Intel Data Center GPU Flex 170, ARK specifications page (first-party)

VERBATIM 2026-10-04 via web_extract; rows quoted exactly as rendered. This is Intel's OWN ARK page for the Flex 170 and it is the document that should have been cited by accelerators/xpu-intel-dc-gpu-flex-170, which instead cites only two GitHub READMEs. ESSENTIALS: Product Collection 'Intel® Data Center GPU Flex Series' | Microarchitecture 'Xe-HPG' | Code Name 'Products formerly Arctic Sound' | Marketing Status Launched | Launch Date "Q3'22" | Warranty Period 3 yrs | Use Conditions Server/Enterprise | Usecase Cloud Computing. GPU SPECIFICATIONS: 'Xe-cores 32' | 'Render Slices 8' | 'Ray Tracing Units 32' | 'Intel® Xe Matrix Extensions (Intel® XMX) Engines 512' | 'Execution Units 512' | 'Graphics Max Dynamic Clock 2050 MHz' | 'Intel® Xe Matrix Extensions (Intel® XMX) Max Dynamic Clock 1950 MHz' | 'TBP 150 W' | 'PCI Express Configurations ‡ Up to PCI Express 4.0 x16' | 'Device ID 0x56C0'. MEMORY SPECIFICATIONS: 'Memory Size 16 GB' | 'Memory Type GDDR6' | 'Graphics Memory Interface 256 bit' | 'Graphics Memory Bandwidth 576 GB/s'. I/O SPECIFICATIONS: '# of Displays Supported‡ 0'. NO COMPUTE FIGURES AT ALL ON THIS PAGE: ARK publishes no TFLOPS and no TOPS row for the Flex 170. The compute numbers exist only in the Flex Series product brief (see w4h-intel-dc-gpu-flex-series-product-brief). SPARSITY CONVENTION: there is NO sparsity footnote anywhere on this page and no row titled 'with Structured Sparsity' or similar. Every figure above is therefore DENSE as published and must NOT be halved. Intel's Arc/Instinct data-centre convention (label sparsity only on rows explicitly so titled) applies here in its strongest form: there is no such row, so there is nothing to halve.