- Joined
- Aug 19, 2017
- Messages
- 2,999 (1.06/day)
Today, we are looking at the latest benchmark results for NVIDIA's upcoming RTX PRO 6000 "Blackwell" workstation-class GPU. Based on the new GB202 GPU, this professional visualization card features an impressive 24,064 CUDA cores distributed across 188 streaming multiprocessors, with boost clocks up to 2,617 MHz. It also introduces 96 GB of GDDR7 memory with full error‑correcting code, a capacity made possible by dual‑sided 3 GB modules. In Geekbench 6.4.0 OpenCL trials, the PRO 6000 Blackwell registered a total score of 368,219. That result trails the gaming‑oriented GeForce RTX 5090, which posted 376,858 points despite having fewer cores (21,760 vs. 24,064 of RTX PRO) and a lower peak clock of 2,410 MHz versus the 2617 MHz of RTX PRO.
A breakdown of subtests reveals that the workstation card falls behind in background blur (263.9 versus 310.7 images per second) and face detection (196.7 versus 241.5 images per second), yet it leads modestly in horizon detection and Gaussian blur. These mixed outcomes are attributed to pre‑release drivers, a temporary cap on visible memory (currently limited to 23.8 GB), and power‑limit settings. If the card ran on release drivers, software (especially OpenCL) could greatly benefit from more cores and higher max frequency. One significant distinction within the RTX PRO 6000 family concerns power consumption. The Max‑Q Workstation Edition is engineered for a 300 W thermal design point, making it suitable for compact chassis and environments where quiet operation is essential. It retains all 24,064 cores and the full 96 GB of memory, but clocks and voltages are adjusted to fit the 300 W budget. By contrast, the standard Workstation and Server models allow a thermal budget of up to 600 W, enabling higher sustained frequencies and heavier compute workloads in full‑size desktop towers and rack‑mounted systems.
Additional PCB leaks on the Chiphell forum and pre‑release listings from Leadtek confirm that both Max‑Q and full‑power variants share the same modular board designs, replacing the conventional 12 V‑6×2 power connector with solder pads. NVIDIA plans to extend this approach to its RTX PRO 5000 line, offering 48 GB configurations, as well as to future Super variants with increased memory. Although official launch dates remain under embargo, these leaks suggest that the complete Blackwell‑based professional visualization portfolio will be released later this quarter, providing customers with flexible power and performance options.
View at TechPowerUp Main Site | Source
A breakdown of subtests reveals that the workstation card falls behind in background blur (263.9 versus 310.7 images per second) and face detection (196.7 versus 241.5 images per second), yet it leads modestly in horizon detection and Gaussian blur. These mixed outcomes are attributed to pre‑release drivers, a temporary cap on visible memory (currently limited to 23.8 GB), and power‑limit settings. If the card ran on release drivers, software (especially OpenCL) could greatly benefit from more cores and higher max frequency. One significant distinction within the RTX PRO 6000 family concerns power consumption. The Max‑Q Workstation Edition is engineered for a 300 W thermal design point, making it suitable for compact chassis and environments where quiet operation is essential. It retains all 24,064 cores and the full 96 GB of memory, but clocks and voltages are adjusted to fit the 300 W budget. By contrast, the standard Workstation and Server models allow a thermal budget of up to 600 W, enabling higher sustained frequencies and heavier compute workloads in full‑size desktop towers and rack‑mounted systems.
Additional PCB leaks on the Chiphell forum and pre‑release listings from Leadtek confirm that both Max‑Q and full‑power variants share the same modular board designs, replacing the conventional 12 V‑6×2 power connector with solder pads. NVIDIA plans to extend this approach to its RTX PRO 5000 line, offering 48 GB configurations, as well as to future Super variants with increased memory. Although official launch dates remain under embargo, these leaks suggest that the complete Blackwell‑based professional visualization portfolio will be released later this quarter, providing customers with flexible power and performance options.
View at TechPowerUp Main Site | Source
