Skip to content
Updated minutes ago
amd

Instinct MI250X

amd · cdna2 · 128 GB HBM2e · 560W TDP

VRAM

128 GB

BF16 TFLOPS

383

Bandwidth

3277 GB/s

From

$0.79/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM128 GB HBM2e
Memory Bandwidth3277 GB/s
BF16 TFLOPS383
FP16 TFLOPS383
FP8 TFLOPS383
INT8 TOPS383
TDP560W
InterconnectINFINITY-FABRIC
Max per Node8
PCIe Gen4
Tensor CoresNo

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
fluidstack$1.09/hr-$0.79/hrCheapest
vast_ai$1.20/hr-$0.85/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
0.80 t/s/WFP8
Qwen 2.5 7B
0.77 t/s/WFP8
Llama 3.1 8B
0.73 t/s/WFP8
DeepSeek V3
0.16 t/s/WFP8
Llama 3.1 70B
0.08 t/s/WFP8
Qwen 2.5 72B
0.08 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
Instinct MI21064 GB1811638$0.99/hr
Gaudi 3128 GB18353700$2.30/hr
Gaudi 3 HL-325L128 GB18353700$2.65/hr
H200 SXM141 GB9904800$2.69/hr
H2096 GB1484000$0.99/hr

Methodology Note

Performance estimates for the Instinct MI250Xare 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 Instinct MI250X's rated 3277 GB/s memory bandwidth and 383 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (cdna2). See our full methodology.

Frequently Asked Questions

How many AI models can run on Instinct MI250X?

The Instinct MI250X 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 Instinct MI250X.

What is the Instinct MI250X inference throughput?

The Instinct MI250X delivers 383 BF16 TFLOPS and 383 FP8 TFLOPS with 3277 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 Instinct MI250X cost per hour?

The Instinct MI250X is available starting from $0.79/hour via fluidstack. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.