[HN Gopher] Every Flop Counts: Scaling a 300B LLM Without Premiu...
___________________________________________________________________
Every Flop Counts: Scaling a 300B LLM Without Premium GPUs
Author : bretpiatt
Score : 109 points
Date : 2025-03-24 12:48 UTC (4 days ago)
(HTM) web link (arxiv.org)
(TXT) w3m dump (arxiv.org)
| osti wrote:
| I think this is the one where they train LLM without NVIDIA
| GPU's.
| cavisne wrote:
| They talk about CUDA level tracing in their framework. I assume
| its just consumer GPU's that Nvidia say arent meant to be used
| in datacenters.
| flowerthoughts wrote:
| They never mention what hardware they're on.
|
| Table 1 is the closest thing. Device specs for six devices:
| 120-989 TFLOPS and 64-96 GB RAM.
|
| An RTX 5090 is about 105 TFLOPS.
|
| https://www.techpowerup.com/gpu-specs/geforce-rtx-5090.c4216
| rahen wrote:
| I'm pretty surprised by the claimed memory usage for 300B
| parameters (table 1). If we compare similar models:
|
| - Llama 3.1 with 405B parameters: 2 TB of memory (FP32), 500 GB
| (FP8)
|
| - DeepSeek R1 with 671B parameters: 1.3 TB (scaling linearly,
| around 600 GB for 300B parameters)
|
| Ling claims no more than 96 GB of memory, most likely for
| inference. That's far more than a 20% reduction. Am I missing
| something?
| fxtentacle wrote:
| Some of these models still produce great results with something
| low like 2.7 bits per variable.
| cavisne wrote:
| I think they only claim their "Ling-Lite" 17B model can fit on
| a single 96GB GPU, their 300B model needs 8 of them (768GB of
| HBM)
___________________________________________________________________
(page generated 2025-03-28 23:02 UTC)