MSI Codex R (i5-12400F Test Bench)
Identifier: Intel Core i5-12400F on MSI PRO B760-VC WIFI (MS-7D98) Form factor: mid-tower desktop (“MSI Codex R” prebuilt chassis) Status: active — primary GPU swap test bench
The CPU+motherboard combination I use as a stable platform to benchmark different GPUs against the same baseline. The 12400F is intentionally modest — it’s a known constant, not the star.
Confirmed via Geekbench detail pages: all 12 captured runs use the same MSI PRO B760-VC WIFI motherboard and 128 GB of RAM — the only thing that swapped was the GPU. The board is the DDR5 variant (the user attributed two Corsair Vengeance RGB DDR5 kits to this rig).
Build & Cost
| Component | Spec | Cost |
|---|---|---|
| Chassis | MSI Codex R prebuilt (case + PSU + base RAM) | $814.68 (2023) |
| Motherboard | MSI PRO B760-VC WIFI (MS-7D98) — DDR5 variant | included |
| CPU | Intel Core i5-12400F (6P/12T, Alder Lake) | included |
| RAM upgrade | 2 × Corsair Vengeance RGB 64 GB (2×32) DDR5-5200 CL40 = 128 GB | $355.98 (2024) |
| GPU: NVIDIA RTX 5090 | 32 GB | $1,999.00 (2025) |
| GPU: NVIDIA RTX 4070 Ti | 12 GB | $653.24 (2024) |
| GPU: NVIDIA RTX 3090 | 24 GB | $1,132.45 (2025) |
| GPU: NVIDIA RTX 4060 | 8 GB | included with Codex R prebuilt (est. ~$300 market value 2023) |
| GPU: NVIDIA RTX 3060 | 12 GB | $706.05 (2022, GPU shortage pricing) |
| Chassis + RAM upgrade | $1,170.66 |
Geekbench AI Runs
GPU swap matrix
| GPU | VRAM | Backend | Single | Half | Quant | Result |
|---|---|---|---|---|---|---|
| RTX 5090 | 32 GB | ONNX/DirectML | 42,496 | 61,328 | 31,856 | 357342 |
| RTX 5090 | 32 GB | ONNX/DirectML | 42,008 | 60,655 | 31,363 | 341704 |
| RTX 5090 | 32 GB | ONNX/DirectML | 41,400 | 60,038 | 30,718 | 357657 |
| RTX 4070 Ti | 12 GB | ONNX/DirectML | 27,553 | 39,999 | 21,827 | 191639 |
| RTX 3090 | 24 GB | ONNX/DirectML | 21,048 | 29,810 | 13,563 | 357654 |
| RTX 3090 | 24 GB | ONNX/DirectML | 20,978 | 29,751 | 13,562 | 341706 |
| RTX 4060 | 8 GB | ONNX/DirectML | 14,422 | 22,768 | 11,696 | 191595 |
| RTX 4060 | 8 GB | ONNX/DirectML | 14,279 | 22,829 | 11,400 | 191524 |
| RTX 3060 | 12 GB | ONNX/DirectML | 13,582 | 23,007 | 9,272 | 191592 |
| RTX 3060 | 12 GB | ONNX/DirectML | 12,346 | 19,867 | 8,819 | 177647 |
CPU baseline
AI Capability
| Configuration | Best Single | Best Half | Best Quant | ACI | Tier | GPU cost | GPU $/ACI |
|---|---|---|---|---|---|---|---|
| + RTX 5090 | 42,496 | 61,328 | 31,856 | 920.4 | S | $1,999 | $2.17 |
| + RTX 4070 Ti | 27,553 | 39,999 | 21,827 | 642.4 | A | $653 | $1.02 ⭐ |
| + RTX 3090 | 21,048 | 29,810 | 13,563 | 444.5 | B | $1,132 | $2.55 |
| + RTX 4060 | 14,422 | 22,829 | 11,696 | 326.7 | B | included (~$300 est.) | ~$0.92 |
| + RTX 3060 | 13,582 | 23,007 | 9,272 | 311.4 | B | $706 | $2.27 |
The **RTX 4070 Ti at 706 looks expensive in 2026 hindsight but reflects 2022 GPU shortage pricing.
The 5090 configuration is the highest-scoring rig in this collection. Note that the i5-12400F bottleneck is real on the 5090 but doesn’t show in Geekbench AI’s compute-bound tests — only LLM serving with high batch sizes will reveal the PCIe/CPU ceiling.
See Also
- Benchmark Methodology
- Proxmox host (Ultra 7 265K) — also tested with 5090 and 3090
- Intel NUC9V7QNX — same physical RTX 4060 GPU on an older 9th-gen Intel mobile platform; direct CPU-bottleneck comparison
- Gemma 4 VRAM Requirements — VRAM-driven model picks