NVIDIA RTX PRO 2000 Blackwell 16GB Specs, Benchmarks & Pricing

The NVIDIA RTX PRO 2000 Blackwell is a compact entry-level professional workstation GPU announced at SIGGRAPH 2025 and released on August 11, 2025. It is built on the NVIDIA Blackwell architecture (GB206 die, TSMC 5nm process) with 4,352 CUDA cores, 136 fifth-generation Tensor Cores, and 34 fourth-generation RT Cores. The card delivers 17.0 TFLOPS of single-precision FP32 compute and 545 AI TOPS (FP4 sparse), making it a notable upgrade over its predecessor the RTX 2000 Ada Generation in both graphics and AI performance.

It features 16 GB of GDDR7 ECC memory on a 128-bit memory bus providing 288 GB/s bandwidth, a significant step up from the Ada predecessor's 224 GB/s GDDR6. The fifth-generation Tensor Cores add FP4 precision support alongside FP8, BF16, TF32, FP16, and INT8, enabling broader AI and machine learning workloads. The fourth-generation RT Cores double the intersection rate of the prior generation, accelerating ray tracing for professional visualization.

With a TDP of only 70W and no external power connector required, the RTX PRO 2000 Blackwell maintains the same power envelope as its Ada predecessor while delivering substantially higher performance. The dual-slot, half-height form factor (2.7" H x 6.6" L) allows installation in both standard and small-form-factor workstations. It connects via PCIe 5.0 x8 and provides four mini-DisplayPort 2.1b outputs supporting up to four independent displays at 4K 165 Hz or two displays at 8K 100 Hz. Video engine support includes one ninth-generation NVENC encoder and one sixth-generation NVDEC decoder.

The RTX PRO 2000 Blackwell succeeds the RTX 2000 Ada Generation and is distributed through PNY, TD SYNNEX, and embedded in compact workstations from BOXX, Dell, HP, and Lenovo.

  • Release Date: August 11, 2025
  • MSRP: $849 USD†
  • GPU Architecture: blackwell
  • Hardware-Accelerated GEMM Operations:
    FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
  • CUDA Compute Capability : 12

† Estimated MSRP of $849 USD based on launch-time market pricing. NVIDIA does not publish official MSRP for workstation GPUs. Newegg lists the PNY variant (VCNRTXPRO2000B-PB) at $849.99 and Wiredzone lists the NVIDIA OEM version (900-5G195-2250-000) at $848.00. CG Channel cited $722 from reseller Connection at announcement time. The $849 price represents the consistent retail list price across major distributors at/near launch.

Considerations

  • Lower memory bandwidth compared to other GPUs
  • Lower INT8 inference performance compared to other GPUs

Specifications for NVIDIA RTX PRO 2000 Blackwell

SpecificationPerformance Ranking
FP32 TFLOPs
49th @ 34 TFLOPs (Entry Tier)(Entry)
FP16 TFLOPs
47th @ 136.25 TFLOPs (Entry Tier)(Entry)
Tensor Core Count
30th @ 136 Cores (Entry Tier)(Entry)
Memory Capacity (GB)
60th @ 16 GB (Mid Tier)(Mid)
Memory Bandwidth (GB/s)
20th @ 288 GB/s
Int8 TOPs
23rd @ 136.25 TOPs
FP8 TFLOPsData not available
FP4 TFLOPs
10th @ 272.5 TFLOPs

Real-time NVIDIA RTX PRO 2000 Blackwell GPU Prices

We're tracking 14 of the NVIDIA RTX PRO 2000 Blackwell GPUs currently available for sale. The lowest average price is $1,426 (what's this?)
Buy Now

Compare Price/Performance to other GPUs

We track real-time prices of other GPUs too so that you can compare the price/performance of the NVIDIA RTX PRO 2000 Blackwell GPU to other GPUs.
Compare GPU Price/Performance

Compare NVIDIA RTX PRO 2000 Blackwell to Another GPU

Compare the NVIDIA RTX PRO 2000 Blackwell directly to another GPU to see specs, benchmarks, and prices side-by-side.
Compare GPUs Side-by-Side

Price History

NVIDIA RTX PRO 2000 Blackwell Price History

This chart tracks NVIDIA RTX PRO 2000 Blackwell prices over the past 6 months based on eBay listings. The yellow line shows the average of the three lowest-priced listings each day—a useful indicator for finding deals.

Trend: Over this period, the lowest average price has been up 53.9%. The current lowest average is $1,425, compared to a typical $1,160 over the period. The lowest average price is calculated as the average of the three lowest-priced listings each day.

gpupoet.com

Product Identifiers

Manufacturer Part Numbers (1)
NVIDIA Part Number
900-5G195-2250-000
Available from 3 Partners (3 products)
PNY
NVIDIA RTX PRO 2000 Blackwell 16GB GDDR7 Graphics Card
VCNRTXPRO2000B-PB(sku)
HP
HP NVIDIA RTX PRO 2000 Blackwell 16GB 4MDP Graphics
B5CH7AA(part number)
Lenovo
NVIDIA RTX PRO 2000 Blackwell 16GB mini-DP x4 GDDR7 Graphics Card
4X61T95638(part number)

References

Notes

  1. FP32 vector (CUDA core) throughput of 17.0 TFLOPS sourced directly from the NVIDIA RTX PRO 2000 Blackwell official datasheet (nvidia.com), representing single-precision CUDA core FP32 performance. Consistent with 4,352 CUDA cores x 1,957 MHz boost clock x 2 ops/cycle / 1000 = 17.03 TFLOPS.
  2. AI TOPS of 545 is the 'AI TOPS' figure from the NVIDIA official datasheet, representing theoretical FP4 TOPS with 2:4 sparsity (confirmed by the RTX PRO 4000 Blackwell datasheet footnote: 'Theoretical FP4 TOPS using the sparsity feature').
  3. int8TOPS of 136.25 represents non-sparse (dense) INT8 performance, derived from FP4 sparse TOPS: 545 (FP4 sparse) / 4 = 136.25 (INT8 dense). Derivation: FP4 sparse = 2x INT8 sparse = 4x INT8 dense per Blackwell 5th-gen tensor core architecture. Dense value used for consistent cross-vendor comparison.
  4. fp16TFLOPS of 136.25 represents Peak FP16 Tensor TFLOPS with sparsity, derived from FP4 sparse TOPS using the Blackwell tensor core ratio (FP16 sparse = FP4 sparse / 4): 545 / 4 = 136.25 TFLOPS. This ratio is consistent with the RTX 5070 where FP16 sparse = 246.9 TFLOPS (from nvidia-geforce-rtx-5070.yaml, sourced from Blackwell GPU architecture whitepaper Table 6). The NVIDIA datasheet does not publish FP16 separately for this workstation card.
  5. Memory bandwidth of 288 GB/s per NVIDIA official datasheet and confirmed by multiple sources. The memory is 16 GB GDDR7 ECC on a 128-bit bus.
  6. GPU die is GB206 on TSMC 5nm process with approximately 21.9 billion transistors. Architecture uses compute capability 12.0 (sm_120), confirmed by the official NVIDIA CUDA GPUs table at developer.nvidia.com/cuda-gpus.
  7. Tensor core count of 136 confirmed by Newegg product listing and multiple third-party sources. The GB206 has 34 SMs x 128 CUDA cores per SM = 4,352 CUDA cores; 34 SMs x 4 tensor core units per SM = 136 tensor cores (same 4-per-SM ratio as Ada Lovelace).
  8. Released at SIGGRAPH 2025 on August 11, 2025. Previously announced via NVIDIA blog in early 2025 as 'coming later this year.' Source: TechPowerUp launch date, SIGGRAPH 2025 announcement via multiple tech news outlets.
  9. FP4 hardware precision is supported by 5th-generation Tensor Cores. FP8 is also supported. Software-level support for FP8 in NVIDIA Transformer Engine requires compute capability 9.x (Hopper) and does not support Blackwell desktop GPUs (sm_120) via that specific library path; lower-level libraries (cuBLASLt, TensorRT) support FP8 on Blackwell.
  10. Estimated MSRP of $849 USD based on launch-time market pricing. NVIDIA does not publish official MSRP for workstation GPUs. Newegg lists the PNY variant (VCNRTXPRO2000B-PB) at $849.99 and Wiredzone lists the NVIDIA OEM version (900-5G195-2250-000) at $848.00. CG Channel cited $722 from reseller Connection at announcement time. The $849 price represents the consistent retail list price across major distributors at/near launch.
  11. NVPN 900-5G195-2250-000 sourced from B&H Photo, Newegg, and Wiredzone product listings for the NVIDIA OEM version. No official board ID documented for this GPU; other NVIDIA workstation GPUs use a PG-prefixed board ID (e.g., L40 uses PG133, RTX A5000 uses PG132) derived separately from the NVPN.
  12. PNY SKU VCNRTXPRO2000B-PB confirmed from Newegg and ProVantage product listings. HP part number B5CH7AA confirmed from ProVantage and Hashrate.no product listings. Lenovo part number 4X61T95638 confirmed from lenovo.com product page listing and eBay listing.
  13. No external power connector required; 70W TDP supplied entirely via PCIe slot. Same power envelope as predecessor RTX 2000 Ada Generation (also 70W).
  14. fp4TFLOPS of 272.5 represents non-sparse (dense) FP4 performance, derived from the official NVIDIA AI TOPS figure of 545 (FP4 sparse, per the NVIDIA official datasheet) ÷ 2. Dense value used for consistent cross-vendor comparison.
  15. NOT FOUND: fp8TFLOPS — the official NVIDIA RTX PRO 2000 Blackwell datasheet and product page (URLs in references) publish only AI TOPS (FP4 sparse), Single-Precision Performance (FP32), and RT Core Performance — no standalone FP8 TFLOPS row, unlike the RTX PRO 4500/6000 Server Editions which do publish one. Left null rather than guessed from the INT8 dense value.
  16. supportedHardwareOperations: FP32 and INT4 removed per the locked GEMM-only definition (2026-09) — the 17.0 TFLOPS FP32 vector figure is plain CUDA-core (non-tensor) throughput, and neither consumer nor datacenter Blackwell documentation lists INT4 Tensor Core support (dropped starting with Blackwell per gpu-researcher.md's canonical Blackwell table). RTX PRO 2000 is a workstation (not B100/B200-class datacenter) SKU, so no FP64 is added either. The array now matches the canonical consumer/professional Blackwell baseline of FP4, FP8, FP16, BF16, TF32, INT8.
  17. fp32TFLOPS of 34 is TF32 Tensor Core throughput without sparsity. NVIDIA does not publish a dense TF32 figure for this card. Derived as the 545 AI TOPS (FP4 with sparsity) on NVIDIA's RTX PRO 2000 page / 16 = 34.06, since professional Blackwell Tensor Cores run FP4 with sparsity at 16 × dense TF32. FP32 vector (CUDA core) throughput is 17.0 TFLOPS; professional parts other than the L40 run dense TF32 at 2 × FP32 vector, as Table 4 of NVIDIA's RTX PRO Blackwell architecture whitepaper shows for the RTX A6000 (77.4 vs 38.7), RTX 6000 Ada (182.1 vs 91.1) and RTX PRO 6000 Workstation (251.9 vs 126.0) (https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/quadro-product-literature/pdf/NVIDIA-RTX-Blackwell-PRO-GPU-Architecture-v1_1.pdf).
This site contains affiliate links. As an eBay Partner and Amazon Associate, gpupoet.com may be compensated if you make a purchase, at no cost to you. Learn more.