about summary refs log tree commit diff
path: root/README.md
diff options
context:
space:
mode:
authorPjotr <wrk@bosa.lan>2026-10-22 09:02:53 +0000
committerPjotr <wrk@bosa.lan>2026-10-22 09:02:53 +0000
commit4c692148e6de551ef7c83677b5f1e5ca11e05df7 (patch)
treee573ca630ad74810e6e92035936905777486faef /README.md
parent07298ef6b812cd07a7c505fec12cc765cfaf319b (diff)
downloadhmf-guix-cwl-main.tar.gz
docs: Record the GPU investigation and its two blockers. main
The host does have an RTX 5080 (driver 610.43.02), but it cannot be used
from here and would not work with the packaged DJL stack anyway:

- the sandbox has no nvidia device nodes (/dev is a private tmpfs with
  only null/zero/full/random/urandom/tty/fuse), CapEff=0 so mknod is
  denied, and there is no libcuda.so* anywhere; only the kernel module is
  visible through /proc/driver/nvidia;
- DJL's newest CUDA native is pytorch-native-cu124 (PyTorch 2.5.1), and
  the libtorch_cuda.so in our cu121 bundle carries sm_35..sm_90 kernels
  only - no sm_100/sm_120, which is what Blackwell needs (CUDA 12.8+).

The plausible route is written down for whoever has the device: DJL 0.39
(pytorch-jni 2.7.1; all 17 DJL classes v-chord/v-cuppa import are
unchanged) plus a system libtorch 2.7.1+cu128 via PYTORCH_LIBRARY_PATH.
README's v-chord paragraph now states the same facts instead of "wait for
a cu128 runtime".
Diffstat (limited to 'README.md')
-rw-r--r--README.md16
1 files changed, 11 insertions, 5 deletions
diff --git a/README.md b/README.md
index 4b01029..a759e92 100644
--- a/README.md
+++ b/README.md
@@ -281,11 +281,17 @@ The scores are meaningless biologically (the circos PNG is a stub), but
 every stage of the DJL pipeline exercises: `libtorch` dlopen, TorchScript
 deserialisation, forward pass, output translation.
 
-The GPU leg needs a DJL runtime with kernels for the installed card.
-CUDA 12.1 (the newest DJL native published for PyTorch 2.2.2) has no
-`sm_120` images, so RTX 50-series (Blackwell) cards fail with *"no kernel
-image is available for execution on the device"*; either wait for a cu128
-runtime or point `PYTORCH_LIBRARY_PATH` at a system libtorch.
+The GPU leg needs a DJL runtime with kernels for the installed card, and
+no DJL release has them for RTX 50-series (Blackwell, `sm_120`): the
+newest native DJL publishes is `pytorch-native-cu124` 2.5.1, i.e. CUDA
+12.4, whose `libtorch_cuda.so` carries `sm_35` … `sm_90` kernels only
+(checked against the 2.2.2/cu121 bundle here).  Blackwell needs CUDA 12.8+
+(PyTorch 2.7+), so the card fails with *"no kernel image is available for
+execution on the device"*.  The plausible route is a newer DJL
+(`pytorch-jni` 2.7.1 exists for DJL 0.34–0.39, and every DJL class v-chord
+uses is unchanged in 0.39) plus a system libtorch 2.7.1+cu128 pointed at
+by `PYTORCH_LIBRARY_PATH` — that has to be tried on a host where the GPU
+device is actually visible.
 
 ## Related