Real GPUs. Real benchmark coverage. Real next clicks. This page now surfaces your actual AI hardware catalog and sends every serious card into its own benchmark detail page.
GPUs surfaced
41
Consumer, pro, and datacenter cards
Benchmark records
354
Linked back to methodology-aware detail pages
Average rating
8.6
Composite scoring from throughput, trust, and coverage
Top consumer pick
5090 32GB
Editorial + benchmark-backed
Fresh from your rig
429.62 tok/s on alibayram/kumru:latest / 160.83 tok/s on brooqs/mistral-turkish-v2:latest. Captured on 2026-05-28 with ollama-local-api.
70B Q4 full-GPU at 40-55 tok/s with 8-16K context — the first consumer card where 70B daily-driving is practical
MyAI rating
9.3
Reference
32 GB
575 W
720 tok/s
$2.0k
Benchmark-backed GPU profile with hardware detail, trust signals, and buying path.
MyAI rating
8.9
Flagship
48 GB
300 W
125 tok/s
$6.8k
Benchmark-backed GPU profile with hardware detail, trust signals, and buying path.
MyAI rating
9.5
Mythic
192 GB
1000 W
520 tok/s
$40k
Quick picks
Best local 70B path
Prioritize VRAM headroom and clean 32B-to-70B scaling.
Best value class
The used-market sweet spot still matters for serious builders.
Best scale-up card
When this turns into serving, capacity and interconnect win.
Top LLM throughput
8× NVIDIA H100 SXM5 80GB
Datacenter GPU / Llama 3 8B FP16
AMD Instinct MI300X 192GB
Datacenter GPU / Llama 3 8B Q4
NVIDIA GeForce RTX 5090 32GB
Consumer GPU / Llama 3 8B Q4
NVIDIA H100 SXM5 80GB
Datacenter GPU / Phi-3 Mini
NVIDIA H200 141GB
Datacenter GPU / Llama 3 70B Q4
NVIDIA GeForce RTX 4090 24GB
Consumer GPU / Llama 3 8B Q4
Browse hardwares
Search by name, slice by class, then jump straight into the hardware brief. These cards are built from your benchmark records, not a static mock list.
Showing 41 GPUs / All classes / all brands.
Rating
9.5
192 GB
1000 W
520 tok/s
$40k
Top benchmark
Embedding throughput / 23500 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.4
80 GB
700 W
612 tok/s
$25k
Top benchmark
Embedding throughput / 12400 emb/s
The default datacenter inference GPU through 2025. Right answer for production multi-tenant serving. Not for homelab — this is enterprise/cloud territory.
Rating
9.4
141 GB
700 W
540 tok/s
$30k
Top benchmark
Embedding throughput / 15200 emb/s
The capacity king. Right answer when 80GB isn't enough and you need 141GB. For everything else, H100 or L40S are more economical.
Rating
9.3
640 GB
5600 W
1980 tok/s
$200k
Top benchmark
Llama 3 8B FP16 / 1980 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.3
192 GB
750 W
920 tok/s
$15k
Top benchmark
Embedding throughput / 9800 emb/s
Best for organizations with existing AMD infrastructure or those specifically chasing $/VRAM ratios at datacenter scale. The software maturity gap vs NVIDIA is real but narrowing.
Rating
9.3
32 GB
575 W
720 tok/s
$2.0k
Top benchmark
Embedding throughput / 12200 emb/s
Buy if 70B Q4 full-GPU is your daily target and you can find one near MSRP. Skip if 4090 covers your model range or if the scalper premium exceeds 20%.
Rating
9.2
192 GB
750 W
248 tok/s
$15k
Top benchmark
Embedding throughput / 14800 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.1
24 GB
450 W
540 tok/s
$1.6k
Top benchmark
Embedding throughput / 8500 emb/s
Buy if you want the best single-card experience for 32B-class models and either bought at MSRP or found a clean used unit. Skip if 70B is your daily target or if 16GB cards cover your needs.
Rating
9.1
16 GB
360 W
165 tok/s
$999
Top benchmark
Embedding throughput / 8200 emb/s
Buy for mixed gaming+AI or as a secondary GPU in multi-card rigs. Skip if your primary use case is local AI — used 3090 gives you 24GB at lower cost, and 5070 Ti 16GB gives similar VRAM for $250 less.
Your PC result
429.62 tok/s on kumru-2b
160.83 tok/s on brooqs-mistral-turkish-v2-latest / Batch=1, 2048 context, 512 generated tokens, consecutive warm-model runs. Generation TPS uses Ollama eval_count/eval_duration.
Rating
9.1
128 GB
240 W
145 tok/s
$3.0k
Top benchmark
DeepSeek-R1 7B / 145 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.1
16 GB
300 W
128 tok/s
$749
Top benchmark
Embedding throughput / 7100 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.0
16 GB
320 W
215 tok/s
$999
Top benchmark
Embedding throughput / 6900 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.0
80 GB
400 W
215 tok/s
$15k
Top benchmark
Embedding throughput / 9500 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
9.0
48 GB
355 W
195 tok/s
$999
Top benchmark
Embedding throughput / 5800 emb/s
Buy for Linux AI builds where $/VRAM is the priority and you're comfortable with the ROCm ecosystem. Skip for Windows, for production, or if you value plug-and-play over tinkering.
Rating
9.0
48 GB
350 W
175 tok/s
$7.8k
Top benchmark
Embedding throughput / 10500 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.9
144 GB
1000 W
380 tok/s
$44k
Top benchmark
Llama 3 8B FP16 / 380 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.9
48 GB
300 W
125 tok/s
$6.8k
Top benchmark
Embedding throughput / 8200 emb/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.7
12 GB
250 W
165 tok/s
$549
Top benchmark
Phi-3 Mini / 165 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.6
8 GB
180 W
158 tok/s
$379
Top benchmark
Phi-3 Mini / 158 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.6
24 GB
350 W
88.0 tok/s
$1.5k
Top benchmark
Embedding throughput / 4900 emb/s
Buy used if you can find a clean unit under $700. The value-for-VRAM proposition is unmatched — nothing else gives you 24GB at this price. Pair with a second used 3090 for 48GB combined VRAM at ~$1,400 total.
Rating
8.5
40 GB
400 W
195 tok/s
$9.0k
Top benchmark
Llama 3 8B FP16 / 195 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.5
128 GB
560 W
188 tok/s
$12k
Top benchmark
Llama 3 8B FP16 / 188 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.5
12 GB
200 W
138 tok/s
$599
Top benchmark
Phi-3 Mini / 138 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.5
48 GB
900 W
138 tok/s
$3.2k
Top benchmark
Gemma 2 9B / 138 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.5
16 GB
180 W
132 tok/s
$449
Top benchmark
Phi-3 Mini / 132 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.5
12 GB
190 W
86.0 tok/s
$249
Top benchmark
Phi-3 Mini / 86.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.4
64 GB
300 W
175 tok/s
$7.5k
Top benchmark
Phi-3 Mini / 175 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.4
12 GB
285 W
165 tok/s
$799
Top benchmark
Phi-3 Mini / 165 tok/s
Buy as a gaming-primary, AI-secondary card. For pure AI, the 12GB ceiling is limiting. Consider used 3090 (24GB) or 4060 Ti 16GB instead.
Rating
8.3
16 GB
165 W
56.0 tok/s
$499
Top benchmark
Mistral 7B / 56.0 tok/s
Buy for budget AI builds where 16GB VRAM capacity matters more than token generation speed. Best new-card value for local AI experimentation. Skip if you can stretch to a used 3090 at $600-700.
Rating
8.2
48 GB
300 W
162 tok/s
$5.5k
Top benchmark
Phi-3 Mini / 162 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.2
96 GB
1400 W
78.0 tok/s
$4.0k
Top benchmark
Qwen 2.5 14B / 78.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.1
8 GB
145 W
145 tok/s
$299
Top benchmark
Phi-3 Mini / 145 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.1
16 GB
263 W
132 tok/s
$499
Top benchmark
Phi-3 Mini / 132 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
8.0
24 GB
230 W
122 tok/s
$2.2k
Top benchmark
Phi-3 Mini / 122 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.9
12 GB
170 W
92.0 tok/s
$329
Top benchmark
Phi-3 Mini / 92.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.9
16 GB
140 W
92.0 tok/s
$1.0k
Top benchmark
Phi-3 Mini / 92.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.9
8 GB
165 W
88.0 tok/s
$269
Top benchmark
Phi-3 Mini / 88.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.8
8 GB
115 W
95.0 tok/s
$299
Top benchmark
Phi-3 Mini / 95.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.8
16 GB
225 W
72.0 tok/s
$329
Top benchmark
Phi-3 Mini / 72.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.8
12 GB
245 W
58.0 tok/s
$449
Top benchmark
Mistral 7B / 58.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.
Rating
7.6
8 GB
225 W
58.0 tok/s
$249
Top benchmark
Phi-3 Mini / 58.0 tok/s
Open the hardware brief for methodology, confidence ranges, and buyer guidance.