microsoft/Phi-3-medium-4k-instruct
14B params · est. 8.4 GB Q4
Head-to-head
Compare recorded specifications and inspect source links for both devices. Speed comparisons appear only when workload, runtime version, quantization, context and batch match.
Device profile
70B FP16 or smaller quantized artifacts with measured context headroom. Standard 405B Q4 exceeds 192GB before runtime overhead.
Matched rows below
192 GB
750 W
Not scored
Device profile
Large quantized models that fit after runtime and context allocations; 405B FP8 does not fit a single 141GB device.
Matched rows below
141 GB
700 W
Not scored
Speed comparisons require matching runtime settings. A missing score means insufficient comparative evidence.
AMD Instinct MI300X 192GB wins VRAM capacity.
NVIDIA H200 141GB has the lower published power rating.
| Metric | AMD Instinct MI300X 192GB | NVIDIA H200 141GB |
|---|---|---|
| MyAI rating | Not scored | Not scored |
| Speed comparison | See matched rows below | See matched rows below |
| VRAM | 192 GB | 141 GB |
| TDP | 750 W | 700 W |
| MSRP | $15k | $35k |
| Workloads | 12 | 11 |
This shortlist cross-references the external OpenEvals snapshot with an approximate Q4-class VRAM estimate. Use it to sanity-check whether the hardware you are comparing can host strong open models, not just benchmark toy workloads.
microsoft/Phi-3-medium-4k-instruct
14B params · est. 8.4 GB Q4
Qwen/Qwen2-72B
73B params · est. 43.6 GB Q4
microsoft/Phi-3.5-mini-instruct
3.8B params · est. 2.3 GB Q4
internlm/internlm2_5-7b-chat
7.7B params · est. 4.6 GB Q4