Skip to content
Updated minutes ago
intel

Gaudi 3

intel · gaudi · 128 GB HBM2e · 900W TDP

VRAM

128 GB

BF16 TFLOPS

1835

Bandwidth

3700 GB/s

From

$2.30/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM128 GB HBM2e
Memory Bandwidth3700 GB/s
BF16 TFLOPS1835
FP16 TFLOPS1835
FP8 TFLOPS3670
INT8 TOPS3670
TDP900W
InterconnectOAM
Max per Node8
PCIe Gen5
Tensor CoresNo

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
gcp$3.20/hr$2.30/hr-Cheapest
aws$3.78/hr$2.65/hr-

Compatible Models (278)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model2 GPUs1 GPU
13B model2 GPUs1 GPU
70B model11 GPUs1 GPU

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
0.56 t/s/WFP8
Qwen 2.5 7B
0.54 t/s/WFP8
Llama 3.1 8B
0.51 t/s/WFP8
DeepSeek V3
0.11 t/s/WFP8
Llama 3.1 70B
0.06 t/s/WFP8
Qwen 2.5 72B
0.06 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
Gaudi 3 HL-325L128 GB18353700$2.65/hr
Gaudi 296 GB4322460$1.65/hr
Instinct MI250X128 GB3833277$0.79/hr
H200 SXM141 GB9904800$2.69/hr
H2096 GB1484000$0.99/hr

Methodology Note

Performance estimates for the Gaudi 3are based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (128 GB HBM2e available), KV-cache allocation, and activation memory. Throughput predictions use the Gaudi 3's rated 3700 GB/s memory bandwidth and 1835 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (gaudi). See our full methodology.

Frequently Asked Questions

How many AI models can run on Gaudi 3?

The Gaudi 3 can run 278 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 8x Gaudi 3.

What is the Gaudi 3 inference throughput?

The Gaudi 3 delivers 1835 BF16 TFLOPS and 3670 FP8 TFLOPS with 3700 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does Gaudi 3 cost per hour?

The Gaudi 3 is available starting from $2.30/hour via gcp. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.