Build snapshot

Build cbagent16k1

Build cbagent16k1 · Saved Jun 8, 2026 · Rules v0.1.0 · Quant Q4_K_M

Immutable snapshot — prices and compatibility frozen at save time

Compare

Profile

Agent Host

Target model

Qwen 3 32B

Snapshot total

$1,711.98

Parts list (snapshot)

NVIDIA GeForce RTX 4090
NVIDIA · Qty 2
$762.00
- AmazonBest price2.1× MSRPnew · in stock$3,399.95as of Jul 27, 2026LIVE(opens in new tab)
View part details →
AMD Ryzen 9 7950X
AMD · Qty 1
Price Unknown
View part details →
ASUS ROG STRIX X670E-E GAMING WIFI
ASUS · Qty 1
Price Unknown
- AmazonBest pricenew · in stock$439.99as of Jul 27, 2026LIVE(opens in new tab)
View part details →
G.Skill Trident Z5 64GB (2x32GB)
G.Skill · Qty 1
$649.99
- AmazonBest price5.2× MSRP$934.99MANUAL(opens in new tab)
View part details →
Samsung 990 PRO 2TB NVMe
Samsung · Qty 1
$299.99
View part details →
Corsair RM1000x (2021) 1000W
Corsair · Qty 1
Price Unknown
View part details →
Fractal Design Meshify 2
Fractal Design · Qty 1
Price Unknown
- AmazonBest pricenew · in stock$164.99as of Jul 27, 2026LIVE(opens in new tab)
View part details →
NZXT Kraken X73 360mm AIO
NZXT · Qty 1
Price Unknown
- AmazonBest pricenew · in stock$103.31as of Jul 27, 2026LIVE(opens in new tab)
View part details →

Issues found

PSU provides 1000 W but estimated load requires 1284 W (1070 W TDP × 1.2 headroom).
PSU lacks 12VHPWR connector required by NVIDIA GeForce RTX 4090.

1 warning

2 GPU(s) would run at PCIe x8 or below in this multi-GPU config.
Use a platform with more CPU PCIe lanes for full x16 per GPU.

Compatibility checks are advisory. Verify fit, power, and cooling before purchase.

Workload Fit

Qwen 3 32BFits

48 GB avail23.4 GB req

Est. Performance

Power Draw

GPU TDP900 W

CPU TDP170 W

System (RAM + storage + fans)55 W

Total estimated draw1125 W

PSU capacity1000 W

PSU utilization113% (high)

System overhead includes 30 W RAM, 10 W storage, 15 W fans/board. Estimated — verify before purchase.

Unable to estimate rating.

Total covers parts with a live or user-entered price; others show “—”.

Consider Apple Silicon

For lighter local inference, Mac Studio M3 Ultra (96GB) ($5,299) runs ~68 tok/s on Llama 8B with MLX — no GPU drivers required.

View Mac config details All options →

Verdict: Cloud-vs-buy recommendation, power TCO, and value/grade — prefilled from this snapshot.

Cloud vs. buy

Verdict:Cloud rental likely cheaper· 19.86 tok/s per $1k

Full buy TCO (power incl.): $1,776.78 vs cloud $508.80

Hours / month80h

Time horizon

Electricity ¢/kWh15¢

Default 15¢/kWh. Changes update power cost + totals live.

Cloud GPU class (auto from build GPU)

Sales tax %

Build cost$1,711.98

Power cost (80h × 12mo @ 15¢)$64.80

Buy total (incl. power)$1,776.78

Cloud (RTX 4090)$508.80 / 12mo

Break-even at ~41 months ($42.40/mo)

Cloud rental likely cheaper

Rent on Vast.ai Rent on RunPod

Cloud pricing is indicative. See affiliate disclosure. Power uses GPU TDP when provided by build.

Export build

| Part | Qty | Price |
|------|-----|-------|
| NVIDIA GeForce RTX 4090 | 2 | $381 |
| AMD Ryzen 9 7950X | 1 | Price Unknown |
| ASUS ROG STRIX X670E-E GAMING WIFI | 1 | Price Unknown |
| G.Skill Trident Z5 64GB (2x32GB) | 1 | $650 |
| Samsung 990 PRO 2TB NVMe | 1 | $300 |
| Corsair RM1000x (2021) 1000W | 1 | Price Unknown |
| Fractal Design Meshify 2 | 1 | Price Unknown |
| NZXT Kraken X73 360mm AIO | 1 | Price Unknown |

**Total:** $1,712
**Profile:** AGENT_HOST
**Model:** qwen-3-coder
**Compa…

Affiliate disclosure