NVIDIA GPUs Info and Used Price

A simple dataframe (and how to collect the info) combining Nvidia GPU information from

• TechPowerUp
• GeekBench
• eBay

---

Repo overview

• data/nvidia_gpu_info.csv contains the final data, also available here

• main.ipynb contains the main notebook used to obtain the data

• analysis.ipynb shows how to query the GPU info dataframe with pandas

• data contains other partial csv tables

---

Example query

We're interested in a GPU with at least 12 GB of memory, a compute score of at least half of RTX 3090, costing less than $1000.00 used.

compute_score = (df.loc[df["Name"] == "RTX 3090"]["Performance Score"].values[0]) / 2.0
mask = (
    (df["Performance Score"] > compute_score)
    & (df["Memory Size (GB)"] > 12)
    & (df["Used Price (eBay US) 2023-03-15"] < 1000)
)
df.loc[mask].sort_values("Used Price (eBay US) 2023-03-15")

giving

	Unnamed: 0	Name	Memory Size (GB)	Performance Score	Used Price (eBay US) 2023-03-15	Score Per Price	Manufacturer	GPU Chip	Released	Bus	GPU clock	Memory clock	Memory Type	Memory Bus Width	Shaders	TMUs	ROPs
17	17	RTX A4000	16	126692	489.99	258.56	NVIDIA	GA104	Apr 12th, 2021	PCIe 4.0 x16	735 MHz	1750 MHz	GDDR6	256 bit	6144	192	96
13	13	TITAN RTX	24	138072	740	186.584	NVIDIA	TU102	Dec 18th, 2018	PCIe 3.0 x16	1350 MHz	1750 MHz	GDDR6	384 bit	4608	288	96
5	5	RTX 3090	24	204921	782.5	261.88	NVIDIA	GA102	Sep 1st, 2020	PCIe 4.0 x16	1395 MHz	1219 MHz	GDDR6X	384 bit	10496	328	112
10	10	RTX A4500	20	144819	799.99	181.026	NVIDIA	GA102	Nov 23rd, 2021	PCIe 4.0 x16	1050 MHz	2000 MHz	GDDR6	320 bit	7168	224	96
22	22	Quadro RTX 5000	16	108130	899.99	120.146	NVIDIA	TU104	Aug 13th, 2018	PCIe 3.0 x16	1620 MHz	1750 MHz	GDDR6	256 bit	3072	192	64