NVIDIA RTX 4000 Ada Generation 20GB Specs, Benchmarks & Pricing
The NVIDIA RTX 4000 Ada Generation is a professional single-slot workstation GPU released in August 2023, built on the Ada Lovelace architecture using the AD104 chip with TSMC's 5nm process. It features 6,144 CUDA cores, 192 fourth-generation Tensor Cores, and 48 third-generation RT Cores, paired with 20GB of GDDR6 ECC memory. The fourth-generation Tensor Cores introduce FP8 precision support and deliver up to 3X AI performance improvement over the previous Ampere generation, with 213.8 TOPS of dense INT8 performance (427.6 TOPS with sparsity). The card's tensor cores support multiple precision formats including FP8 (in both E4M3 and E5M2 formats), TF32, BF16, FP16, INT8, and INT4, making it highly versatile for AI inference and machine learning workloads. With a TDP of only 130W in a single-slot form factor, the RTX 4000 Ada offers exceptional power efficiency for professional visualization, 3D rendering, and AI applications. The GPU includes two eighth-generation NVENC encoders with AV1 support (40% more efficient than H.264) and two fifth-generation NVDEC decoders. Memory bandwidth of 360 GB/s is provided through a 160-bit memory interface running at 18 Gbps effective speed. With PCI Express 4.0 x16 interface and four DisplayPort 1.4a outputs, it delivers up to 1.7X higher performance than the previous generation RTX A4000.
- Release Date: August 9, 2023
- MSRP: $1,250 USD
- GPU Architecture: ada
- Hardware-Accelerated GEMM Operations:FP16 FP32 BF16 FP8 INT8 INT4 TF32 FP64 INT1
- CUDA Compute Capability : 8.9
Strengths
- 20GB VRAM for model and data capacity
- 26.7 TFLOPS FP32 compute performance
Specifications for NVIDIA RTX 4000 Ada Generation
| Specification | Performance Ranking |
|---|---|
| FP32 TFLOPs | |
| FP16 TFLOPs | |
| Tensor Core Count | |
| Memory Capacity (GB) | |
| Memory Bandwidth (GB/s) | |
| Int8 TOPs |
Gaming Benchmarks for NVIDIA RTX 4000 Ada Generation
Rankings based on 1,262 real-world benchmark results from OpenBenchmarking.org
| Benchmark | Performance Ranking |
|---|---|
| 3DMark Wild Life Extreme (4K) | |
| 3DMark Wild Life Extreme (1440p) | |
| 3DMark Wild Life Extreme (1080p) |
Real-time NVIDIA RTX 4000 Ada Generation GPU Prices
Compare Price/Performance to other GPUs
Compare NVIDIA RTX 4000 Ada Generation to Another GPU
Price History
NVIDIA RTX 4000 Ada Generation Price History
This chart tracks NVIDIA RTX 4000 Ada Generation prices over the past 6 months based on eBay listings. The green line shows the median price across all listings each day, while the yellow line shows the average of the three lowest-priced listings—a useful indicator for finding deals.
Trend: Over this period, the lowest average price has been down 23.0%. The current lowest average is $1,175, compared to a typical $1,482 over the period. The lowest average price is calculated as the average of the three lowest-priced listings each day.
Product Identifiers
Manufacturer Part Numbers (4)
- NVIDIA Part Number
- 900-5G190-2570-000
- NVIDIA Part Number
- 900-5G190-2270-000
- NVIDIA Part Number
- 900-5G190-2770-000
- NVIDIA Part Number
- 900-5G190-0070-000
Available from 6 Partners (9 products)
- Dell NVIDIA RTX 4000 Ada Generation, 20 GB GDDR6, full height, PCIe 4.0x16, 4 DP Graphics Card
- 490-BKHG(part number)
- Dell NVIDIA RTX 4000 SFF Ada Generation, 20 GB GDDR6, half height, PCIe 4.0x16, 4 mDP Graphics Card
- 490-BKPJ(part number)
- NVIDIA RTX 4000 Ada Graphics Accelerator for HPE
- S3T54C(sku)
- HP NVIDIA RTX 4000 Ada Generation Graphics Card
- 8D6B7AA(part number)
- ThinkSystem NVIDIA RTX 4000 Ada 20GB PCIe Active GPU
- 4X67A97287(part number)
- NVIDIA RTX 4000 SFF Ada 20GB mini-DP x4 GDDR6 Graphics Card
- 4X61Q50293(part number)
- PNY NVIDIA RTX 4000 Ada Generation Single-Slot
- VCNRTX4000ADA-PB(sku)
- PNY NVIDIA RTX 4000 SFF Ada Generation Dual-Slot
- VCNRTX4000ADALP-PB(sku)
- Supermicro NVIDIA RTX 4000 SFF Ada Generation 20GB
- GPU-NVQRTX4000-ADA-SFF(part number)
References
- https://www.nvidia.com/en-us/products/workstations/rtx-4000/
- https://www.techpowerup.com/gpu-specs/rtx-4000-ada-generation.c4171
- https://lenovopress.lenovo.com/lp2144-thinksystem-nvidia-rtx-4000-ada-20gb-pcie-active-gpu
- https://www.leadtek.com/eng/products/workstation_graphics(2)/NVIDIA_RTX_4000_Ada_Generation(40988)/detail
- https://images.nvidia.com/aem-dam/en-zz/Solutions/technologies/NVIDIA-ADA-GPU-PROVIZ-Architecture-Whitepaper_1.1.pdf
- https://videocardz.net/nvidia-rtx-4000-ada
- https://www.cgchannel.com/2023/08/nvidia-unveils-rtx-4000-4500-and-5000-workstation-gpus/
- https://develop3d.com/workstations/nvidia-rtx-4000-4500-5000-ada-generation-gpus-launch/
- https://www.bhphotovideo.com/c/product/1882046-REG/nvidia_900_5g190_2270_000_rtx_4000_ada_graphic.html
- https://www.amazon.com/NVIDIA-Workstation-Architecture-Professional-900-5G190-2570-000/dp/B0D4V35VBP
- https://www.dell.com/en-us/shop/dell-nvidia-rtx-4000-ada-generation-20-gb-gddr6-full-height-pcie-40x16-4-dp-graphics-card/apd/490-bkhg/graphic-video-cards
- https://buy.hpe.com/us/en/options/server-accelerators/computational-graphics-accelerators-for-servers/computational-graphics-accelerators-with-hpe/nvidia-rtx-4000-ada-graphics-accelerator-for-hpe/p/s3t54c
- https://www.bhphotovideo.com/c/product/1816528-REG/hp_8d6b7aa_nvidia_rtx_4000_ada.html
- https://www.lenovo.com/us/en/p/accessories-and-software/graphics-cards/graphics_cards/4x61q50293
- https://www.pny.com/nvidia-rtx-4000-ada
- https://www.provantage.com/pny-technologies-vcnrtx4000ada-pb~4PNY92E7.htm
- https://www.senetic.co.uk/product/GPU-NVQRTX4000-ADA-SFF
- https://github.com/NVIDIA/open-gpu-kernel-modules
- https://devicehunt.com/view/type/pci/vendor/10DE/device/27B0
Notes
- FP32 TFLOPS of 26.7 calculated as 6,144 CUDA cores × 2.175 GHz boost clock × 2 ops/cycle ÷ 1000 = 26.73 TFLOPS (rounded to 26.7 as published by NVIDIA).
- Tensor performance of 427.6 TFLOPS represents peak FP8 tensor performance with sparsity, as published by NVIDIA and confirmed across multiple official sources including Leadtek and NVIDIA product pages.
- FP16 TFLOPS of 213.8 is calculated based on the Ada Lovelace tensor core architecture where FP16 with sparsity delivers approximately half the performance of FP8 with sparsity (427.6 TFLOPS ÷ 2 = 213.8 TFLOPS).
- int8TOPS of 213.8 represents non-sparse (dense) INT8 performance, derived from sparse INT8 of 427.6 TOPS ÷ 2 per NVIDIA Ada Lovelace architecture. With 2:4 structured sparsity, INT8 performance is 427.6 TOPS. Dense value used for consistent cross-vendor comparison.
- Memory bandwidth calculated as 160-bit interface × 18 Gbps ÷ 8 bits/byte = 360 GB/s.
- Fourth-generation Tensor Cores support FP8 in both E4M3 and E5M2 formats, along with TF32, BF16, FP16, INT8, and INT4 precision modes.
- CUDA Compute Capability 8.9 confirmed from TechPowerUp specifications for Ada Lovelace AD104 chip.
- The GPU uses the AD104-400 variant with 48 of 60 streaming multiprocessors enabled, fabricated on TSMC 5nm process with 35.8 billion transistors on a 294mm² die.
- Power consumption (TDP) is 130W, representing the maximum thermal design power for the single-slot form factor.
- Some sources (Lenovo) list tensor performance as 327.6 TFLOPS, but NVIDIA official sources and Leadtek consistently show 427.6 TFLOPS, which appears to be the correct peak FP8 tensor performance with sparsity.
- MSRP of $1,250 USD confirmed by two independent sources at launch (August 2023, available September 2023). CG Channel reported "The RTX 4000 has a MSRP of $1,250" at https://www.cgchannel.com/2023/08/nvidia-unveils-rtx-4000-4500-and-5000-workstation-gpus/. Develop3D reported "Nvidia RTX 4000 Ada is available in September with an ESP of $1,250" at https://develop3d.com/workstations/nvidia-rtx-4000-4500-5000-ada-generation-gpus-launch/.
- NVIDIA part numbers found include 900-5G190-2570-000 (retail/standard), 900-5G190-2270-000 (OEM variant), 900-5G190-2770-000 (Lenovo ThinkSystem variant), and 900-5G190-0070-000 (active cooling variant). Sources from B&H Photo, Amazon, Exxact, and Lenovo Press documentation.
- PCI Device ID 27B2 and Subsystem Device ID 181B identified from NVIDIA open-gpu-kernel-modules repository on GitHub. The SFF variant uses PCI Device ID 27B0 with Subsystem Device ID 16FA.
- Dell offers two variants with part numbers 490-BKHG (full height) and 490-BKPJ (half height/SFF). Source from dell.com official product pages.
- HPE part number S3T54C for server version, HP part number 8D6B7AA for workstation version. Sources from HPE Store and B&H Photo.
- Lenovo part numbers include 4X67A97287 (ThinkSystem server variant, feature code C4S1) and 4X61Q50293 (ThinkStation SFF variant). Sources from Lenovo Press and lenovo.com.
- PNY offers VCNRTX4000ADA-PB (single-slot) and VCNRTX4000ADALP-PB (SFF dual-slot). Sources from pny.com and ProVantage.
- Supermicro part number GPU-NVQRTX4000-ADA-SFF for the SFF variant. Source from Senetic UK.
- Multiple NVIDIA part number variants exist for different OEM configurations and form factors. The base part number format is 900-5G190-XXXX-XXX with different suffixes indicating specific SKU configurations.