Publish Sign in

cuDF GPU DataFrames

Official NVIDIA-authored guidance for cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O and…

You say
Install this skill Read the source first Free Written by NVIDIA · unverified publisher
Context cost
12.9k tokensestimated from the bundle, loaded when it triggers
Bundle
7 files · 51.8 kBtext throughout, nothing executable
Licence
CC-BY-4.0 AND Apache-2.0free to use
Last change
—no release on file
Servers it uses
Noneruns standalone

What it does

Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads.

Installed, it changes the agent in these ways.

What this skill changes about the agent is not written down here yet. The listing was collected from its source, and the description is in its own SKILL.md.

Expertise

Domain judgement the base model does not have.

cudfdataframespandasgpu
Filed under

Data Science & ML

The skill itself

This is the whole product. A skill is instructions the model reads, so there is nothing behind the listing you cannot see first — the front matter loads with every session, and the body below it loads when the skill triggers.

SKILL.md9.4 kB · 204 lines
--- name: accelerated-computing-cudf description: Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads. license: CC-BY-4.0 AND Apache-2.0 metadata: author: NVIDIA tags: - cudf - dataframes - pandas - dask-cudf - etl ---
15# cuDF & dask-cuDF Implementer's Guide
16
17## Compatibility
18
19- Release tracked by this skill: 26.04.
20- Requires NVIDIA Volta or newer on CUDA 12, or Turing or newer on CUDA 13. Release 26.04 supports CUDA 12.2-12.9 with driver 535+ or CUDA 13.0-13.1 with driver 580+, and Python 3.11-3.14. cuDF sweet spot: >100K rows.
21
22## Naming
23
24Use NVIDIA library-first wording in user-facing answers. Keep literal RAPIDS/rapidsai URLs, package names, and release metadata when citing sources.
25
26## Role
27
28You are a cuDF expert helping an implementer work with GPU DataFrames. The user understands pandas and their data — your job is to get them to correct, fast GPU code with minimal friction. Choose the path from the user's intent: cudf.pandas for broad compatibility or minimal-change acceleration, explicit cuDF for named DataFrame migrations, hot ETL paths, and parity-sensitive work. Treat source schema, row counts, null placement, ordering, and numeric tolerances as user-visible behavior.
29
30## Critical Rules
31
321. **Choose the right cuDF path.** Use cudf.pandas for broad compatibility or minimal-change acceleration. Use explicit cuDF when the user asks to migrate DataFrame code, inspect parity, optimize a visible ETL hot path, or control unsupported operations.
332. **Size gate: 100K rows minimum.** Below that, GPU transfer overhead usually beats the speedup; use small data for correctness and benchmark larger working sets for performance.
343. **Keep conversions at boundaries.** Use .to_pandas(), .values, or .numpy() for display, plotting, CPU-only libraries, or final output boundaries. Keep intermediate ETL data on GPU.
354. **Float32 is your friend.** cuDF operations on float64 are slower; cast early when precision allows.
365. **Validate semantics on representative slices.** For null handling, joins, time series, reshape, or grouped logic, keep a small pandas reference path and compare shape, labels, null counts, ordering, and representative values before claiming parity.
376. **For data > GPU memory**, move to dask-cuDF with enable_cudf_spill=True. See references/dask-cudf-patterns.md.
38
39## Three Paths to GPU DataFrames
40
41### Path 1: cudf.pandas Accelerator (Compatibility / Minimal Change)
42
43Use when the user needs a small code change, third-party pandas compatibility,
44or one code path that can keep running while unsupported operations fall back.
45
46**Jupyter/IPython:**
47```python
48%load_ext cudf.pandas
49import pandas as pd # now GPU-backed; falls back silently for unsupported ops
50```
51
52**Script:**
53```bash
54python -m cudf.pandas my_script.py
55```
56
57**With multiprocessing:**
58```python
59import cudf.pandas
60cudf.pandas.install() # must come BEFORE pandas import, before Pool creation
61from multiprocessing import Pool
62```
63
64Confirm acceleration with the cudf.pandas profiler before claiming speedup.
65For notebook, CLI, and stats examples, read
66references/cudf-pandas-accelerator.md. If the profile shows the hot path
67running on CPU, use Path 2 for explicit cuDF control.
68
69### Path 2: Explicit cuDF API
70
71For full control, hot-path optimization, named DataFrame migrations, and
72parity-sensitive operations:
73
74```python
75import cudf
76
77# Read data directly to GPU
78df = cudf.read_parquet("data.parquet")
79
80# Operations mirror pandas
81result = df.groupby("key")["value"].sum()
82merged = df.merge(lookup, on="id", how="left")
83filtered = df[df["amount"] > 1000]
84
85# String operations
86df["clean"] = df["name"].str.strip().str.lower()
87
88# To check API coverage before committing to migration:
89# See references/api-patterns.md for known gaps and workarounds
90```
91
92**Keep data on GPU end-to-end.** Only call .to_pandas() at the very end for display or CPU or non-GPU handoff.
93
94Prefer explicit cuDF for tasks involving read_csv/read_parquet, joins,
95groupby, reshape, nullable types, fillna/where, time buckets, rolling
96windows, or CPU/GPU parity checks. Add a small CPU/GPU validation path when
97semantics matter instead of relying on successful execution alone.
98
99For pandas code with null handling, reshape, or time-series behavior, read
100references/api-patterns.md for the relevant semantic checklist before
101rewriting. A cudf.pandas bootstrap is enough for a minimal-change request; an
102implementation request should make the hot path explicit and observable.
103
104For reshape-heavy pandas code (pivot_table, melt, stack/unstack,
105crosstab), keep the source schema as part of the contract: index labels,
106column labels or levels, fill_value, aggfunc, margins, and normalization.
107Use explicit cuDF where the equivalent is supported; use cudf.pandas or a
108narrow compatibility boundary when exact pandas reshape semantics matter more
109than rewriting every operation. Add a small pandas-reference parity check for
110shape, labels, and representative values before finalizing. See
111references/api-patterns.md.
112
113### Path 3: dask-cuDF (Multi-GPU / Large Data)
114
115When dataset exceeds GPU memory. See references/dask-cudf-patterns.md for full patterns.
116
117```python
118from dask_cuda import LocalCUDACluster
119from dask.distributed import Client
120import dask_cudf
121
122cluster = LocalCUDACluster(enable_cudf_spill=True) # one worker per GPU
123client = Client(cluster)
124
125ddf = dask_cudf.read_parquet("s3://bucket/data/*.parquet")
126result = ddf.groupby("key").agg({"value": "sum"}).compute()
127```
128
129## Memory Management
130
131**Enable spill before OOM happens** (not after):
132```python
133import cudf
134cudf.set_option("spill", True) # spill to host RAM when GPU is full
135```
136
137**RMM pool allocator** (reduces cudaMalloc overhead in pipelines with many allocations):
138```python
139import rmm
140rmm.set_current_device_resource(rmm.mr.CudaAsyncMemoryResource())
141# Must be called BEFORE any cuDF operations
142```
143
144| GPU Free vs Dataset | Strategy |
145|---|---|
146| Free > 2× dataset | Single GPU cuDF |
147| Free 1–2× dataset | cuDF + cudf.set_option("spill", True) |
148| Dataset > GPU mem | dask-cuDF |
149| Dataset > node mem | dask-cuDF + multi-node (see accelerated-computing-mpf) |
150
151## Troubleshooting
152
153**No speedup vs pandas:**
154- Data < 100K rows? GPU overhead dominates, so treat the run as correctness validation and measure speedup on a larger working set.
155- Run %%cudf.pandas.profile — high CPU % means many fallbacks. Identify and fix those ops.
156- Check references/api-patterns.md for known gaps.
157
158**OOM (CUDA out of memory):**
1591. Enable spill: cudf.set_option("spill", True)
1602. If allocator fragmentation or repeated allocation overhead is visible, use the accelerated-computing-rmm memory-resource setup guidance before GPU allocations
1613. Still failing: move to dask-cuDF
162
163**AttributeError / NotImplementedError:**
164- Check references/api-patterns.md for the specific operation
165- Keep that one operation on CPU at a narrow boundary and continue the supported pipeline on GPU
166- Use .to_pandas() only for the unsupported op, then .from_pandas() back
167
168**Wrong results vs pandas:**
169- Null/NaN handling differs: cuDF uses <NA> (nullable) by default, pandas uses NaN. See references/api-patterns.md.
170- Sort stability: cuDF sort is not guaranteed stable unless stable=True is passed
171- If the difference is due to floating point differences, try casting to higher precision floats (e.g. float64 instead of float32). If the results are still different, stop. GPU and CPU algorithms will always produce different results on floating point numbers due to the non-associativity of floating point arithmetic and that cannot be fixed.
172
173## Nullable and Fill Semantics
174
175When the user explicitly cares about pandas nullable dtypes, fillna,
176where/mask, or grouped null behavior, treat parity checks as part of the
177implementation. See references/api-patterns.md for nullable dtype examples.
178
179- Preserve nullable integer/string columns instead of filling them with sentinel
180 values unless the source code already did that.
181- Keep where/mask semantics when they encode a condition. Use broad
182 fillna only when the condition is exactly null-only.
183- Compare with to_pandas(nullable=True) when the pandas reference uses
184 nullable extension dtypes.
185- Put the parity check in a reusable helper next to the GPU path, so future
186 changes exercise the same nullable conversion and aggregation checks.
187- Validate row counts, null counts, mask truth tables, grouped aggregates, and
188 representative dtypes before claiming semantic parity.
189
190## Reference Files
191
192- references/cudf-pandas-accelerator.md — Profiling, fallback detection, cudf.pandas deep dive
193- references/api-patterns.md — Known API gaps, workarounds, semantic differences
194- references/dask-cudf-patterns.md — Multi-GPU patterns, best practices, partition tuning
195
196## External Documentation
197
198Use WebFetch to retrieve detailed API signatures, parameter descriptions, and examples on demand.
199
200- **cuDF Documentation:** https://docs.rapids.ai/api/cudf/stable/
201- **dask-cuDF API Reference:** https://docs.rapids.ai/api/dask-cudf/stable/api/
202- **GitHub:** https://github.com/rapidsai/cudf
203- **CHANGELOG:** https://github.com/rapidsai/cudf/blob/main/CHANGELOG.md
204
In the file
SKILL.md1,249 words
Files7
LicenceCC-BY-4.0 AND Apache-2.0
Why you can read it

Nothing in a skill executes. The client loads the text and the model follows it, so a skill can be audited the way a runbook is — by reading it.

What it costs in context

Skills are not billed by the call. They are paid for in context: every token the instructions occupy is a token your code, your diff and your conversation cannot use. Here is what this one takes and when it takes it.

≈100
always loaded
The name and description, so the model knows the skill exists and when to reach for it.
12,850
on trigger
The instruction body and 6 supporting files, read only when the skill fires.
6.5%
of a 200k window
Ten skills this size would take about 65% of the window before you open a file.
050k100k150k200k context window

12.9k tokens, estimated from the bundle at four bytes to the token, held for the rest of the session once it triggers. Heavy. Teams tend to install this one per project rather than globally, and load it only when the job comes up.

Servers bill, skills cost

A server charges by the month. A skill charges once per session, in context, and then keeps charging it for as long as the session lives.

Before and after

The same question, put to the same model twice: once as it comes, and once with these instructions loaded.

No worked example has been published for this skill yet.

Adoption
Installsnone yet
Ratingno reviews yet

The procedure it runs

The procedure has not been published here. It is in the skill’s own SKILL.md, which its author has not sent to the marketplace yet.

Prose, not code

These steps are written for a model to follow, not executed by a runtime. It can still be told to skip one, and it will say so when it does.

Servers it uses

None. This skill calls no MCP servers at all.

Everything it needs is in the instructions, so it works in a project with nothing connected — the model reads the file and changes how it works with what it can already reach.

It writes no files and reaches no network. All it changes is how the model reasons and writes.

What it asks for
Writes filesno
Network accessno

Read from the allowed-tools line of this skill’s own SKILL.md. A skill grants no permissions of its own — it can only ask for tools your client already has.

What it will not do

Every skill is narrow, and the useful ones say where they stop. These are the jobs this one is the wrong tool for.

What this skill is not for has not been published here. Nothing is implied by that: it is a section the author has not filled in.

What is in the bundle

7 files, 51.8 kB on disk. A bundle is text throughout: the instructions the model reads, plus the templates it fills in.

  • BENCHMARK.md4.3 kB
  • SKILL.md9.4 kB
  • skill-card.md4.1 kB
  • evals/evals.json16.2 kB
  • references/api-patterns.md7.0 kB
  • references/cudf-pandas-accelerator.md3.7 kB
  • references/dask-cudf-patterns.md7.1 kB
What is not in it

No dependencies and nothing executable: a skill is text the agent reads, so the bundle is 7 files you can review in full before installing. The CC-BY-4.0 AND Apache-2.0 licence covers the templates and examples as well as the instructions.

Install

Installing copies the bundle into your project. Nothing runs at install time — the files sit on disk until the model reads them.

# cuDF GPU DataFrames · 12.9k tokens when loaded npx mcprush@latest skill add nvidia/cudf-gpu-dataframes

Writes to .claude/skills/cudf-gpu-dataframes/ in the current project. Add --global to put it in your home directory instead, for every project.

Which clients pick it up on their own

A skill is a folder of text. A client with a skills folder reads it without being told; everywhere else the same text works, it is just handed to the model rather than found.

Claude Code.claude/skills/
Claude Desktop
ChatGPT
Cursor.cursor/skills/
VS Code.github/skills/
Codex CLI.agents/skills/
Gemini CLI.gemini/skills/
Grok.grok/skills/
Zed.agents/skills/
Windsurf.windsurf/skills/
Agent SDK.claude/skills/
HTTP / API
This release
Versionnot versioned
Publishedno release date on file
PriceFree
Referencenvidia/cudf-gpu-dataframes

Versions

Its author publishes no version number, so there is nothing here to pin to: what you install is the folder as it stands today. Instructions change more often than APIs do — a skill can be rewritten entirely without anything it depends on moving.

v
  • No earlier releases have been published to the marketplace.
Pinning

Nothing to pin to: this skill carries no version number of its own. What you install is what the folder holds on the day you install it.

Reviews

—no reviews yet · no installs yet

Nobody has reviewed this skill yet. The rating is the mean of the reviews written here, so there is none until somebody writes the first.

Who can post

Only accounts that have had the skill installed for fourteen days, so a review is written after living with it rather than after reading it. Publishers may reply once.

Who wrote it

NV
NVIDIA

Publishes on mcprush.

0 servers listed9 skills listednot claimed
Profile
Publisher
Servers0