diff --git a/Cargo.toml b/Cargo.toml
index 7473a0ef77..c9c3bf0d8f 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -45,7 +45,7 @@ candle-onnx = { path = "./candle-onnx", version = "0.8.0" }
 candle-transformers = { path = "./candle-transformers", version = "0.8.0" }
 clap = { version = "4.2.4", features = ["derive"] }
 criterion = { version = "0.5.1", default-features=false }
-cudarc = { package = "mistralrs_cudarc_fork", version = "0.12.2", features = ["std", "cublas", "cublaslt", "curand", "driver", "nvrtc", "f16", "cuda-version-from-build-system", "dynamic-linking"], default-features=false }
+cudarc = { version = "0.13.3", features = ["std", "cublas", "cublaslt", "curand", "driver", "nvrtc", "f16", "cuda-version-from-build-system", "dynamic-linking"], default-features=false }
 fancy-regex = "0.13.0"
 gemm = { version = "0.17.0", features = ["wasm-simd128-enable"] }
 hf-hub = { version = "0.3.3", package = "candle-hf-hub" }
diff --git a/candle-core/Cargo.toml b/candle-core/Cargo.toml
index 6440a81ce5..752e478d9b 100644
--- a/candle-core/Cargo.toml
+++ b/candle-core/Cargo.toml
@@ -40,8 +40,9 @@ criterion = { workspace = true }
 
 [features]
 default = []
-cuda = ["cudarc", "dep:candle-kernels", "float8/mistralrs_cudarc_fork"]
+cuda = ["cudarc", "dep:candle-kernels", "float8/cuda"]
 cudnn = ["cuda", "cudarc/cudnn"]
+nccl = ["cuda", "cudarc/nccl"]
 mkl = ["dep:libc", "dep:intel-mkl-src"]
 accelerate = ["dep:libc", "dep:accelerate-src"]
 metal = ["dep:metal", "dep:candle-metal-kernels"]