Pulse report — Nvidia DGX Spark¶

Submitted 2026-05-15T17:27:40+00:00 · fingerprint 5c832197980607c5

Summary¶

26 ok · 3 skipped · 0 failed

benchmark	status	duration	summary
`cpu_python`	ok	0.20s	sieve 219.76 Mops/s, arith 26.50 Mops/s
`cpu_hash`	ok	4.00s	SHA-256 single-thread (16 MiB chunks): 2370 MiB/s
`cpu_compress`	ok	2.89s	zlib c/d 61/337 MiB/s, lzma 14 MiB/s, zlib ratio 0.66
`sqlite_oltp`	ok	8.03s	WAL mode: inserts=73491 selects=3825 updates=70435 ops/s
`memory`	ok	2.05s	memcpy 26350 MiB/s (ctypes), 1MiB alloc 208857/s
`disk_seq`	ok	0.28s	seq write 1836 MiB/s, seq read 1888 MiB/s (256 MiB, O_DIRECT)
`disk_rand`	ok	7.67s	random 4K read: 2607 IOPS (O_DIRECT)
`network_download`	ok	3.20s	max 9.2 MiB/s (78 Mbps), median 9.0 MiB/s, σ=0.3 (3 samples)
`llm_inference`	ok	5.67s	tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf on llama-cpp-python (20 threads): pp512=233.9 tok/s · tg128=38.3 tok/s
`gpu_compute`	ok	0.74s	NVIDIA CUDA / NVIDIA GB10 (pyopencl-naive): 232.7 GFLOPS @ 2048^3
`sysbench_cpu`	ok	30.01s	sysbench cpu @ 20t, prime=20000: 32778 events/s
`sysbench_memory`	ok	0.07s	sysbench memory read: 67374 MiB/s
`sysbench_threads`	ok	20.01s	sysbench threads @ 16t: 7188 events/s
`stream`	ok	0.64s	STREAM Triad 114.7 GB/s (Copy 115.7 · Scale 119.4 · Add 115.8) · 1144 MiB working set
`fio_disk`	ok	61.34s	4K randread 583260 IOPS · 1M seqwrite 11494 MiB/s
`stress_ng_cpu`	ok	30.07s	stress-ng cpu @ 20t (matrixprod): 2199 bogo-ops/s
`stress_ng_memory`	ok	30.05s	stress-ng memrate: 7 bogo-ops/s
`7zip`	ok	19.59s	compress 100 MIPS, decompress 8443 MIPS
`openssl_speed`	ok	18.00s	AES-256-GCM (16K blocks): 7719 MiB/s
`c_ray`	ok	5.97s	c-ray-mt 7680x4320 rays=16 threads=20: 4.72 s
`ffmpeg_x264_encode`	ok	3.19s	x264 encode 900 frames @ 1080p in 3.2s = 282.4 fps (last reported: 291.0 fps)
`blender_bmw27`	ok	133.47s	BMW27 (CYCLES, 100% resolution): 132.8 s (wall: 133.5 s)
`glmark2`	skipped	0.00s	no DISPLAY/WAYLAND_DISPLAY — needs a graphical session
`docker`	skipped	0.04s	/usr/bin/docker info failed (daemon not running or no permission): permission denied while trying to connect to the dock
`compile_sqlite`	ok	12.34s	sqlite3.c compiled in 12.34s (20732 lines/s) using cc
`compile_cpp`	ok	0.96s	json.hpp compiled in 0.96s (25728 header lines/s) using c++
`compile_java`	ok	10.26s	javac compiled 500 classes / 786000 LOC in 10.25s (76647 lines/s)
`compile_redis`	ok	6.57s	redis built in 6.6s with -j8 (151 object files)
`browser_js`	skipped	0.02s	no chromium-family browser found

Subscores¶

Geomean of the per-bench headline metric within each category, normalised against the reference baseline (100 = parity).

category	score
compress	0.1
cpu	83.0
crypto	96.5
disk	16.3
gpu	1.9
media	141.2
memory	444.3
ml	47.9
render	87.5

System¶

field	value
platform	`linux/ubuntu/aarch64`
CPU	Cortex-X925 (20 threads)
RAM	121.7 GiB
motherboard	NVIDIA P4242
governor	`performance`
NUMA	1 node(s)
GPU	nvidia: NVIDIA GB10 (unified memory, drv 580.142)
Submitter	Patrik maly.patrik007@gmail.com

Detail¶

`cpu_python` — cpu¶

status: ok · duration: 0.20s

sieve 219.76 Mops/s, arith 26.50 Mops/s

metric	value	unit
sieve_throughput	219.8	Mops/s
arith_throughput	26.5	Mops/s
primes_found	1.489e+05	primes

`cpu_hash` — cpu¶

status: ok · duration: 4.00s

SHA-256 single-thread (16 MiB chunks): 2370 MiB/s

metric	value	unit
sha256_throughput	2370	MiB/s
chunk_size_mib	16	MiB

`cpu_compress` — cpu¶

status: ok · duration: 2.89s

zlib c/d 61/337 MiB/s, lzma 14 MiB/s, zlib ratio 0.66

metric	value	unit
zlib_compress	60.62	MiB/s
zlib_decompress	336.6	MiB/s
lzma_compress	14.11	MiB/s
zlib_ratio	0.665	ratio

`sqlite_oltp` — database¶

status: ok · duration: 8.03s

WAL mode: inserts=73491 selects=3825 updates=70435 ops/s

metric	value	unit
inserts_per_s	7.349e+04	ops/s
selects_per_s	3825	ops/s
updates_per_s	7.044e+04	ops/s
txn_per_s	1.478e+05	ops/s

`memory` — memory¶

status: ok · duration: 2.05s

memcpy 26350 MiB/s (ctypes), 1MiB alloc 208857/s

metric	value	unit
memcpy	2.635e+04	MiB/s
alloc_1mib_per_sec	2.089e+05	allocs/s

`disk_seq` — disk¶

status: ok · duration: 0.28s

seq write 1836 MiB/s, seq read 1888 MiB/s (256 MiB, O_DIRECT)

metric	value	unit
seq_write	1836	MiB/s
seq_read	1888	MiB/s
file_size_mib	256	MiB

`disk_rand` — disk¶

status: ok · duration: 7.67s

random 4K read: 2607 IOPS (O_DIRECT)

metric	value	unit
rand_read_iops	2607	IOPS
rand_read_throughput	10.19	MiB/s
file_size_mib	512	MiB

`network_download` — network¶

status: ok · duration: 3.20s

max 9.2 MiB/s (78 Mbps), median 9.0 MiB/s, σ=0.3 (3 samples)

metric	value	unit
download_max_mibps	9.247	MiB/s
download_median_mibps	8.962	MiB/s
download_max_mbps	77.57	Mbps
download_median_mbps	75.18	Mbps
download_stdev_mibps	0.3229	MiB/s
download_samples	3	n

`llm_inference` — ml¶

status: ok · duration: 5.67s

tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf on llama-cpp-python (20 threads): pp512=233.9 tok/s · tg128=38.3 tok/s

metric	value	unit
llm_pp512_tokens_per_sec	233.9	tok/s
llm_tg128_tokens_per_sec	38.35	tok/s

`gpu_compute` — gpu¶

status: ok · duration: 0.74s

NVIDIA CUDA / NVIDIA GB10 (pyopencl-naive): 232.7 GFLOPS @ 2048^3

metric	value	unit
gpu_sgemm_gflops	232.7	GFLOPS
gpu_matmul_size	2048	N

`sysbench_cpu` — cpu¶

status: ok · duration: 30.01s

sysbench cpu @ 20t, prime=20000: 32778 events/s

metric	value	unit
sysbench_cpu_events_per_sec	3.278e+04	events/s
sysbench_cpu_threads	20	threads

`sysbench_memory` — memory¶

status: ok · duration: 0.07s

sysbench memory read: 67374 MiB/s

metric	value	unit
sysbench_memory_read	6.737e+04	MiB/s

`sysbench_threads` — cpu¶

status: ok · duration: 20.01s

sysbench threads @ 16t: 7188 events/s

metric	value	unit
sysbench_threads_events_per_sec	7188	events/s
sysbench_threads_count	16	threads

`stream` — memory¶

status: ok · duration: 0.64s

STREAM Triad 114.7 GB/s (Copy 115.7 · Scale 119.4 · Add 115.8) · 1144 MiB working set

metric	value	unit
stream_copy_gb_s	115.7	GB/s
stream_scale_gb_s	119.4	GB/s
stream_add_gb_s	115.8	GB/s
stream_triad_gb_s	114.7	GB/s

`fio_disk` — disk¶

status: ok · duration: 61.34s

4K randread 583260 IOPS · 1M seqwrite 11494 MiB/s

metric	value	unit
fio_randread_iops	5.833e+05	IOPS
fio_randread_throughput	2278	MiB/s
fio_seqwrite_throughput	1.149e+04	MiB/s

`stress_ng_cpu` — cpu¶

status: ok · duration: 30.07s

stress-ng cpu @ 20t (matrixprod): 2199 bogo-ops/s

metric	value	unit
stress_ng_cpu_bogo_ops_per_sec	2199	bogo-ops/s
stress_ng_cpu_threads	20	threads

`stress_ng_memory` — memory¶

status: ok · duration: 30.05s

stress-ng memrate: 7 bogo-ops/s

metric	value	unit
stress_ng_memrate_bogo_ops_per_sec	6.76	bogo-ops/s

`7zip` — compress¶

status: ok · duration: 19.59s

compress 100 MIPS, decompress 8443 MIPS

metric	value	unit
7z_compress_rating	100	MIPS
7z_decompress_rating	8443	MIPS

`openssl_speed` — crypto¶

status: ok · duration: 18.00s

AES-256-GCM (16K blocks): 7719 MiB/s

metric	value	unit
aes256_gcm	7719	MiB/s

`c_ray` — render¶

status: ok · duration: 5.97s

c-ray-mt 7680x4320 rays=16 threads=20: 4.72 s

metric	value	unit
c_ray_render_time_s	4.722	s
c_ray_threads	20	threads

`ffmpeg_x264_encode` — media¶

status: ok · duration: 3.19s

x264 encode 900 frames @ 1080p in 3.2s = 282.4 fps (last reported: 291.0 fps)

metric	value	unit
ffmpeg_x264_encode_fps	282.4	fps
ffmpeg_x264_output_mib	22.87	MiB

`blender_bmw27` — render¶

status: ok · duration: 133.47s

BMW27 (CYCLES, 100% resolution): 132.8 s (wall: 133.5 s)

metric	value	unit
blender_bmw27_render_seconds	132.8	s

`glmark2` — gpu¶

status: skipped · duration: 0.00s

skipped: no DISPLAY/WAYLAND_DISPLAY — needs a graphical session

`docker` — container¶

status: skipped · duration: 0.04s

skipped: /usr/bin/docker info failed (daemon not running or no permission): permission denied while trying to connect to the docker API at unix:///var/run/docker.sock

`compile_sqlite` — compile¶

status: ok · duration: 12.34s

sqlite3.c compiled in 12.34s (20732 lines/s) using cc

metric	value	unit
compile_time	12.34	s
compile_throughput	2.073e+04	lines/s
download_time	2.476	s

`compile_cpp` — compile¶

status: ok · duration: 0.96s

json.hpp compiled in 0.96s (25728 header lines/s) using c++

metric	value	unit
cpp_compile_time	0.9626	s
cpp_compile_throughput	2.573e+04	lines/s
download_time	0.4226	s

`compile_java` — compile¶

status: ok · duration: 10.26s

javac compiled 500 classes / 786000 LOC in 10.25s (76647 lines/s)

metric	value	unit
javac_time	10.25	s
javac_lines_per_sec	7.665e+04	lines/s
javac_classes_per_sec	48.76	classes/s

`compile_redis` — compile¶

status: ok · duration: 6.57s

redis built in 6.6s with -j8 (151 object files)

metric	value	unit
redis_build_time	6.565	s
redis_objs_per_sec	23	objs/s
redis_jobs	8	threads
download_time	0.5063	s

`browser_js` — browser¶

status: skipped · duration: 0.02s

skipped: no chromium-family browser found

Pulse report — Nvidia DGX Spark¶

Summary¶

Subscores¶

System¶

Detail¶

cpu_python — cpu¶

cpu_hash — cpu¶

cpu_compress — cpu¶

sqlite_oltp — database¶

memory — memory¶

disk_seq — disk¶

disk_rand — disk¶

network_download — network¶

llm_inference — ml¶

gpu_compute — gpu¶

sysbench_cpu — cpu¶

sysbench_memory — memory¶

sysbench_threads — cpu¶

stream — memory¶

fio_disk — disk¶

stress_ng_cpu — cpu¶

stress_ng_memory — memory¶

7zip — compress¶

openssl_speed — crypto¶

c_ray — render¶

ffmpeg_x264_encode — media¶

blender_bmw27 — render¶

glmark2 — gpu¶

docker — container¶

compile_sqlite — compile¶

compile_cpp — compile¶

compile_java — compile¶

compile_redis — compile¶

browser_js — browser¶