Full codebase review

Feed a mid-size repository through a coding model and get notes back on every file.

The pipeline

Read and annotate ~60k tokens of code, qwen2-5-coder-14b

Fastest card where every stage is a real measurement: NVIDIA B300, 6.3 min for the whole job.

Every GPU, slowest job to fastest

GPU	Total time	Compute	Energy	Basis
NVIDIA H100 NVL	5.9 min	5.9 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA B300	6.3 min	6.3 min	38.61 Wh	all 1 stages measured
NVIDIA GH200 Grace Hopper	6.6 min	6.6 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA B200	6.6 min	6.6 min	44.06 Wh	all 1 stages measured
NVIDIA GeForce RTX 5090	6.7 min	6.7 min	13.03 Wh	all 1 stages measured
NVIDIA H200	6.7 min	6.7 min	13.86 Wh	all 1 stages measured
NVIDIA RTX PRO 6000 Blackwell Workstation Edition	6.8 min	6.8 min	27 Wh	all 1 stages measured
NVIDIA H100 80GB HBM3	6.9 min	6.9 min	26.8 Wh	all 1 stages measured
NVIDIA H800 80GB	6.9 min	6.9 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA B100	7 min	7 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition	7.1 min	7.1 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX PRO 6000 Blackwell Server Edition	7.4 min	7.4 min	32.21 Wh	all 1 stages measured
NVIDIA RTX PRO 5000 Blackwell	8.7 min	8.7 min	26.33 Wh	all 1 stages measured
NVIDIA GeForce RTX 4090	10.5 min	10.5 min	30.47 Wh	all 1 stages measured
NVIDIA A800 80GB	11.2 min	11.2 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA A100 80GB SXM4	11.2 min	11.2 min	32.59 Wh	all 1 stages measured
NVIDIA A100 80GB PCIe	11.3 min	11.3 min	33.67 Wh	all 1 stages measured
NVIDIA GeForce RTX 3090 Ti	11.3 min	11.3 min	49.88 Wh	all 1 stages measured
NVIDIA RTX 6000 Ada Generation	11.4 min	11.4 min	43.65 Wh	all 1 stages measured
NVIDIA H100 PCIe	11.6 min	11.6 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA A100 40GB SXM4	11.7 min	11.6 min	35.27 Wh	all 1 stages measured
GeForce RTX 5070 Ti	12 min	12 min	44.78 Wh	all 1 stages measured
GeForce RTX 5080	12.2 min	12.2 min	29.75 Wh	all 1 stages measured
NVIDIA RTX PRO 4500 Blackwell	12.5 min	12.5 min	21.85 Wh	all 1 stages measured
NVIDIA GeForce RTX 3090	12.7 min	12.7 min	37.63 Wh	all 1 stages measured
NVIDIA GeForce RTX 3080 Ti	12.8 min	12.8 min	64.96 Wh	all 1 stages measured
NVIDIA L40S	13.5 min	13.4 min	49.6 Wh	all 1 stages measured
NVIDIA L40	13.5 min	13.5 min	37.92 Wh	all 1 stages measured
NVIDIA A100 40GB PCIe	14.1 min	14.1 min	n/a	anchored estimate (0/1 stages measured)
GeForce RTX 4080 Super	14.2 min	14.2 min	33.05 Wh	all 1 stages measured
NVIDIA GeForce RTX 4080	14.3 min	14.3 min	44.74 Wh	all 1 stages measured
NVIDIA RTX A5500	14.3 min	14.3 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 7900 XTX	14.3 min	14.3 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX A6000	14.6 min	14.6 min	41.58 Wh	all 1 stages measured
AMD Radeon Pro W7900	15 min	15 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA GeForce RTX 4070 Ti Super	15.2 min	15.2 min	50.72 Wh	all 1 stages measured
NVIDIA RTX A5000	15.4 min	15.4 min	46.13 Wh	all 1 stages measured
NVIDIA RTX 5880 Ada Generation	15.8 min	15.8 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 7900 XT	16.1 min	16.1 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX PRO 4000 Blackwell	16.5 min	16.5 min	33.79 Wh	all 1 stages measured
NVIDIA Titan RTX	16.7 min	16.7 min	42.48 Wh	all 1 stages measured
NVIDIA A40	17.3 min	17.3 min	54.16 Wh	all 1 stages measured
NVIDIA RTX 5000 Ada Generation	17.6 min	17.6 min	59.48 Wh	all 1 stages measured
NVIDIA TITAN V	18.4 min	18.4 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX A4500	18.4 min	18.4 min	42.18 Wh	all 1 stages measured
NVIDIA GeForce RTX 4070 Super	19.5 min	19.5 min	46.03 Wh	all 1 stages measured
NVIDIA GeForce RTX 4070	19.7 min	19.7 min	43.13 Wh	all 1 stages measured
NVIDIA GeForce RTX 4070 Ti	19.7 min	19.7 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA TITAN Xp	19.9 min	19.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 9070 XT	21.2 min	21.2 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 9070	21.2 min	21.2 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA A10G	21.3 min	21.2 min	43.41 Wh	all 1 stages measured
NVIDIA Quadro RTX 8000	21.4 min	21.4 min	49.24 Wh	all 1 stages measured
NVIDIA Quadro RTX 6000 (Turing)	21.5 min	21.5 min	62.2 Wh	all 1 stages measured
GeForce GTX 1080 Ti	22.1 min	22.1 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA Quadro RTX 5000	22.2 min	22.2 min	n/a	anchored estimate (0/1 stages measured)
GeForce RTX 5060 Ti	22.2 min	22.2 min	45.98 Wh	all 1 stages measured
NVIDIA TITAN X (Pascal)	22.3 min	22.3 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon Pro W7800	22.7 min	22.7 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 6950 XT	22.7 min	22.7 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX 4500 Ada Generation	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon Pro W6800	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 6800 XT	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 6800	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 6900 XT	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 7800 XT	22.9 min	22.9 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX A4000	24.3 min	24.3 min	30.11 Wh	all 1 stages measured
AMD Radeon RX 7700 XT	26.9 min	26.9 min	n/a	anchored estimate (0/1 stages measured)
Intel Arc A770 Limited Edition	27.3 min	27.3 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX 4000 (Ada Generation)	27.5 min	27.5 min	38.11 Wh	all 1 stages measured
NVIDIA GeForce RTX 3060	28.2 min	28.2 min	69.16 Wh	all 1 stages measured
NVIDIA GeForce RTX 4060 Ti 16GB	32.1 min	32.1 min	67.32 Wh	all 1 stages measured
NVIDIA L4	36.4 min	36.4 min	37.51 Wh	all 1 stages measured
Intel Arc Pro A60	36.8 min	36.8 min	n/a	anchored estimate (0/1 stages measured)
AMD Radeon RX 6700	37.3 min	37.3 min	n/a	anchored estimate (0/1 stages measured)
NVIDIA RTX 2000 Ada Generation	42.8 min	42.8 min	32.76 Wh	all 1 stages measured
NVIDIA T4	50.7 min	50.6 min	51.37 Wh	all 1 stages measured

Cards that can't run this pipeline

Every one of these fails on the same kind of wall, a stage that will not fit in VRAM.

AMD Radeon RX 7600, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
GeForce RTX 4060, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
Intel Arc A750, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce GTX 1070 Ti, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce GTX 1080, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce GTX 1660, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce GTX 1660 Super, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce GTX 1660 Ti, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2060, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2060 Super, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2070, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2070 SUPER, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2080 Founders Edition, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2080 Super, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 2080 Ti Founders Edition, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 3050, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 3060 Ti, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 3070 Founders Edition, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 3070 Ti, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 3080, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 5050, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA GeForce RTX 5060, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)
NVIDIA RTX A2000, blocked at "Read and annotate ~60k tokens of code" (Qwen2.5-Coder 14B)

How these numbers are built

Each stage time is the quantity of work divided by that card's measured throughput for that model, from our own bench. The pipeline is assumed to run batched, every image, then every clip, so each model loads once. Model load time is added where we recorded it; our text-generation runs don't carry a load measurement yet, so pipelines with a language-model stage are slightly optimistic. Nothing here is a single timed run of the whole pipeline, and we don't present it as one.