NVIDIA RTX PRO 6000 vs RTX PRO 4500 Blackwell datasheets - Compute/Graphics API rows compared (silent-documentation-revision evidence)

SILENT-REVISION EVIDENCE, both PDFs re-downloaded and text-extracted 2026-10-04. This record exists to hold the side-by-side API-row reading, because the finding is a COMPARISON and neither sheet states it. EXACT ROWS AS PRINTED, RTX PRO 6000 Blackwell Workstation Edition (doc 3519208, footer '3519208. Mar25', HTTP Last-Modified Tue, 18 Mar 2025): 'Graphics APIs DirectX 12, Shader Model 6.6, OpenGL 4.63, Vulkan 1.33' / 'Compute APIs CUDA 11.6, OpenCL 3.0, DirectCompute'. RTX PRO 4500 Blackwell Workstation Edition (doc 5108623, footer '5108623. Apr26', HTTP Last-Modified Tue, 07 Jul 2026): 'Graphics APIs DirectX 12, Shader Model 6.6, OpenGL 4.63, Vulkan 1.43' / 'Compute APIs CUDA 12.8, OpenCL 3.0, DirectCompute'. TWO CORRECTIONS TO THE EXISTING REPO TEXT. (1) The note on vb-nv-rtx-pro-4500-blackwell-datasheet-spec-table previously rounded these to 'OpenGL 4.6, Vulkan 1.4' and 'Vulkan 1.3'; the sheets actually print 4.63, 1.43 and 1.33 - the trailing digit matters here because 1.33 vs 1.43 is the entire delta. That record has been corrected. (2) The claim that the 6000's row is 'stale' is right but under-argued; this record supplies the missing evidence. THE 6000'S API ROW IS INHERITED FROM THE ADA TEMPLATE, NOT A BLACKWELL PROPERTY: the RTX 6000 ADA Generation datasheet (doc 2504660) prints the identical 'Compute APIs CUDA 11.6, OpenCL 3.0, DirectCompute' with 'Graphics APIs ... OpenGL 4.66, Vulkan 1.36'. Two different Ada-vs-Blackwell sheets both carrying 'CUDA 11.6' shows the string is copy-paste from a pre-Blackwell template that NVIDIA did not update, which is a DOCUMENTATION defect. It is not a hardware difference: the RTX PRO 4500 is the same architecture (both print 'GPU architecture NVIDIA Blackwell', 5th Gen Tensor Cores, 4th Gen RT Cores) and the 4500 sheet says CUDA 12.8. NVIDIA ships one driver stack for both parts. DO NOT propagate 'CUDA 11.6' or 'Vulkan 1.33' into any accelerator record as a capability ceiling on the RTX PRO 6000. CONVERSE HAZARD, SYMMETRIC AND WORSE: a silent revision can also go the other way, so the 4500 sheet is not automatically right about the 6000's future contents - it is right about TODAY's 4500 sheet only. Anyone citing a driver/API support level from a workstation datasheet should cite the NVIDIA driver release notes, not the datasheet row. THE FIGURES THAT MATURE ON BOTH SHEETS ARE UNCHANGED, which is the control: CUDA Cores 24,064 vs 10,496, AI TOPS 4000 vs 1,617, 96 GB/1792 GB/s vs 32 GB/896 GB/s, 600 W vs 200 W all re-verified byte-for-byte. The revision is confined to the two API rows.