L4 vs T4
Both run on the same nodes, the same templates and the same tiers here — so the choice comes down to memory, bandwidth and price. Every figure below is from the catalogue or the vendor datasheet.
- L4
- $0.05/h24 GB · Ada Lovelace · 96 available
- T4
- $0.04/h16 GB · Turing · 96 available
- Price gap
- 20%On spot, per GPU-hour
- Over 100 h
- $1.00Difference for the same GPU-hours
What actually separates them.
Four differences that change a decision, computed from the numbers rather than asserted.
Bigger models fit on the L4
24 GB against 16 GB. At FP16 that is about 9B parameters on one GPU, against 6B.
The T4 costs 20% less per hour
$0.04 against $0.05 on spot — $1.00 of difference over 100 GPU-hours.
Token generation is faster on the T4
320 GB/s against 300 GB/s. Inference reads the whole model from memory for every token, so bandwidth, not FLOPS, usually sets the ceiling.
More memory per dollar on the L4
480 GB per dollar per hour against 400 GB. The usual tie-breaker when both cards fit the job.
Every figure we hold on both cards.
A row is marked only when both values exist and one is genuinely better. L4 leads on 6 rows, T4 on 4.
| L4 | T4 | |
|---|---|---|
| GPU memoryDecides which models fit at all | 24 GB GDDR6 | 16 GB GDDR6 |
| Memory bandwidthSets token throughput on inference | 300 GB/s | 320 GB/s |
| FP16 tensor | 121 TFLOPS | 65 TFLOPS |
| FP32 | 30 TFLOPS | 8 TFLOPS |
| 8-bit float (FP8)Halves the memory a model needs, where the stack supports it | Supported | Not supported |
| Architecture | Ada Lovelace (2023) | Turing (2018) |
| Board powerLower is cheaper to run at scale — for us, not for your bill | 72 W | 70 W |
| GPU-to-GPU linkOnly matters for multi-GPU training | PCIe only | PCIe only |
| Node sizes | 1×, 2×, 4× | 1×, 2×, 4× |
| Spot price | $0.05/h | $0.04/h |
| On-demand price | $0.08/h | $0.07/h |
| Memory per dollarGPU memory per $1 of spot time, per hour | 480 GB | 400 GB |
| 30-day reclaim rateShare of spot instances reclaimed | <5% | <5% |
| Available now | 96 GPUs | 96 GPUs |
Specifications come from each vendor's datasheet — linked on the L4 page and the T4 page. Prices are ours, per GPU-hour, billed per minute.
The same models, on one GPU of each.
Weights only, with 20% of the memory kept free for activations and the KV cache. A cross means one GPU is not enough — the model page has the multi-GPU arithmetic.
| Model | Precision | L4 24 GB | T4 16 GB |
|---|---|---|---|
| Llama 3.1 8B8B parameters | FP16 · 16 GB | Fits | Needs 2× node |
| INT4 · 4 GB | Fits | Fits | |
| Qwen2.5 32B32B parameters | FP16 · 64 GB | Needs 4× node | Needs more than one node |
| INT4 · 16 GB | Fits | Needs 2× node | |
| Llama 3.3 70B70B parameters | FP16 · 140 GB | Needs more than one node | Needs more than one node |
| INT4 · 35 GB | Needs 2× node | Needs 4× node | |
| Mixtral 8x22B141B parameters | FP16 · 282 GB | Needs more than one node | Needs more than one node |
| INT4 · 71 GB | Needs 4× node | Needs more than one node |
The same 100 GPU-hours on each.
Compute only, billed per minute. Storage is $0.08 per GB-month on either card.
| L4 | T4 | |
|---|---|---|
| One hour, spot | $0.05 | $0.04 |
| One day, spot | $1.20 | $0.96 |
| 100 GPU-hours, spot | $5.00 | $4.00 |
| 100 GPU-hours, on-demand | $8.00 | $7.00 |
| A month, reserved | $43.80 | $36.50 |
Take the L4 when…
- Fine-tuning small models — LoRA on 7B–13B models, and full fine-tuning below 3B.
- Image and video generation — Diffusion pipelines run entirely in memory at this capacity.
- Anything that can checkpoint — On the spot tier this model is 38% below its own on-demand price, with a 2-minute notice before a reclaim.
Take the T4 when…
- Quantised inference — Small and medium models in 4-bit or 8-bit form, with the whole model resident in memory.
- Computer vision and batch jobs — Detection, classification, embeddings and transcoding at a very low hourly rate.
- Anything that can checkpoint — On the spot tier this model is 43% below its own on-demand price, with a 2-minute notice before a reclaim.
L4 or T4?
Which is cheaper, the L4 or the T4?
The T4, at $0.04 per GPU-hour on spot against $0.05 — 20% less. On-demand: $0.07 against $0.08.
Which one should I take for inference?
Take the one that holds the model with room for the KV cache, then the one with more memory bandwidth. Here that is the L4 on capacity (24 GB) and the T4 on bandwidth.
Can I run both on the same job?
Not inside one instance — a node holds GPUs of a single model. You can run two instances in parallel, on the same persistent volume in turn, or split the work: a big card for the model that needs capacity, a cheap one for the parts that do not.
Other comparisons: the full list · all 47 models
Try both for less than the price of deciding.
An hour on each costs $0.09 on spot. Nothing renews, and your disk moves between them.
Billed per minute from the moment the instance is reachable. Minimum credit $40, no subscription.