French
AMD has long dominated high-core workstations for a decade, with Threadripper PRO outperforming Intel’s previous-gen Xeon-W series capped at 60 cores. Intel’s new 600-series Granite Rapids-WS platform changes this, delivering up to 86 cores and 128 PCIe 5.0 lanes — powering HP’s new Z8 Fury G6i AI-focused workstation.
Positioned for AI workloads rather than traditional CAD tasks, the Z8 Fury G6i supports one Xeon 600-series CPU, up to four 300W RTX PRO 6000 Blackwell Max-Q GPUs (384GB total VRAM) or a single 600W high-performance RTX PRO 6000 GPU. It accommodates 2TB DDR5-6400 ECC memory, multiple M.2 NVMe drives, nine PCIe slots in a 54L chassis, and optional 5RU rack mounting for centralized deployment.
The review unit features a 64-core Xeon 696X CPU, dual RTX PRO 6000 Max-Q GPUs (192GB combined VRAM), 128GB DDR5 RAM, and a four-NVMe storage setup (one Gen5 boot drive + three Gen4 data drives). Priced starting at $7,900 (base) and $74,878 (review configuration), it typically offers discounted corporate volume pricing.
Key Specifications
Powered by the Intel W890 chipset, the Xeon 600 series spans 18-core to 86-core models with 200–350W TDP, up to 4.9GHz boost and 336MB L3 cache. The system supports 2TB max DDR5-6400 ECC memory, 104TB total storage (eight M.2 NVMe + SATA HDDs), and full Blackwell-generation RTX PRO GPU options alongside data center-grade A800. I/O includes front/rear USB arrays, optional 20Gbps USB-C and Thunderbolt 5 ports, optional 10GbE/25GbE networking and Wi-Fi 7. It features ISV certification, HP Wolf security suite, 80 Plus Platinum power supplies, 60% recycled materials, and configurable 1350W–2700W dual redundant PSUs.
Design, Build & Expandability
Adopting HP’s matte black professional workstation design, the tower features a full-height mesh grille for airflow, tool-free drive bays and GPU latches, and a foldable rear carrying handle. The robust chassis supports high-power multi-GPU setups, with dual hot-swappable PSUs adaptable to 120V/240V power environments.
Storage includes four heatsink-equipped Gen5 M.2 slots and two 3.5-inch SATA HDD bays. Expansion covers nine PCIe slots (four Gen5 x16) for quad-GPU deployment, plus a dedicated MCIO connector for high-speed network upgrades without occupying PCIe slots. Premium front I/O with dual 20Gbps USB-C is optional and equipped on the review unit.
Benchmark Test Setup
Testing compares the Z8 Fury G6i against two platforms: a 24-core Xeon 658x system with RTX 4090, and a 96-core AMD Threadripper 9995WX Dell Precision 7875 with dual RTX PRO 6000 GPUs and 512GB RAM. Tests cover AI inference, 3D rendering, system performance, storage, professional graphics, and LLM serving workloads.
Core Benchmark Performance
AI Computer Vision (Procyon)
The G6i scores 207 in CPU AI vision (16.5% lower than Xeon 658x, 31.8% higher than AMD Dell) and 1151 in GPU AI vision (29% lower than Dell’s dual-GPU setup), delivering balanced professional AI inference performance.
Blender 3D Rendering
CPU rendering is 90% faster than the low-core Xeon platform but 50–66% slower than the high-core AMD workstation. GPU rendering is nearly identical to the Dell system, with marginal ±1.3% variance across all scenes.
System & Storage Benchmarks
PCMark 10 scores 7742, trailing both comparison systems in general desktop productivity. It excels in Blackmagic RAW media processing, topping rivals with 311FPS 8K CPU decoding and 650FPS 8K GPU decoding. The primary Gen5 NVMe drive hits 8911.5MB/s read and 8166.5MB/s write, slightly trailing the Dell’s storage speed. 3DMark CPU multi-threaded performance falls between the two competitors, while storage and ray-tracing LuxMark results are nearly on par with the Dell workstation.
Specialized CPU Workloads
Y-Cruncher mathematical testing shows the G6i outperforms both rivals on small-scale datasets, while the higher-core AMD system leads large-scale multi-billion-digit calculations. The G6i dominates 7-Zip compression/decompression, achieving 344.96 GIPS total rating for a 38% lead over the baseline Xeon platform. V-Ray rendering scores are 7% lower than the Dell system with near-comparable professional visualization performance.
Professional Graphics & AI Video
In SPECviewperf 15, the G6i matches Dell in Blender, Enscape, medical and energy visualization workloads, while the AMD system leads CAD-focused tasks like Creo, CATIA and Unreal Engine. For Topaz Video AI, the Dell leads most heavy upscaling tasks, while the G6i excels in specific slow-motion interpolation scenarios.
vLLM LLM Serving Performance (Key Advantage)
Testing across diverse LLMs under balanced, prefill-heavy and decode-heavy workloads reveals the G6i’s core strength. With identical dual RTX PRO 6000 GPUs and no NVLink support on either platform, the G6i’s full PCIe Gen5 x16 GPU slots and superior Xeon 6 CPU scheduling deliver clear advantages. It leads by 30–40% on small/quantized models at high concurrency, with gains narrowing to 3–10% on large MoE models where GPU compute dominates. The Dell’s asymmetric PCIe Gen5/Gen4 GPU layout and Threadripper CPU scheduling limit multi-GPU inference efficiency.
Conclusion (Z8 Fury G6i)
The HP Z8 Fury G6i re-establishes Intel’s competitiveness in high-end AI workstations. It outperforms lower-tier Xeon systems across most workloads and trades performance evenly with AMD’s flagship Threadripper workstation in GPU rendering and media tasks. Its defining strength is multi-GPU AI inference, outperforming the AMD platform heavily on common small and quantized LLMs thanks to full Gen5 PCIe bandwidth and optimized CPU-side scheduling.
While the higher-core AMD system remains superior for extreme multi-threaded CPU and memory-bound workloads, the G6i is a highly serviceable, enterprise-grade AI workstation. It serves as an ideal choice for enterprises standardizing on Intel platforms and prioritizing GPU-accelerated AI inference, content creation and professional visualization.
Intel Xeon 658X Processor Review (Compressed)
Intel launched the Granite Rapids-WS Xeon 600 series on February 2, 2026, with retail chips releasing in late March. Unifying legacy Xeon W-2500/W-3500 lines on the new W890 chipset and LGA4710 socket, the 11-SKU lineup tops out at the 86-core Xeon 698X ($7,699). This review covers the 24-core/48-thread Xeon 658X, a mid-entry model retailing for $2,000–$2,300 (above its $1,869 list price), tested on a bare Asus Pro WS W890E-SAGE SE platform for pure CPU performance data.
The Xeon 658X features a 3.0GHz base, 4.9GHz boost clock, 144MB L3 cache, 250W base TDP (300W max turbo). It retains the full platform specs of high-end 600-series chips: eight DDR5-6400 memory channels, 4TB max RAM, 128 PCIe 5.0 lanes, plus universal AMX and AVX-512 acceleration. Unlike high-core variants, it does not support DDR5-8000 MRDIMM. Industry rumors suggest Granite Rapids-WS is a terminal platform with no confirmed mainstream successor until the potential 2028 Coral Rapids refresh, creating minor upgrade uncertainty for custom builders.
Xeon 600 Series Full SKU Overview
The lineup ranges from 12-core to 86-core models. Top five SKUs (674X and above) support DDR5-8000 MRDIMM; lower-tier 634/636/638 cut to four memory channels and 80 PCIe 5.0 lanes. All models support 4TB RAM, CXL 2.0 and per-core FP16 AMX acceleration, while X-series SKUs are unlocked.
Test Platform & Comparison Setup
The Xeon 658X test rig uses 128GB DDR5-6400 ECC, Samsung 9100 Pro 1TB storage and Noctua NH-U12S DX-4677 cooling. Benchmarks pit it against the 64-core Xeon 696X (HP Z8 Fury G6i) and 96-core AMD Threadripper 9995WX (Dell Precision 7875), focusing on CPU-native performance and mitigating GPU variable interference.
Benchmark Performance Results
Procyon AI Computer Vision
The 658X leads the test group with a 248 overall score, 20% faster than the 64-core Xeon 696X and 58% faster than AMD’s 9995WX, leveraging native AMX acceleration for superior CPU-based AI inference on mainstream vision models.
Blender 4.5/5.0 CPU Rendering
Limited by lower core count, the 658X lags both rivals in CPU rendering. In Blender 4.5, it scores 365.3/234.1/186.2 samples/min across three scenes, with the HP system 87–92% faster and the Dell system 184–218% faster. Blender 5.0 results show identical performance gaps versus the Xeon 696X platform.
PCMark 10
Scoring 9657, the 658X delivers stronger general system performance than the Z8 G6i (7742, +25%) but falls 18% short of the high-core AMD Dell workstation.
Blackmagic RAW Speed Test
The 658X hits 205FPS 8K CPU decoding and 181FPS 8K GPU decoding. Its CPU result is 34% lower than the HP system but 30% higher than AMD; GPU performance trails both competitors significantly.
3DMark CPU Profile & Geekbench 6
The 658X outperforms the higher-core Xeon 696X across all thread counts in 3DMark CPU tests, with a 16890 max-thread score (+7% vs HP). In Geekbench 6, it achieves near parity with the 696X (2383 single-core / 21447 multi-core) thanks to shared Redwood Cove core architecture, while falling 33–36% behind the AMD flagship.
Y-Cruncher Standard & BBP Tests
For small-scale Pi calculations, the 658X beats the AMD system but lags the 64-core Xeon. Large-scale workloads favor higher-core competitors. The compute-bound BBP test shows linear core-count scaling, with the 24-core 658X delivering roughly half the throughput of the 64-core 696X.
7-Zip Compression
The 658X scores 250.08 GIPS total rating, 38% lower than the HP Z8 G6i but vastly outperforming the AMD system in compression/decompression efficiency.
Conclusion (Xeon 658X)
The Xeon 658X excels as a platform-focused workstation CPU rather than a raw multi-threaded performer. It suffers predictable losses in core-scaling workloads like Blender rendering and large-scale mathematical computation against higher-core Xeon and AMD rivals. However, it boasts group-leading CPU AI inference via AMX acceleration, superior lightly-threaded and daily system performance, and single-core parity with top-tier Xeon 600-series chips.
Retaining full 128 PCIe 5.0 lanes, eight DDR5 memory channels and 4TB RAM capacity of flagship models, it delivers enterprise-grade I/O and expandability at a lower cost. While platform upgrade uncertainty exists for custom builds, it is a cost-effective choice for OEM workstation deployments and users prioritizing full platform features, AI inference and general productivity over extreme multi-threaded rendering power.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!