Hardware Board

Rank the rigs.

64 hardware tags from 420 tagged runs, each ranked by the best score anyone has posted on that rig. Same testpack everywhere — the hardware is the variable. Sort by value to see score per dollar.

Your rig missing? Run the CLI with --hardware-tag and put it on the board.

64 rigs · sorted by best score (high → low) · pick any two rigs to compare

#	Hardware	Best PipelineScore ▾	Best model on this rig	Value	Est. price	Runs	Latency	Tier
1	m5-max-48gb	93.2	openai/gpt-oss-20blocal	—	—	2	1,082ms	TRUNK
2	lab-b200-192gb	89.5	DeepSeek R1 671B-A37Bdeepseek	0.9	$38,000	2	893ms	MAINLINE
3	m2-ultra-192gb	88.8	Qwen 3 235B-A22B MoEalibaba	5.9	$5,600	7	2,422ms	MAINLINE
4	b200-192gb	88.1	DeepSeek V3 671B-A37Bdeepseek	0.9	$38,000	10	892ms	MAINLINE
5	rtx-5080-16gb	87.5	qwen3.6-35b-a3blocal	—	—	4	10,386ms	MAINLINE
6	lab-h200-141gb	87.3	DeepSeek Coder V2 236Bdeepseek	1.0	$32,000	4	876ms	MAINLINE
7	dgx-h100	86.0	DeepSeek R1 671B-A37Bdeepseek	0.1	$270,000	6	902ms	MAINLINE
8	a100-80gb	85.3	DeepSeek Coder V2 236Bdeepseek	2.3	$14,000	14	876ms	MAINLINE
9	h200-141gb	85.1	Qwen 2.5 72B Instructalibaba	1.0	$32,000	7	896ms	MAINLINE
10	lab-dgx-h100	85.0	Qwen 3 72B Instructalibaba	0.1	$270,000	6	889ms	MAINLINE
11	h100-80gb	84.8	Hermes 3 Llama 3.1 405Bnous	1.2	$27,000	7	875ms	MAINLINE
12	cloud-api	84.7	DeepSeek Coder V2 236Bdeepseek	—	—	10	864ms	MAINLINE
13	2x-rtx-4090	84.5	DeepSeek Coder V2 236Bdeepseek	8.7	$3,600	7	3,010ms	MAINLINE
14	rtx-3080-10gb	84.0	DeepSeek V4deepseek	69.5	$450	10	817ms	MAINLINE
15	m3-max-64gb	84.0	DeepSeek R1 Distill Qwen 32Bdeepseek	9.8	$3,200	8	1,498ms	MAINLINE
16	4x-rtx-3090	83.8	Qwen 3 72B Instructalibaba	8.7	$3,600	5	3,009ms	MAINLINE
17	lab-baseline	83.7	DeepSeek V4deepseek	—	—	5	805ms	MAINLINE
18	m3-ultra-96gb	83.7	mlx-community/Qwen3.6-35B-A3B-4bitlocal	—	—	1	13,463ms	MAINLINE
19	m3-max-128gb	83.3	DeepSeek V4deepseek	6.6	$4,700	12	1,156ms	MAINLINE
20	lab-m3-ultra-256gb	83.1	Hermes 3 Llama 3.1 405Bnous	4.1	$7,500	2	3,008ms	MAINLINE
21	m4-pro-48gb	83.1	DeepSeek V4deepseek	12.9	$2,400	4	783ms	MAINLINE
22	rtx-4090-24gb	82.5	DeepSeek V4deepseek	17.1	$1,800	16	1,154ms	MAINLINE
23	lab-m3-max-64gb	82.4	Qwen 3 32B Instructalibaba	9.6	$3,200	3	1,550ms	MAINLINE
24	m3-ultra-256gb	82.3	Qwen 3 235B-A22B MoEalibaba	4.1	$7,500	8	2,924ms	MAINLINE
25	rtx-4080-16gb-offload	82.3	DeepSeek R1 Distill Qwen 32Bdeepseek	32.3	$950	7	1,543ms	MAINLINE
26	rtx-3090-24gb	81.3	Qwen 3.6 72Balibaba	33.7	$900	10	1,286ms	MAINLINE
27	lab-m2-ultra-192gb	81.2	Llama 4 70B Instructmeta	5.4	$5,600	4	2,959ms	MAINLINE
28	lab-a100-40gb	81.2	DeepSeek R1 Distill Qwen 32Bdeepseek	4.0	$7,500	4	617ms	MAINLINE
29	rtx-3070-8gb	80.9	gemma4:12b-it-qat_gpulocal	—	—	1	10,202ms	MAINLINE
30	ryzen-7950x-cpu-only	80.5	Qwen 3.6 72Balibaba	37.5	$800	23	692ms	MAINLINE
31	lab-h100-80gb	79.9	Qwen 2.5 72B Instructalibaba	1.1	$27,000	2	897ms	MAINLINE
32	rtx-4080-16gb	79.9	Qwen 3 14B Instructalibaba	31.3	$950	14	796ms	MAINLINE
33	ryzen-5950x-rtx-3060	79.6	Qwen 3.6 72Balibaba	37.1	$800	3	770ms	MAINLINE
34	a100-40gb	79.3	Devstral Small 24Bmistral	3.9	$7,500	4	619ms	MAINLINE
35	macbook-pro-m3-pro-18gb	79.0	Qwen 3 14B Instructalibaba	18.4	$1,600	17	606ms	MAINLINE
36	lab-4x-rtx-3090	78.8	Qwen 2.5 VL 72Balibaba	8.2	$3,600	2	3,011ms	MAINLINE
37	lab-rtx-3090-24gb	77.9	Codestral 22Bmistral	32.2	$900	3	1,473ms	MAINLINE
38	lab-ryzen-7950x-cpu-only	76.9	Qwen 2.5 14B Instructalibaba	35.8	$800	9	686ms	MAINLINE
39	a6000-48gb	76.9	DeepSeek Coder V2 16Bdeepseek	6.8	$4,200	4	1,501ms	MAINLINE
40	lab-a6000-48gb	76.7	Devstral Small 24Bmistral	6.8	$4,200	2	1,543ms	MAINLINE
41	m2-pro-16gb	75.7	Gemma 3 12B ITgoogle	21.7	$1,300	15	798ms	MAINLINE
42	lab-macbook-pro-m3-pro-18gb	75.3	DeepSeek R1 Distill Llama 8Bdeepseek	17.5	$1,600	13	665ms	MAINLINE
43	rtx-3060-12gb	75.2	Qwen 2.5 14B Instructalibaba	100.0	$280	25	619ms	MAINLINE
44	rtx-4070-12gb	74.7	Qwen 2.5 14B Instructalibaba	53.5	$520	17	815ms	FEEDER
45	lab-rtx-3080-10gb	74.2	Qwen 3 14B Instructalibaba	61.4	$450	5	775ms	FEEDER
46	ryzen-7950x-rtx-3090	73.4	Codestral 22Bmistral	16.1	$1,700	4	1,465ms	FEEDER
47	lab-m3-max-128gb	73.4	Gemma 3 27B ITgoogle	5.8	$4,700	2	1,533ms	FEEDER
48	m3-pro-36gb	73.2	Gemma 3 12B ITgoogle	13.6	$2,000	17	803ms	FEEDER
49	lab-m2-pro-16gb	72.6	Qwen 3 8B Instructalibaba	20.8	$1,300	4	780ms	FEEDER
50	lab-rtx-4080-16gb	72.2	Gemma 3 12B ITgoogle	28.3	$950	2	779ms	FEEDER
51	lab-rtx-4080-16gb-offload	70.0	Aya 23 35Bcohere	27.4	$950	1	1,528ms	FEEDER
52	lab-a100-80gb	68.7	Mixtral 8x7B Instructmistral	1.8	$14,000	1	903ms	FEEDER
53	lab-ryzen-7950x-rtx-3090	68.4	Phi 3.5 MoE 42Bmicrosoft	15.0	$1,700	1	1,411ms	FEEDER
54	lab-rtx-4070-12gb	68.4	Yi Coder 9Byi	49.0	$520	2	798ms	FEEDER
55	lab-m3-pro-36gb	68.3	Llama 4 8B Instructmeta	12.7	$2,000	3	807ms	FEEDER
56	lab-rtx-3060-12gb	66.9	InternLM 2.5 7B Chatinternlm	89.0	$280	6	660ms	FEEDER
57	lab-snapdragon-x-elite-32gb	64.6	Qwen 3 4B Instructalibaba	21.9	$1,100	1	357ms	FEEDER
58	m2-air-16gb	59.5	Qwen 2.5 3B Instructalibaba	23.3	$950	5	343ms	TAP
59	m1-air-8gb	58.8	Gemma 3 4B ITgoogle	31.3	$700	9	351ms	TAP
60	lab-m3-air-16gb	55.1	Qwen 2.5 3B Instructalibaba	18.7	$1,100	3	349ms	TAP
61	m3-air-16gb	52.9	Phi 3 Mini 3.8Bmicrosoft	17.9	$1,100	7	347ms	TAP
62	snapdragon-x-elite-32gb	52.4	Phi 3 Mini 3.8Bmicrosoft	17.7	$1,100	4	354ms	TAP
63	lab-m2-air-16gb	49.3	Granite 3.1 2B Instructibm	19.3	$950	3	352ms	TAP
64	lab-m1-air-8gb	28.9	TinyLlama 1.1B Chatcommunity	15.4	$700	1	356ms	DRIP

Click a hardware tag to see every run on that rig; tick two rigs to go head-to-head. A rig's rank reflects its best showing, so it rewards the best model the hardware can hold. Value is score per dollar (best rig = 100) using hand-maintained street-price approximations in USD; cloud and unpriced rigs are excluded from the value ranking.

← Model board Users board →