New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
I have experimented with local AI as well, nothing serious yet though, but I was running CPU inference on models less than 32b. They are impressive but without a GPU, painfully slow for the most part.
That bad, huh? I'll assume decade old Xeons then, the original budget driver. Not a dealbreaker mind, since the Ryzen of same RAM for example is a 400% increase in cost for less vCores and everything else...
Seems there was a giveaway 4GB RAM VPS for 1 year in Toronto a few pages back.
LET’S GO! RACKNERD BLACK FRIDAY 2026 IS LIVE: https://www.racknerd.com/BlackFriday/ ⚡
LET’S GO! RACKNERD BLACK FRIDAY 2026 IS LIVE: https://www.racknerd.com/BlackFriday/ ⚡
Not my point. I have used both the Xeons and Ryzens from RackNerd. Some of my projects were really happy on the Xeons, but I can't remember which versions there, only remember I never faced any issues with them.
I moved to their Ryzen cause I got a good deal a few years ago. I just meant if you can get the Ryzen go for it. Else, the Intel-based VPSes from RackNerd will also serve your purpose well.
I think you missed to add the message
LET’S GO! RACKNERD BLACK FRIDAY 2026 IS LIVE: https://www.racknerd.com/BlackFriday/ ⚡
Yeah that's what I just saw and noticed the location. They don't have anything in stock in Toronto right now, not even dedicated servers.
It's been about a year since Toronto has been in stock (source: https://racknerdtracker.com/?q=Toronto&stock=0 including out of stock)
Combined Specs (Two 128 GB Units)
• AI Compute: Up to 2 PFLOPS (FP4 with sparsity); 12,288 total CUDA cores (6,144 per unit)
• CPUs: 40 total Arm cores (two 20-core processors, each with 10 Cortex-X925 + 10 Cortex-A725)
• Memory: 256 GB aggregate LPDDR5x unified memory across two nodes (128 GB per unit; 273 GB/s bandwidth per node, 546 GB/s aggregate)
• Model Capacity: Supports models up to ~400B–405B parameters across distributed workloads
As for the model I am still using Deepseek v4 flash 0731 due to slow internet but It hasn't failed any tasks yet
LET’S GO! RACKNERD BLACK FRIDAY 2026 IS LIVE: https://www.racknerd.com/BlackFriday/ ⚡
Ah yes. I believe we will have a few more though rare during the flash sales on BF.
LET’S GO! RACKNERD BLACK FRIDAY 2026 IS LIVE: https://www.racknerd.com/BlackFriday/ ⚡
Just curious, what else are you doing/generating and what are your machine specs? I am not stranger to AI (use it at my day job and during personal projects), but I am also not rich to buy hardware for anything better than a 9b that fits in locally. And during my playing with a 27b model I was very disappointed.
Awesome. I am confused. Is this a Snapdragon AI processor?
Wish I was this rich. I am jealous, but yeah, that looks lots of fun. Enjoy!
BTW, I am genuinely curious as to why no one on LET is talking about GTA VI. Not that I have played or planning to play it, but it's the buzz all around.
Me too. I am running on 32GB RAM, no GPU
DGX spark or similar
Order Number: 5020484975, please double my bandwidth on both orders. Thanks!
Didn't know DGX was ARM.
Yep two dgx sparks
At this point, is it better to get gaming GPUs or NPUs like Ryzen AI Max for local LLM inference?
How big of a hole did it burn in your pockets?
Hello, I would like to double my bandwidth. Invoice #26921412 Thanks!
And does anyone have experience with Alibaba Cloud for DeepSeek 4? Any good?
I mean I don't disagree with you, and if they put the Ryzen's on sale that would be good, but as of today it's a 400% increase along the same RAM line with less cores, storage, etc....of course those less quantity cores are probably orders of magnitude better at the job..and it's not a critical or profitable workload so I'm hard pressed to justify all costs, hence the deal hunting and interest in what the sacrifices are to get to this cost.
Alright, for anyone following along I did get a quick bite from sales, they've got a mix of Xeon 2680 V2s and Gold 6252. So some old and some newer, luck of the draw.
Order Number: 9900760405
Please double the transfer.
Thank you!