VRAM
128 GB
TDP
560 W
MSRP
$14k
Perf/W
0.09 tok/s/W
Cost/1K tok
$0.0028/k
Tested
2024-11-12
AMD Instinct MI250X 128GB produces approximately 52.0 tok/s on Llama 3 70B at Q4_K_M quantization (batch 1, 8192-token context, 128GB VRAM, 560W TDP). That figure comes from 1 measured run on llama.cpp. The 70B model needs roughly 40GB of VRAM at Q4, so headroom and KV cache budget matter as much as raw throughput.
AMD Instinct MI250X 128GB — CDNA2 datacenter accelerator. Dual-die design with 128GB HBM2e at 3.2 TB/s, 560W TDP. AMD's pre-MI300 datacenter GPU with mature ROCm support.
128GB VRAM at 3.2 TB/s bandwidth — competitive with H100 for memory-bound inference. ROCm 6.x support is production-grade for vLLM and llama.cpp. ~110 tok/s on 12B Q4. Best for: AMD-friendly datacenter deployments, large-model inference (70B+ FP16), and organizations with existing ROCm infrastructure.
AMD Instinct MI250X 128GB with 128GB VRAM at 560W TDP, scored across 4 workloads with 4 benchmark records.
Best workload
Llama 3 8B FP16
188 tok/s
Quantization
FP16
4K context · batch 1
| Workload | Score | Quant | Context | σ | Status | Tested |
|---|---|---|---|---|---|---|
| Llama 3 70B Q4 llm | 52.0tok/s | Q4_K_M | 8K | , | Curated Aggregate | 2024-11-12 |
| Llama 3 8B FP16 llm | 188.0tok/s | FP16 | 4K | , | Curated Aggregate | 2024-11-19 |
| Mistral 7B llm | 135.0tok/s | Q4_K_M | 4K | , | Curated Aggregate | 2024-04-08 |
| Gemma 2 9B llm | 110.0tok/s | Q4_K_M | 8K | , | Curated Aggregate | 2024-07-31 |
AMD Instinct MI250X 128GB clears a 8.5/10 based on workload-normalized throughput, memory headroom, efficiency, value, trust, and coverage.
What models fit this 128GB card at different quantization levels.
Trust score
6/10
MyAI rating
8.5
Runs
1
Freshness
Stale
Source-linked row with explicit verification status.
Tested on 2024-11-19; 646 days old.
Enterprise buying note
Retailer we'd check first
Amazon search plus reseller quotes
For datacenter and accelerator parts, Amazon is useful for spotting live listings, accessories, or used pulls, but serious procurement usually happens through integrators, brokers, or cloud partners.
Enterprise procurement, not retail
This silicon is typically acquired through an authorized OEM partner, system integrator, or hyperscaler reseller. For on-demand access, compare hourly rates at RunPod, Vast.ai, Lambda Labs, or your existing cloud provider before committing to capital expenditure.
This is a strong buy when the listing stays close to reference pricing and matches the workload you actually run.
Affiliate note: we do not have a device-level ASIN yet, so this opens tagged Amazon search results for the exact product name.