RTX PRO 4000 Blackwell vs RTX 4000 Ada: what the upgrade buys, including both SFF cards

Two generations, the same formats

The RTX 4000 Ada and its small-form-factor sibling have been NVIDIA’s midrange workstation cards since 2023. Their Blackwell successors keep the formats, a full-height single-slot card and a low-profile 70 W card, and change almost everything inside. The full-size RTX PRO 4000 Blackwell has 24 GB of GDDR7 at 672 GB/s of memory bandwidth, against 20 GB of GDDR6 at 360 GB/s on the RTX 4000 Ada; between the SFF cards the bandwidth rises from 280 to 432 GB/s. In Puget Systems’ tests of December 2025 the new card was 34 to 83 per cent faster in V-Ray, Blender, Octane, Unreal Engine, After Effects and Topaz Video AI, while for CAD the midrange Blackwell cards gained 12 to 21 per cent in SOLIDWORKS and Revit depends mainly on the processor.

NVIDIA datasheets: RTX 4000 Ada (2023) and RTX PRO 4000 Blackwell (June 2026). The two SFF cards are compared further down.

What changed from Ada to Blackwell

Memory moved from GDDR6 to GDDR7, which is where most of the bandwidth gain comes from, and grew from 20 to 24 GB. The Tensor Cores are a generation newer and add FP4; the RT cores are a generation newer too. The host link doubles its bandwidth per lane with PCIe 5.0, the display outputs move to DisplayPort 2.1b, and the video engines add 4:2:2 encoding and decoding for H.264 and HEVC, which the Ada cards lack. The full-size card now draws 145 W instead of 130 W through the same single 16-pin connector; an adapter from one 8-pin cable comes with the card. Like the Ada card, it needs an auxiliary power cable; the SFF card still needs none.

NVIDIA publishes two sets of figures for the full-size card, so check which one a specification sheet uses. Its product page gives 1,290 AI TOPS, 40 TFLOPS of FP32 and 122 RT TFLOPS; the newer datasheet of June 2026 gives 1,178, 37 and 112, and this article uses the datasheet. Both say 145 W, while some partner pages still show the launch figure of 140 W.

Measured: rendering and content creation

Puget Systems tested the full-size cards against each other in its December 2025 roundup; the SFF cards were not part of it.

Puget Systems, 2025 Professional GPU Content Creation Roundup, 18 December 2025.

Local AI: bandwidth and 4 GB more memory

For language models the bandwidth line matters most, because token generation reads the weights once per token: 672 against 360 GB/s puts the ceiling 87 per cent higher on the same model. Compute rises less than bandwidth. NVIDIA quotes the Blackwell card’s AI TOPS in FP4 and the Ada card’s in FP8, both with sparsity. Halved to FP8 at the 2:1 ratio NVIDIA’s Server Edition specifications show, the full-size card goes from 427.6 to about 589 by our arithmetic, 38 per cent more. The 427.6 is from the Ada card’s product page; its 2023 datasheet prints 327.6, which does not match the clock implied by the card’s FP32 rate.

The extra 4 GB changes which models fit. By our users-per-card rule (90 per cent of the card’s memory, less 3 GiB, for weights and KV cache), a 24 GB card has about 18.6 GiB and a 20 GB card about 15 GiB. A 14B model in FP8, such as Qwen3-14B at 15.2 GiB, fits the 24 GB card with an FP8 cache for about five conversations of 8,192 tokens and does not fit the 20 GB card. gpt-oss-20b, a 13.8 GB checkpoint, fits both. A 32B model in 4-bit is about 18 to 21 GB of weights, depending on the format; only the smallest checkpoints fit the 24 GB card, with cache for about one such conversation, and none fits the 20 GB card. For a 32B model with several users, the 32 GB RTX PRO 4500 is the next step. Only the Blackwell card runs NVFP4 natively, the 4-bit format NVIDIA’s TensorRT-LLM supports on Blackwell and not on Ada; the Ada card runs 4-bit weights as INT4 through AWQ or GPTQ instead. Our VRAM guide has the full arithmetic.

We supply the RTX PRO 4000 and, for models that need more memory, the 32 GB RTX PRO 4500. Tell us the model and the context length, and we will say which one fits.

The SFF cards: 70 W, low profile

The two SFF cards share the format that makes them useful: 2.7 by 6.6 inches, low profile, dual slot, powered from the slot, actively cooled, and supplied with both a full-height and a low-profile bracket. They suit compact workstations and small servers that have no GPU power cable, provided the maker lists a 70 W low-profile card for that chassis. The 16 GB RTX PRO 2000 in the same format is set against the RTX 4000 Ada in our RTX PRO 2000 vs RTX 2000 Ada and RTX 4000 Ada comparison, and every other tier in Blackwell vs Ada, tier by tier.

NVIDIA datasheets: RTX 4000 SFF Ada (April 2023) and RTX PRO 4000 SFF Blackwell (June 2026). NVIDIA’s launch documents of March 2023 gave the Ada card 320 GB/s.

Inside, the Blackwell card has the same 8,960 cores as its full-size sibling, but at 70 W it runs them at a much lower clock and with slower memory. It reaches 60 to 65 per cent of the full-size card’s FP32 rate, depending on which of NVIDIA’s figures is used, and 64 per cent of its bandwidth at 48 per cent of the power. Against the RTX 4000 SFF Ada it brings 24 instead of 20 GB and 432 instead of 280 GB/s, or 320 GB/s by NVIDIA’s launch documents of March 2023.

NVIDIA’s launch blog claims “up to 2.5x higher AI performance, 1.7x higher ray-tracing performance and 1.5x more bandwidth” than the previous generation. The last two match NVIDIA’s published figures: 73 against 44.3 RT TFLOPS is 1.65 times, and 432 against 280 GB/s is 1.54 times. The first matches an FP4 figure, 770 TOPS, divided by an FP8 figure, 306.8 TFLOPS; NVIDIA’s blog gives no test conditions. At the same precision the gain is about 25 per cent by our arithmetic. We have not found an independent benchmark of the two SFF cards against each other.

One detail matters for upgrades in older machines. The Blackwell SFF card connects at PCIe 5.0 x8 through a full-length connector, and the Ada card at PCIe 4.0 x16. In a PCIe 4.0 workstation the new card runs at 4.0 x8, half the link bandwidth of the card it replaces. For inference with the model resident in memory that rarely matters; for workloads that stream data to the GPU all the time, it can.

We supply the RTX PRO 4000 SFF too. Write to us with the workstation model, and we tell you in advance whether the card will work in it.

CAD and BIM: the smallest step

Puget’s engineering roundup of December 2025 is the reference for design departments. In SOLIDWORKS, the midrange Blackwell cards, the 2000, 4000 and 4500, gained 12 to 21 per cent over their Ada predecessors; in SOLIDWORKS drawings every AMD card tested beat the fastest NVIDIA model by at least 18 per cent; Autodesk Inventor ran well on every card tested; and Revit, in Puget’s words, “is primarily CPU-dependent”. A CAD seat that is fast enough today will not feel much faster with the new card. The reasons to upgrade a CAD workstation are the display outputs for high-resolution monitors, the extra memory for very large assemblies and the ISV certifications of the current generation.

When to upgrade from the RTX 4000 Ada

The upgrade is worth it for GPU rendering, video and visualisation, where Puget measured 34 to 83 per cent in V-Ray, Blender, Octane, Unreal Engine, After Effects and Topaz Video AI and about 25 per cent overall in DaVinci Resolve, and for local AI, where the bandwidth nearly doubles and the extra 4 GB brings 14B models in FP8 into range.

For CAD and BIM seats, the gain is smaller: 12 to 21 per cent in SOLIDWORKS and close to nothing in Revit.

For small-form-factor machines, the Blackwell SFF card is ahead in every published performance figure, but check the host’s PCIe generation first and expect a smaller step than the full-size card makes.

Neither generation of these cards supports MIG or vGPU. For MIG, the Blackwell workstation cards with documented driver and vBIOS requirements start at the RTX PRO 5000; NVIDIA’s MIG guide (updated 4 November 2026) also lists an RTX PRO 4500 with two instances of 16 GB and a minimum driver of 575.64, but names no edition and no vBIOS for it, and the workstation 4500’s datasheet and product page do not mention MIG, so have MIG support on that card confirmed in writing. Virtual machines with vGPU need a card on NVIDIA’s vGPU list, such as the RTX PRO 6000 Server Edition or the L4; our RTX PRO family comparison explains which RTX PRO cards qualify.

What we supply

We supply the RTX PRO 4000 Blackwell and the RTX PRO 4000 SFF with manufacturer warranty on one EU contract and invoice, as single cards or in workstations built to order, together with the rest of the professional GPU range. For rack servers with vGPU, we supply the RTX PRO 6000 and RTX PRO 4500 Server Editions, the L40S and the L4.

FAQ