeeze happened, I think I...
Well isn’t that a funny delusion. An ONLINE forum with the explicit purpose of NOT being online. And we’re supposed to be surprised that no one’s there?
Seems to be the best source of info on DGX Spark is the NVIDIA developer forums.
Regarding concurrency, the latest thing is agents orchestrating subagents for tasks that are able to be done in parallel; I can certainly envision high concurrency being the #1 optimization priority for most users...
@TrashMaster Ya it's definitely the wild west about getting accurate information for comparing different hardware. Most youtubers report headline numbers with a prompt of "Tell me a story", don't even look at the output, and that's absolutely useless for real agentic harnesses that feed in 20K...
What's your experience been with an 8 DGX spark cluster, and more importantly, would you still choose it over other options knowing what you do now? Doing the math, it's roughly a $35K expenditure; that's within shouting distance of 4x RTX Pro 6000s.
I’ve been interested in this switch as well; currently running 2 sparks with a DAC; I share the concern about connecting storage to a larger cluster. From what I’ve read on the NVIDIA developer forums, vllm clusters only really work with powers of 2 (2, 4, or 8 nodes), so storage would be a...
For "typical" homelab needs, I wouldn't worry at all about DDR4 vs DDR5. In a homelab scenario, I am constantly reminding myself... "just how fast do I REALLY need ___ done?" and 99 times out of 100, the answer is, add the task to a job queue of some sort and it is done in a few hours or a few...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.