Nvidia gives its tiniest workstation GPUs a Blackwell boost
(2025/08/12)
- Reference: 1754954196
- News link: https://www.theregister.co.uk/2025/08/11/nvidia_workstation_cards_blackwell_updates/
- Source link:
Nvidia’s latest Blackwell GPUs are a pair of itty-bitty workstation cards that aim to deliver the highest performance possible for professional visualization and local AI workloads within a 70-watt energy diet.
Unveiled at the Siggraph conference in Vancouver, British Columbia, on Monday, the RTX Pro 4000 small form factor (SFF) and RTX Pro 2000 share the same half-high dual-slot cooler design. But despite their similarities, the two cards are very different beasts, with the RTX 4000 SFF packing more than twice the CUDA cores at 8,960 versus the RTX Pro 2000's 4,352.
Nvidia [1]says the RTX 4000 Ada SFF is about 1.7x faster in ray tracing and offers 2.5x higher AI performance compared to its previous model. With 280 tensor cores, Nvidia says the chip is capable of delivering 770 teraFLOPS of FP4 performance.
[2]
While that technically is a 2.51x improvement in floating point math, most of that comes from the move to FP4 rather than architectural gains. When normalized to FP8 the chip is only about 25 percent faster.
[3]
[4]
While floating point performance may not make a huge leap with the diminutive new card, memory bandwidth – a key metric when it comes to local inference – certainly does. With 24GB of GDDR7 memory good for 432GB/s of bandwidth the card should be able to churn out tokens in LLMs like OpenAI's newly released gpt-oss-20b roughly 54 percent faster than Nvidia’s last offering.
Nvidia suggests the RTX Pro 2000 is best used for professional visualization workloads, however it also promises a sizable performance uplift. The GPU giant says you can expect a 1.6x improvement in 3D modeling, 1.4x higher performance in CAD, and 1.6x faster rendering compared to its Ada Generation predecessor.
[5]
As for AI performance, the card falls well short of its larger or more power hungry siblings but is no slouch considering its meager 70W TDP. The card boasts up to 545 teraFLOPS of sparse FP4 compute and 280GB/s of memory bandwidth fed by 16GB of GDDR7.
Both the RTX Pro 4000 SFF and 2000 will be available starting later this year from PNY and TD SYNNEX and will be offered in OEM systems from BOXX, Dell, HP, and Lenovo.
[6]Trump seeing green as he weighs deal to allow Nvidia Blackwell GPU sales to China
[7]Stacking up Huawei's rack-scale boogeyman against Nvidia's best
[8]Intel bets you'll stack cheap GPUs to avoid spending top dollar on Nvidia Pros
[9]AMD puts Intel in rear view mirror with Threadripper Pro 9000 high-end desktop chips
The cards fill out Nvidia's existing line up of Blackwell workstation GPUs [10]announced back at its March GTC conference in March, when the company revealed its 96GB RTX Pro 6000.
Also at Siggraph, Nvidia [11]showed off a 2U server platform that can run a pair of the 600W RTX Pro 6000 Server edition cards. Each of the GPUs boasts up to 4 petaFLOPS of sparse FP4 performance.
The systems are available now from Cisco, Dell, HPE, Lenovo, and Supermicro, among others. ®
Get our [12]Tech Resources
[1] https://blogs.nvidia.com/blog/blackwell-ai-acceleration-workstation-rtx-pro/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/08/11/trump_seeing_green_as_he/
[7] https://www.theregister.com/2025/07/29/huawei_rackscale_boogeyman/
[8] https://www.theregister.com/2025/05/20/intel_arc_pro_b60/
[9] https://www.theregister.com/2025/05/21/amd_threadripper_radeon_workstation/
[10] https://www.theregister.com/2025/03/18/gtc_frame_nvidias_budget_blackwell/
[11] https://nvidianews.nvidia.com/news/nvidia-rtx-pro-servers-with-blackwell-coming-to-worlds-most-popular-enterprise-systems
[12] https://whitepapers.theregister.com/
Unveiled at the Siggraph conference in Vancouver, British Columbia, on Monday, the RTX Pro 4000 small form factor (SFF) and RTX Pro 2000 share the same half-high dual-slot cooler design. But despite their similarities, the two cards are very different beasts, with the RTX 4000 SFF packing more than twice the CUDA cores at 8,960 versus the RTX Pro 2000's 4,352.
Nvidia [1]says the RTX 4000 Ada SFF is about 1.7x faster in ray tracing and offers 2.5x higher AI performance compared to its previous model. With 280 tensor cores, Nvidia says the chip is capable of delivering 770 teraFLOPS of FP4 performance.
[2]
While that technically is a 2.51x improvement in floating point math, most of that comes from the move to FP4 rather than architectural gains. When normalized to FP8 the chip is only about 25 percent faster.
[3]
[4]
While floating point performance may not make a huge leap with the diminutive new card, memory bandwidth – a key metric when it comes to local inference – certainly does. With 24GB of GDDR7 memory good for 432GB/s of bandwidth the card should be able to churn out tokens in LLMs like OpenAI's newly released gpt-oss-20b roughly 54 percent faster than Nvidia’s last offering.
Nvidia suggests the RTX Pro 2000 is best used for professional visualization workloads, however it also promises a sizable performance uplift. The GPU giant says you can expect a 1.6x improvement in 3D modeling, 1.4x higher performance in CAD, and 1.6x faster rendering compared to its Ada Generation predecessor.
[5]
As for AI performance, the card falls well short of its larger or more power hungry siblings but is no slouch considering its meager 70W TDP. The card boasts up to 545 teraFLOPS of sparse FP4 compute and 280GB/s of memory bandwidth fed by 16GB of GDDR7.
Both the RTX Pro 4000 SFF and 2000 will be available starting later this year from PNY and TD SYNNEX and will be offered in OEM systems from BOXX, Dell, HP, and Lenovo.
[6]Trump seeing green as he weighs deal to allow Nvidia Blackwell GPU sales to China
[7]Stacking up Huawei's rack-scale boogeyman against Nvidia's best
[8]Intel bets you'll stack cheap GPUs to avoid spending top dollar on Nvidia Pros
[9]AMD puts Intel in rear view mirror with Threadripper Pro 9000 high-end desktop chips
The cards fill out Nvidia's existing line up of Blackwell workstation GPUs [10]announced back at its March GTC conference in March, when the company revealed its 96GB RTX Pro 6000.
Also at Siggraph, Nvidia [11]showed off a 2U server platform that can run a pair of the 600W RTX Pro 6000 Server edition cards. Each of the GPUs boasts up to 4 petaFLOPS of sparse FP4 performance.
The systems are available now from Cisco, Dell, HPE, Lenovo, and Supermicro, among others. ®
Get our [12]Tech Resources
[1] https://blogs.nvidia.com/blog/blackwell-ai-acceleration-workstation-rtx-pro/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aJq8V97OWsXPNMCfV7LYIwAAARc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/08/11/trump_seeing_green_as_he/
[7] https://www.theregister.com/2025/07/29/huawei_rackscale_boogeyman/
[8] https://www.theregister.com/2025/05/20/intel_arc_pro_b60/
[9] https://www.theregister.com/2025/05/21/amd_threadripper_radeon_workstation/
[10] https://www.theregister.com/2025/03/18/gtc_frame_nvidias_budget_blackwell/
[11] https://nvidianews.nvidia.com/news/nvidia-rtx-pro-servers-with-blackwell-coming-to-worlds-most-popular-enterprise-systems
[12] https://whitepapers.theregister.com/