NVIDIA DGX-1 DGX OS 6 / Ubuntu 22.04 Xeon E5-2698v4 メモリ512GB GPU 128GB )RTX6000

Description: DescriptionThis product is listed upon the request of the owner. Please only purchase if you have thoroughly read the description and are fully satisfied with the details. [Latest Environment Configured] NVIDIA DGX-1 DGX OS 6 / Ubuntu 22.04 Xeon E5-2698v4 Memory 512GB GPU Memory 128GB (Search terms: RTX 5000 6000 8000 4090 P100 V100 A100 H100) NVIDIA DGX-1 DGX OS 6 / Ubuntu 22.04Ubuntu 22.04 LTS (jammy) + CUDA 12.2 (nvidia-driver-535-server)The product has been configured with the latest DGX OS 6 / Ubuntu 22.04 environment as of April 2024. This price is set under the assumption that it is a functional NVLink parallel GPU & CPU high-memory environment, not treated as junk. For those interested, even if this is an older generation, it may still be useful in scenarios where a large amount of physical memory is required alongside GPUs, such as for mid to large-scale LLM development. Depending on the model, for example, when running llama3-8B with Transformers using device_map="auto", the load is balanced across 8 GPUs via NVLink, sometimes delivering faster performance than a single GPU with the same or higher specs. As an inference machine or for training smaller models, it should still be quite useful. If there is room in your server room, this could serve as an environment for hosting and publishing LLMs of sizes between 7B-30B (and other uses too). We have actually used it this way for testing the performance of ultra-large quantized models and small-scale training, but due to certain constraints related to power supply and noise issues involving other servers in our office, we decided to part with it. If it doesn’t sell, it might be used as a sub-environment or cluster when the main environment's resources are insufficient. That’s why it's priced for those who want it. Depending on the shipping lot, the reality is that non-supported units may require considerable investigation, with a high price tag akin to a "junk lottery." Sometimes it may not be possible to create a DGX OS Ubuntu environment (this was the case when I first bought it, and it caused quite a bit of frustration). Without any guidance, configuring the environment can be disheartening. The original price exceeds 10 million yen, and I suspect that NVIDIA representatives were very supportive during maintenance, given the product's high cost. [Hardware Details] Xeon E5-2698v4 (2 CPUs) 20 cores, 40 threads, 2.20-3.60GHzMemory: 512GB (32GB x16)Disk: SSD 480GB + 1.92TB (less than DGX-1 default, additional storage possible)GPU: Tesla P100 128GB (16GB x8)Power: 200V 50/60Hz 10A (x4) *Only 200V use is possibleSize: 3U (W484 x D950 x H135 mm) / Approx. 60kgNote: Only the DGX-1 unit is included; nothing else is provided. Standard 200V server power taps or cables are required for operation.The exact state at initial shipment is unclear, but there are no fixing pins for the lid that hides the front disk mounts. However, the lid is magnetically secured, so it stays in place even without pins. There are also numerous scratches and stains. Due to its current placement, it's not rack-mounted but sitting on a desk, and there may be other minor external defects (though there seems to be no visible damage, but please be aware it is sold as-is). [Description] Now, to provide a bit more specific detail about the contents: The disks are set up in RAID0, and the DGX OS6 is configured with the latest environment, Ubuntu 22.04 LTS (jammy) + CUDA 12.2 (nvidia-driver-535-server). It will be handed over in its current state with the configured Ubuntu environment. Briefly explaining the condition, it is accessible via SSH login within our internal network, with 8 GPUs connected via NVLink. We have confirmed that LLM operates on the connected GPUs with PyTorch installed. Please note that not all DGX OS functions have been tested. The device was originally acquired as a used product with the disks individually initialized and in an unknown state of operation. It took over a year, including some downtime, to configure the environment from an initialized state to the latest DGX environment. Due to the premium price and the limited number of people who set up from initialization to the latest environment without support, there was almost no information available online. Moreover, even existing setup instructions were lacking, making the Ubuntu installation particularly challenging. [Shipping Information] The GPU modules will be shipped separately. Upon arrival, you will need to insert the separated GPU modules. This unit will be handed over in a state where the DGX OS and CUDA installation are completed, and PyTorch has been verified as operational, making it ideal for those who want to use a high-memory environment immediately with GPUs included. Again, please understand that no support will be provided after the unit is delivered and you can log in to the environment (as the installation method did not follow publicly available information). Since this is an expensive item, I have tried to provide a detailed explanation, but there may be inaccuracies as I have been handling this DGX-1 with trial and error. Please only bid if you agree to the terms that this is a used product and is sold as-is with no claims or returns.Please carefully check the photos and purchase only if you are satisfied.Ask if you can lower the price before you buy.If you return an item for any reason, you will only be refunded the value of the item minus the shipping costs for both shipping and return shipping.If you do not agree to these conditions, we will not be able to sell to you.By purchasing, you agree to these terms and conditions.ShippingItems are shipped by FedEx.Please allow 5 business days for handling time.Please allow 3 to 5 business days from the time of shipment for delivery. Please note that the number of days may vary depending on the situation.International Buyers - Please Note:Import duties, taxes, and charges are not included in the item price or shipping cost. These charges are the buyer's responsibility. Please check with your country's customs office to determine what these additional costs will be prior to bidding or buying.Thank you for your understanding.

Price: 17000 USD

Location: kyoto

End Time: 2024-10-05T21:07:12.000Z

Shipping Cost: 1200 USD