|
Very low CUDA GEMM throughput on RTX Spark N1X (Windows, PyTorch 2.15 cu134) – ~1.7 TFLOPS FP32, slower than an RTX 4060 Laptop
|
|
2
|
84
|
November 16, 2026
|
|
Faster Scientific Simulations with New FP64 Algorithms in NVIDIA cuBLAS
|
|
0
|
22
|
November 13, 2026
|
|
Deterministic Xid 120 GSP kernel panic on RTX PRO 6000 Blackwell under sustained cuBLAS workload (595.84)
|
|
29
|
3173
|
November 13, 2026
|
|
Jetson Thor R38.2.1: ~13% GEMV timing shift with unchanged sampled GPU/EMC clocks — how can CUDA kernel benchmarks be made reproducible?
|
|
3
|
112
|
November 12, 2026
|
|
DGX Spark / GB10: ACPI TGPU vs nvidia-smi, continuous thermal limits and expected fan response
|
|
0
|
133
|
November 12, 2026
|
|
cuDSS 1.0.0: silent wrong answers in multi-RHS dense solve and RHS pointer handling (Windows)
|
|
2
|
99
|
November 11, 2026
|
|
RTX 4060 Laptop: BF16 torch.bmm illegal memory access across CUDA 12.4 and 13.0
|
|
3
|
77
|
November 10, 2026
|
|
A silent bug in cuSolverDx heev when exactly one eigenvalue is exactly zero
|
|
5
|
129
|
November 2, 2026
|
|
cublasStrmv 3-17x slower than hand-written CUDA/OpenCL/WebGPU in non-transpose configs
|
|
3
|
91
|
October 28, 2026
|
|
DGPP: a GB10-optimized C++/CUDA inference engine
|
|
112
|
8629
|
October 21, 2026
|
|
Two models on one DGX Spark: Qwen3.6-35B NVFP4 + Nemotron 3.5 Lightning with DSpark, numbers under shared load and a fix for cuDNN FP8 crashes
|
|
0
|
282
|
October 21, 2026
|
|
Silent CUDA Compute Corruption on RTX PRO 6000 Blackwell
|
|
9
|
627
|
October 16, 2026
|
|
cuBLASLt returns silently wrong FP16/BF16/FP8 GEMM results on Hopper under CUDA_MPS_ACTIVE_THREAD_PERCENTAGE (algoId 66, cluster kernels)
|
|
2
|
128
|
October 6, 2026
|
|
RTX 5060 Ti (SM120): GPU engine hang / device loss under sustained low-precision tensor-core workloads — reproducible in ~2.5 minutes
|
|
0
|
149
|
September 30, 2026
|
|
TensorFlow 2.21 GPU wheel for JetPack 7.2 / Jetson Orin
|
|
2
|
92
|
September 29, 2026
|
|
RTX PRO 4500 Blackwell – repeated TDR (117/141/1B8) and “GPU is lost” during LLM/CUDA workloads on Windows 11
|
|
0
|
201
|
September 26, 2026
|
|
RTX 5080 (GB203): 29× Xid 79 on one card — reproduced across Linux open modules (580/595/610) AND Windows (610.88), PSU rail measured steady through t
|
|
7
|
653
|
September 24, 2026
|
|
TensorFlow 2.21 GPU wheel for JetPack 7.2 / Jetson Orin
|
|
0
|
52
|
September 22, 2026
|
|
Crash when trying to use Opencv Cuda functions on Jetson
|
|
6
|
140
|
September 21, 2026
|
|
CUDA Graph updates report success but retain stale cluster dimensions on driver 580.159.03
|
|
0
|
76
|
September 18, 2026
|
|
Blackwell 6000 Pro Workstation - SW Power Cap, stuck at 577Mhz, claims 600 Watts but 35c - PyTorch/cuBLAS GEMM testing
|
|
1
|
472
|
September 18, 2026
|
|
RTX PRO 4000 Blackwell mixed-precision training: FP16/BF16, FP8, FP6, FP4 and FP32 accumulation
|
|
0
|
206
|
September 11, 2026
|
|
[580.105.08] cuMemSetAccess returns OOM near 512K aggregate VMM mappings across GPUs with free VRAM
|
|
0
|
122
|
September 7, 2026
|
|
GB10 / DGX Spark: CUDNN_FE failure 11 and CUBLAS_STATUS_INTERNAL_ERROR under batch load - fixed by driver 580.173
|
|
1
|
311
|
August 26, 2026
|
|
Call to cublasDgetrsBatched from a Fortran code fails
|
|
2
|
90
|
August 26, 2026
|
|
Inkling-NVFP4 (975B) on 8× DGX Spark
|
|
2
|
368
|
August 18, 2026
|
|
GPU Utilization Bottleneck: Single-Process with 8 Streams vs. Multi-Process with 1 Stream each (DeepStream 7.1 / PyServiceMaker)
|
|
14
|
589
|
August 17, 2026
|
|
NVIDIA nvmath-python v1.0 is now GA
|
|
0
|
107
|
July 31, 2026
|
|
INT8 Tensor Core corruption on Turing sm_75 cuBLAS llama.cpp BPE tokenizer
|
|
3
|
500
|
July 31, 2026
|
|
Green Context SM Partitioning on RTX PRO 4000 Blackwell — Hard 48 SM Cap and DRAM Bandwidth Contention
|
|
0
|
235
|
July 29, 2026
|