I'd love to hear what you think! Please drop me a line and let me know what you like and what could be better. 🙏
NVIDIA L4 24GB Specs, Benchmarks & Pricing
The NVIDIA L4 GPU was introduced in March 2023 with part number 900-2G193-0000-000. It, along with the L40 are the NVIDIA Ada Lovelace architecture built using the Built on the 5 nm process. The L4 uses the AD104 graphics processor, fourth-generation Tensor Cores, deep learning super sampling (DLSS 3) and 24 GB GDDR6 memory. The L4 is unique in that its maximum power draw is only 72 W which is quite low compared with other Tensor Core GPUs such as the such as the L40 at 300W and L40S at 350W. It is a single-slot PCIe 4.0 x16 card without connectivity for monitors.
- Release Date: March 21, 2023
- MSRP: $7,000 USD†
- GPU Architecture: ada
- Hardware-Accelerated GEMM Operations:FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
- CUDA Compute Capability : 8.9
† Estimated MSRP of $7,000 USD for the NVIDIA L4 based on OEM list prices. NVIDIA does not publish official MSRP for data center GPUs. OEM list prices corroborated by 2 independent sources: Lenovo ThinkSystem NVIDIA L4 (4X67A84824) has MSRP of $7,049 per Bluum, and Cisco UCSC-GPU-L4 is listed at $6,788 per CompSource. Note: Current market/resale prices (~$2,500-$3,000) are significantly lower due to depreciation but do not reflect original launch list pricing.
Considerations
- Lower memory bandwidth compared to other GPUs
- Lower FP8 matrix performance compared to other GPUs
Specifications for NVIDIA L4
| Specification | Performance Ranking |
|---|---|
| FP32 TFLOPs | |
| FP16 TFLOPs | |
| Tensor Core Count | |
| Memory Capacity (GB) | |
| Memory Bandwidth (GB/s) | |
| Int8 TOPs | |
| FP8 TFLOPs | |
| FP4 TFLOPs | Data not available |
Real-time NVIDIA L4 GPU Prices
We're tracking 65 of the NVIDIA L4 GPUs currently available for sale. The lowest average price is $3,189 (what's this?)
Buy NowCompare 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 L4 GPU to other GPUs.
Compare GPU Price/PerformanceCompare NVIDIA L4 to Another GPU
Compare the NVIDIA L4 directly to another GPU to see specs, benchmarks, and prices side-by-side.
Compare GPUs Side-by-SidePrice History
NVIDIA L4 Price History
This chart tracks NVIDIA L4 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 26.0%. The current lowest average is $3,632, compared to a typical $3,365 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 (6)
- Product SKU
- PG193 SKU 200
- NVIDIA Part Number
- 699-2G193-0200-xxx
- NVIDIA Part Number
- 900-2G193-0000-000
- NVIDIA Part Number
- 900-2G193-0000-001
- NVIDIA Part Number
- 900-2G193-0100-001
- NVIDIA Part Number
- 900-2G193-0200-000
Available from 6 Partners (13 products)
Dell
- NVIDIA L4, PCIe, 72-Watt, 24GB Passive, Single Wide Full Height GPU
- 540-BDTP(sku)
- NVIDIA L4, PCIe, 72-Watt, 24GB Passive, Single Wide Full Height GPU
- NG3PY(part number)
- NVIDIA L4, PCIe, 72-Watt, 24GB Passive, Single Wide Low Profile GPU
- 540-BDTM(sku)
- NVIDIA L4 PCIe 72-Watt 24GB Passive Single Wide Full Height GPU (Dell OEM)
- 900-2G193-0100-001(part number)
HPE
- NVIDIA L4 24GB PCIe Accelerator for HPE
- S0K89C(sku)
- NVIDIA L4 24GB PCIe Accelerator for HPE
- P59360-001(part number)
Lenovo
- ThinkSystem NVIDIA L4 24GB PCIe Gen4 Passive GPU
- 4X67A84824(part number)
Cisco
- NVIDIA L4 70W 24GB 1-Slot GPU
- UCSC-GPU-L4(part number)
- NVIDIA L4 70W 24GB 1-Slot GPU
- UCSC-GPU-L4=(part number)
- NVIDIA L4 Tensor Core, 70W, 24GB (X-Series)
- UCSX-GPU-L4=(part number)
PNY
- NVIDIA L4 Graphic Card - Low-profile - 24 GB
- NVL4TCGPU-KIT(part number)
- NVIDIA L4 Graphic Card - Full Height - 24 GB
- NVL4TCGPUATX-KIT(part number)
Supermicro
- NVIDIA Ada L4 24GB GDDR6 PCIe 4.0 Graphics Card
- GPU-NVL4(part number)
References
- https://www.nvidia.com/en-us/data-center/l4/
- https://nvdam.widen.net/s/rvq98gbwsw/l4-datasheet-2595652
- https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf
- https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
- https://www.techpowerup.com/gpu-specs/l4.c4091
- https://www.nvidia.com/en-us/technologies/ada-architecture/
- https://developer.nvidia.com/blog/supercharging-ai-video-and-ai-inference-performance-with-nvidia-l4-gpus/
- https://buy.bluum.com/lenovo-thinksystem-nvidia-l4-pcie-gen4-passive-graphic-card-24-gb/
- https://www.compsource.com/buy/UCSCGPUL4/Cisco-91
- https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/l4/PB-11316-001_v01.pdf
- https://www.dell.com/en-us/shop/nvidia-l4-pcie-72-watt-24gb-passive-single-wide-full-height-gpu-customer-install/apd/540-bdtp/graphic-video-cards
- https://www.hpe.com/us/en/product-catalog/options/pip.specifications.nvidia-accelerators.1014731952.html
- https://lenovopress.lenovo.com/lp1717-thinksystem-nvidia-l4-24gb-pcie-gen4-passive-gpu
- https://www.cisco.com/c/dam/en/us/products/collateral/servers-unified-computing/ucs-c-series-rack-servers/nvidia-l4-gpu.pdf
- https://www.pny.com/nvidia-l4
- https://store.supermicro.com/supermicro-nvidia-ada-l4-24gb-gddr6-pcie-4-0-graphics-card.html
Notes
- int8TOPS of 242.5 represents non-sparse (dense) INT8 performance, derived from sparse INT8 of 485 TOPS ÷ 2 per NVIDIA Ada architecture whitepaper. With 2:4 structured sparsity, INT8 performance is 485 TOPS. Dense value used for consistent cross-vendor comparison.
- fp8TFLOPS of 242.5 represents non-sparse (dense) FP8 performance. Per the official NVIDIA L4 datasheet (PB-11316-001_v01.pdf), the FP8 Tensor Core row and INT8 Tensor Core row publish identical dense/sparse figures — NVIDIA's Ada Lovelace 4th-generation Tensor Cores process FP8 and INT8 at identical per-clock throughput since both are 8-bit formats. Dense FP8 = dense INT8 = 242.5 TFLOPS. Dense value used for consistent cross-vendor comparison.
- Ada Lovelace has no FP4 hardware path (FP4 tensor core support was introduced in Blackwell) — fp4TFLOPS is left null and is a confirmed hardware negative, not a research gap, so it is not added to unverifiedSpecs.
- Estimated MSRP of $7,000 USD for the NVIDIA L4 based on OEM list prices. NVIDIA does not publish official MSRP for data center GPUs. OEM list prices corroborated by 2 independent sources: Lenovo ThinkSystem NVIDIA L4 (4X67A84824) has MSRP of $7,049 per Bluum, and Cisco UCSC-GPU-L4 is listed at $6,788 per CompSource. Note: Current market/resale prices (~$2,500-$3,000) are significantly lower due to depreciation but do not reflect original launch list pricing.
- Manufacturer identifier 'PG193 SKU 200' sourced from NVIDIA official product brief PB-11316-001_v01 at https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/l4/PB-11316-001_v01.pdf
- NVPN 699-2G193-0200-xxx is the official NVIDIA part number designation per NVIDIA product brief PB-11316-001_v01. Multiple variants exist with different suffix codes (900-2G193-0000-000, 900-2G193-0000-001, 900-2G193-0100-001, 900-2G193-0200-000) representing different form factors and OEM configurations.
- Dell part numbers 540-BDTP (full-height) and 540-BDTM (low-profile) sourced from official Dell product pages at dell.com
- HPE part numbers S0K89C (SKU) and P59360-001 (spare part) sourced from https://www.hpe.com/us/en/product-catalog/options/pip.specifications.nvidia-accelerators.1014731952.html
- Lenovo part number 4X67A84824 sourced from Lenovo Press product guide at https://lenovopress.lenovo.com/lp1717-thinksystem-nvidia-l4-24gb-pcie-gen4-passive-gpu
- Cisco part numbers UCSC-GPU-L4 (C-Series) and UCSX-GPU-L4= (X-Series) sourced from Cisco official datasheet at https://www.cisco.com/c/dam/en/us/products/collateral/servers-unified-computing/ucs-c-series-rack-servers/nvidia-l4-gpu.pdf
- PNY part numbers NVL4TCGPU-KIT (low-profile) and NVL4TCGPUATX-KIT (full-height) sourced from https://www.pny.com/nvidia-l4 and verified via multiple retailers
- Supermicro part number GPU-NVL4 sourced from https://store.supermicro.com/supermicro-nvidia-ada-l4-24gb-gddr6-pcie-4-0-graphics-card.html
- fp32TFLOPS of 60 is TF32 Tensor Core throughput without sparsity, per NVIDIA's L4 page (nvidia.com/en-us/data-center/l4/), which lists 120 TFLOPS with sparsity. FP32 vector (CUDA core) throughput is 30.3 TFLOPS.