{"id":912031,"date":"2026-07-21T22:55:11","date_gmt":"2026-07-21T22:55:11","guid":{"rendered":"https:\/\/cri-one.com\/blog\/2026\/07\/21\/case-study-wonderphi-compute-burn-in-on-a-dual-xeon-workstation\/"},"modified":"2026-07-21T23:15:22","modified_gmt":"2026-07-21T23:15:22","slug":"case-study-wonderphi-compute-burn-in-on-a-dual-xeon-workstation","status":"publish","type":"post","link":"https:\/\/cri-one.com\/blog\/2026\/07\/21\/case-study-wonderphi-compute-burn-in-on-a-dual-xeon-workstation\/","title":{"rendered":"Case Study \u2014 WonderPhi Compute Burn-In on a Dual-Xeon Workstation"},"content":{"rendered":"

Software acceleration for existing hardware.<\/p>\n

Theory is cheap. We measured WonderPhi Compute on real hardware \u2014 a workstation with dual Intel Xeon Gold 6138T processors (40 physical cores, 80 logical), an NVIDIA A10M GPU with 19 GB of VRAM, running Windows 11 Pro. Every number below is wall-clock time, single run, cold cache. No averaging. No cherry-picking.<\/p>\n

The Machine<\/h2>\n\n\n\n\n\n\n
CPU<\/td>\n2x Intel Xeon Gold 6138T @ 2.00 GHz \u2014 40 physical \/ 80 logical cores<\/td>\n<\/tr>\n
GPU<\/td>\nNVIDIA A10M \u2014 19 GB VRAM, Ampere architecture, 7th-gen NVENC<\/td>\n<\/tr>\n
OS<\/td>\nWindows 11 Pro<\/td>\n<\/tr>\n
WPC Config<\/td>\n64 workers (cap 64, cores 80), GPU exposed as $WPC_GPU=1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

This isn’t a cloud instance. This is a machine sitting on a desk, running production workloads around the clock, with WonderPhi’s three watchers armed and autonomous.<\/p>\n

Test 1 \u2014 SHA-256 Hashing (Pure CPU Parallel)<\/h2>\n

The simplest possible benchmark: hash the same 500 MB file 40 times. CPU-bound, memory-bandwidth-sensitive.<\/p>\n\n\n\n\n\n
Serial (bash for-loop)<\/td>\n96.61 seconds<\/strong><\/td>\n<\/tr>\n
WPC parallel (64 workers)<\/td>\n6.84 seconds<\/strong><\/td>\n<\/tr>\n
Speedup<\/td>\n14.1x<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Why not the full 64x? All 40 threads read the same file concurrently \u2014 memory bandwidth saturates before CPU does. That’s a DDR4 bottleneck, not a WPC limit. Workloads with distinct files per task (like the 541-domain health check we ran) hit 40-50x<\/strong> because each worker reads independent bytes.<\/p>\n

Test 2 \u2014 1080p H.264 Video Encoding (CPU vs. GPU)<\/h2>\n

Same 3-minute 1080p30 source (5,400 frames), 3 parallel encodes on each side.<\/p>\n\n\n\n\n\n
CPU (libx264, CRF 23)<\/td>\n51.38s \u2014 315 fps aggregate<\/td>\n<\/tr>\n
GPU (A10M NVENC, CQ 23)<\/td>\n48.07s \u2014 337 fps aggregate<\/td>\n<\/tr>\n
GPU speedup<\/td>\n1.1x (6% faster)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

On 1080p H.264, this box’s 40-core CPU is competitive with the A10M. Use the GPU when you want to keep CPU free for other work<\/strong>, not for raw throughput.<\/p>\n

Test 3 \u2014 4K HEVC Encoding (CPU vs. GPU)<\/h2>\n

30-second 3840×2160 30fps source (900 frames), 2 parallel encodes.<\/p>\n\n\n\n\n\n
CPU (libx265 fast, CRF 28)<\/td>\n36.27s \u2014 50 fps aggregate<\/td>\n<\/tr>\n
GPU (A10M NVENC, CQ 28)<\/td>\n24.36s \u2014 74 fps aggregate<\/td>\n<\/tr>\n
GPU speedup<\/td>\n1.5x (33% faster)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

At 4K, the GPU pulls ahead. HEVC at this resolution is far heavier than H.264 at 1080p. NVENC is silicon \u2014 it doesn’t care about resolution the way software encoders do. The gap widens with resolution: at 8K, expect 3-5x.<\/p>\n

Test 4 \u2014 AV1 Encoding<\/h2>\n

An honest caveat: the A10M does not<\/strong> support AV1 hardware encoding. Ampere NVENC (7th gen, 2020) has H.264 + HEVC only. AV1 NVENC arrived with Ada Lovelace (2022). All AV1 results are CPU.<\/p>\n\n\n\n\n\n
libaom-av1 (reference encoder)<\/td>\n219.87s \u2014 4.1 fps<\/td>\n<\/tr>\n
libsvtav1 (fast encoder)<\/td>\n12.33s \u2014 73.0 fps<\/td>\n<\/tr>\n
Encoder choice speedup<\/td>\n17.8x<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Encoder choice matters more than parallelism on AV1. And WPC parallelism helps for multi-file batches: 6 videos in parallel finished 3.1x faster than serial \u2014 but libsvtav1 is already multi-threaded internally, so you get diminishing returns.<\/p>\n

The 541-Domain Health Check<\/h2>\n

We also ran a real-world operational test: a WPC plan with 541 independent curl<\/code> health checks against live domains. Results: 442 reachable, 99 unreachable, total wall time 17.1 seconds.<\/strong> That’s 541 network requests, each with a 5-second timeout, completed in under 18 seconds.<\/p>\n

Honest Takeaways<\/h2>\n\n\n\n\n\n\n\n\n\n\n
Workload<\/th>\nBest Option on This Box<\/th>\n<\/tr>\n<\/thead>\n
Batch hashing, checksumming, validation<\/td>\nWPC parallel CPU<\/strong> (14-50x)<\/td>\n<\/tr>\n
Batch web scraping \/ health checks<\/td>\nWPC parallel CPU<\/strong> (bounded by network)<\/td>\n<\/tr>\n
1080p H.264 encoding<\/td>\nEither \u2014 CPU or NVENC depending on what else is running<\/td>\n<\/tr>\n
4K HEVC encoding<\/td>\nA10M NVENC<\/strong> (1.5x, grows with resolution)<\/td>\n<\/tr>\n
AV1 encoding<\/td>\nlibsvtav1 on CPU<\/strong> (17.8x over libaom; no GPU AV1 HW)<\/td>\n<\/tr>\n
Real-time streaming<\/td>\nA10M NVENC<\/strong> \u2014 consistent fps regardless of CPU load<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Every claim on the product page is backed by a number we measured. That’s the difference between marketing and evidence.<\/p>\n

Get WonderPhi Compute \u2014 $99.99 \u2192<\/a><\/strong><\/p>\n

Browse the full WonderPhi product line<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Software acceleration for existing hardware. Theory is cheap. We measured WonderPhi Compute on real hardware \u2014 a workstation with dual Intel Xeon Gold 6138T processors (40 physical cores, 80 logical), an NVIDIA A10M GPU with 19 GB of VRAM, running Windows 11 Pro. Every number below is wall-clock time, single run, cold cache. No averaging. […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[362,353,357,1],"tags":[],"class_list":["post-912031","post","type-post","status-publish","format-standard","hentry","category-case-study","category-computing","category-product-insights","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/posts\/912031","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/comments?post=912031"}],"version-history":[{"count":1,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/posts\/912031\/revisions"}],"predecessor-version":[{"id":912034,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/posts\/912031\/revisions\/912034"}],"wp:attachment":[{"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/media?parent=912031"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/categories?post=912031"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cri-one.com\/blog\/wp-json\/wp\/v2\/tags?post=912031"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}