NVIDIA CUDA: results and suite pointers

host-only pointers

Why it exists

Find live CUDA parity/timing matrices after the small usage chapters.

What it is

Print Go v1.4.0 / welvet.cuda 0.79 pins and commands for w2a go_v_cuda and cuda_smoke. No GPU required.

Where live results live

This chapter is host-only. It does not claim timings from loading the page. Run the sibling matrices on a machine with an NVIDIA GPU:

cd ../../w2a && go test ./tests/go_v_cuda -count=1 -timeout 180m
cd ../../welvet.cuda/examples/cuda_smoke && go run .
cd welvet/examples && ./run_nvidia.sh

Engine pin: welvet.cuda 0.79, model ABI 1, Go binding nativecuda (Welvet v1.4.0). Platforms: Linux x86_64 + Windows amd64 for Mount; Linux amd64 for EnableRouting.

Requirements overview: chapter 89.

Go example

examples/95-nvidia-cuda-results/main.go

Run:cd welvet/examples/95-nvidia-cuda-results && source ../env.sh && go run .

// Pointers to CUDA validation matrices; no GPU run is implied.
package main

import "fmt"

func main() {
	fmt.Println("NVIDIA CUDA results pointers (not a live benchmark)")
	fmt.Println()
	fmt.Println("Engine pin: welvet.cuda 0.79 · model ABI 1 · Go binding: nativecuda (Welvet v1.4.0)")
	fmt.Println("Usage tour: 89 mount · 90 Mamba dtypes · 91 train modes · 92 SyncToGo/ENTITY · 93 JSON · 94 mixed")
	fmt.Println()
	fmt.Println("Live matrices (sibling checkouts):")
	fmt.Println("  cd ../../w2a && go test ./tests/go_v_cuda -count=1 -timeout 180m")
	fmt.Println("  cd ../../welvet.cuda/examples/cuda_smoke && go run .")
	fmt.Println()
	fmt.Println("Install shared library into Go:")
	fmt.Println("  cd ../../welvet.cuda && ./scripts/linux_install_go.sh")
	fmt.Println()
	fmt.Println("Small usage tour without the large matrix:")
	fmt.Println("  ../run_nvidia.sh")
}

Output

exit 0 · last run via go run .

NVIDIA CUDA results pointers (not a live benchmark)

Engine pin: welvet.cuda 0.79 · model ABI 1 · Go binding: nativecuda (Welvet v1.4.0)
Usage tour: 89 mount · 90 Mamba dtypes · 91 train modes · 92 SyncToGo/ENTITY · 93 JSON · 94 mixed

Live matrices (sibling checkouts):
  cd ../../w2a && go test ./tests/go_v_cuda -count=1 -timeout 180m
  cd ../../welvet.cuda/examples/cuda_smoke && go run .

Install shared library into Go:
  cd ../../welvet.cuda && ./scripts/linux_install_go.sh

Small usage tour without the large matrix:
  ../run_nvidia.sh

Validate (harness)

cd welvet/examples/95-nvidia-cuda-results && source ../env.sh && go run .