NVIDIA L4
A low-power 24 GB inference and media GPU. More compute does not imply more memory bandwidth: its published 300 GB/s is below A10.
GDDR6
Published peak, not measured application throughput
A larger L4 VM adds CPU and host memory but does not enlarge each GPU’s 24 GB. On AWS, fractional G6f allocations can expose less than a full L4.
When this is a sensible choice
Start here if…
Try L4 for embeddings, vision, video processing and small language models that fit on one device. It is a practical cloud entry point when inference is only one part of the service.
Choose another configuration if…
If decode is dominated by repeatedly reading a large weight payload, compare bandwidth and measured latency with A10 or an HBM part. Do not choose from FP8 TOPS alone.
Specifications with their boundaries attached
- Architecture
- Ada Lovelace
- Memory
- 24 GB GDDR6
- Memory bandwidth
- 0.3 TB/s per accelerator
- Peak compute
- 121 TFLOPS dense BF16/FP16 (Google machine table)
- Scale-up interconnect
- PCIe only; no NVLink
- Host attachment
- PCIe 4.0 x16
- Power
- 72 W
- Partitioning
- No MIG
- Catalogue status
- Documented product
Compute figures are theoretical peaks at the stated precision. Dense and structured-sparse rates must not be mixed. Bandwidth labelled bidirectional combines both directions. See the source documents.
Three bandwidths, three different jobs
PCIe only; no NVLink
InfiniBand, RoCE, EFA or provider-specific transport
Where it appears in provider documentation
| Provider / machine | Network scope | What changes the decision |
|---|---|---|
AWS G4dn / G5 / G6 / G6e ↗ | Depends on VM size; no NVLink assumption | Respectively T4, A10G, L4 and L40S. A10G is AWS’s A10-family device. Fractional and full GPU offerings are different products. |
Google Cloud G2 / G4 ↗ | VM-size-specific network | G2 uses L4; G4 uses RTX PRO 6000. Check exact GPU count and guest-visible memory. |
Model fit and software support
These publisher or serving-engine documents mention this hardware family. They have not been reproduced on our machines.
No model-specific recipe in our reviewed set certifies this exact hardware. The memory calculator can narrow candidates, but it cannot establish software support. Read the model register.
What is the memory floor?
Start with total parameters, then add the memory the workload needs. This arithmetic does not certify a serving configuration. All output sizes below are decimal GB.
Override device capacity with the memory exposed by your allocation, particularly for cloud B300 and partitioned devices. The starting 32 GB budget and 15% reserve are editable teaching assumptions. They are not measurements for the selected model. Mixed-precision tensors, quantization scales, vision encoders and draft models can increase the weight payload.
Raw weights only
320B × 8 bits ÷ 8
Budget per device
24 GB × (1 − 15%)
Arithmetic lower bound
round up ((weights + 32) ÷ budget)
This assumes perfectly balanced sharding. A result of three does not prove that a three-device parallel layout is supported. Check layer/expert divisibility, actual allocatable memory, precision kernels and the fabric before renting.
320B is the model card’s total; 18B active is not its storage size. The vLLM recipe reports about 306 GiB for the native FP8 checkpoint. Hopper requires BF16 KV for this model; the documented ROCm path targets gfx950, not every Instinct GPU. Read the model source ↗
For full training, also budget gradients, optimizer states, activations and communication buffers. The inference weight estimate above is insufficient.
Tokens per second, TTFT and TPOT are not certified for this hardware in our reference. Use the benchmark checklist to compare an exact model, software revision and workload.
Nearby memory capacities, different tradeoffs
These are comparison candidates selected by memory capacity, not performance rankings or drop-in replacements.
NVIDIA A10
A 24 GB graphics-capable server GPU. It offers more memory than T4, without the HBM or switched links of A100.
NVIDIA A30
A 24 GB HBM2 accelerator for shared enterprise servers. It has more memory bandwidth than L4, but an older compute architecture.
NVIDIA T4
A small inference GPU for a model that already fits. Its 16 GB capacity is usually the first limit a new LLM user encounters.
Sources and review scope
Manufacturer and cloud documentation checked 2026-09-12. We reviewed specifications and the stated product boundaries; we did not run training or serving benchmarks on this device. Cloud memory and availability can differ by configuration.
- NVIDIA L4 specifications ↗ · checked 2026-09-12
- Google Compute Engine GPU machine types ↗ · checked 2026-09-12
Cite this reference
AI Infra Interviews. NVIDIA L4: specifications and workload fit. Reviewed . https://aiinfrainterviews.com/hardware/l4
Include your access date when citing a changing specification. Link to the specification section for hardware figures or the explanation for a sizing or workload decision.
Research method and limits
We compile manufacturer specifications, cloud documentation, model cards and serving recipes. The reference preserves source units and distinguishes individual devices from nodes and racks. Conflicting figures and unknown fields remain labelled.
Our contribution is the comparison, unit reconciliation, worked arithmetic and workload explanation. Published peaks are vendor specifications. Calculator results are estimates under the displayed assumptions. Neither is a measurement from our own accelerator lab.
For a manufacturer’s specification, consult the original source documents. Cite our page when using its analysis, and retain primary-source attribution for underlying figures. This curated reference does not establish market share, live availability or a universal performance ranking.
Found a discrepancy? Send a correction with the page URL, exact variant, disputed figure and supporting primary document.