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
You can ask deepseek what deepseek needs to run itself, deepseek-r1:14b.
I asked deepseek and it said you do not meet minimum requirements.
You can run very very light task though.
Minimum Requirements:
CPU: 8 cores (Intel Xeon or equivalent)
RAM: 32 GB
GPU: 1x NVIDIA A100 (40 GB VRAM) or equivalent
Storage: 100 GB SSD
Operating System: Ubuntu 20.04 or later
Recommended Requirements:
CPU: 16 cores (Intel Xeon or equivalent)
RAM: 64 GB
GPU: 2x NVIDIA A100 (40 GB VRAM) or equivalent
Storage: 200 GB NVMe SSD
Operating System: Ubuntu 20.04 or later
https://tensordock.com/
You don't really need a dedicated GPU to run deepseek-r1:14b. I'm running this exact same model on my KS-LE-1 (32 GB RAM, E3-1245 V2), and it runs just fine. It'll spit out 4 - 6 words each second, which is not fast but kinda works for my case. Of course, if you want it to handle heavy workloads, then a GPU is needed.
The model itself requires about 10 GB of storage, and it costs ~ 12 GB of RAM during running on my server.
Checkout our dedicated VDS GPU lineup
https://crunchbits.com/gpu/cloud
Deepseek-r1:14b is based on Qwen 2.5 with 14 billion parameters. It needs minimum of 8 GB VRAM with 4 bit quantization. To conformably use it, I would go with 12 GB VRAM GPU.
Of course it can be quantized even more (it will be dumb) or run on a CPU (on RAM instead of VRAM).
I recommend running it on Ollama. Btw there are now some better models (look in Ollama library). Qwen is bad at instruction following. Not sure how much deepseek improves that.
Just asking: is that possible training AI module on GPU server and move that data to low end server
I mean simple server without gpu, does able to respond according to those data?
Better look for Mac mini with m4
for this model it performs gr8 and I think it will be cheaper to buy one and host by self.
We now offer GeForce RTX 4090 upgrades on any of our Ryzen 7 or Ryzen 9 dedicated servers.
https://deals.fiberstate.com/
Let us know if we can help!
how much does it cost? I don't see it as an upgrade option on the order form
https://billing.fiberstate.com/index.php?rp=/store/bare-metal/amd-ryzen-9-7950x
If you select our Ryzen 9 9750X package for instance, on the configuration page you'll see a drop down option for GPU: Integrated, you can select the 4090 add-on here.
RTX 4090 is a $200/mo option.
Its also available on our Ryzen 7 5700G servers, but these are currently sold out. We should have restock on them in the next week.
good topic for AI hosting
Hi,
we can offer: 9900X, 64 GB RAM, 1 TB NVMe, RTX A6000 for 421,21 EUR per month excl. VAT.
We have a few 4090s in stock right now: https://puregpu.com/
32GB memory probably not enough. As the prompt token input size increases, as does memory requirements as well. I started playing with local AI and 32GB memory isn't enough. Especially if you're dealing with large context token inputs >100,000 and later want to add other tooling like RAG and vector databases.
Instead check out Openrouter.ai for cloud based have both free and paid LLM models https://openrouter.ai/models. With free Google Gemini 2.0 models and DeepSeek, I've pushed 40 million tokens/month
Some examples
Hello @comXyz
You can check our GPU servers in Chicago, Florida, Germany, Japan, and Washington.
Specifications:
If you need any information, please DM me.
Depending on what you want to do with the server, personal or business, esp. expected utilization rate, you might find this open source project helpful: https://docs.skypilot.co/en/latest/docs/index.html
CPU: E5-2670V3 (12×2.3GHZ)
GPU: GeForce GTX 1080 Ti
RAM: 64GB DDR4
DRIVE: 250Gb SSD
PORT: 1 GBPS
$230
CPU: E5-2670V3 (12×2.3GHZ)
GPU: 2 x GeForce GTX 1080 Ti
RAM: 128GB DDR4
DRIVE: 250Gb SSD
PORT: 1 GBPS
$330
Or check other our plans for dedicated servers with GPU
If interested - contact us using DM at LET, live chat at https://vsys.host/, or email [email protected] to get a discounted price, kindly mention that you are from the LET!