GPU HUNTER/v0.4.1
BrowseCompareCalculatorBlog
⌘K
Find your GPU
GPU HUNTER

Independent benchmarks for local AI inference. Built for engineers who run models on their own metal.

Last sync · 2h agoAPI operational
Hardware
  • All GPUs
  • Workstation
  • Consumer
  • Apple Silicon
Tools
  • Compare
  • Calculator
  • Model Fit
Resources
  • Blog
  • llms.txt
© 2026 GPU HUNTER · Not affiliated with NVIDIA, AMD, or AppleSome links are affiliate links. We may earn a commission at no extra cost to you.build a3f4c2 · 2026.04.30
browse/nvidia/rtx-5070
R5
NVIDIAConsumerEntry Blackwell

GeForce RTX 5070

Blackwell · TSMC 4NP · released 2025-03

Entry-level Blackwell. 12GB limits model size but fine for 7B-14B models and Stable Diffusion.

VRAM
12 GB
Bandwidth
672 GB/s
TDP
250 W
8B Q4
65 t/s
Score
74 /100
Current price
$549
Buy on Amazon Newegg
price tracker
Price tracking coming soon
Coming soon
Affiliate link — we may earn a commission
01  //  Inference benchmarks

Single-stream decode · llama.cpp

Llama 8B · Q4_K_M
65 t/s
Llama 8B · Q8_0
42 t/s
Llama 8B · FP16
23 t/s
# env llama.cpp b4732 · 4096 ctx · batch=1 · prompt=512 · temp=0.0 · median of 5 runs
02  //  Hardware specs
ArchitectureBlackwell
Process nodeTSMC 4NP
Memory12 GB
Memory bandwidth672 GB/s
FP16 compute30.9 TFLOPS
INT8 compute62 TOPS
TDP250 W
PCIeGen 5 x16
Form factorDual-slot
CoolingAxial
03  //  Model fit

Approximate VRAM required to load weights + 4096 ctx KV cache.

Qwen3 32B
128k ctx
Q4
19 GB
NO
Q8
36 GB
NO
FP16
64 GB
NO
Qwen3 72B
128k ctx
Q4
42 GB
NO
Q8
78 GB
NO
FP16
144 GB
NO
Qwen3 235B
128k ctx
Q4
132 GB
NO
Q8
240 GB
NO
FP16
470 GB
NO
Llama 3.3 70B
128k ctx
Q4
40 GB
NO
Q8
75 GB
NO
FP16
140 GB
NO
DeepSeek V3
128k ctx
Q4
380 GB
NO
Q8
700 GB
NO
FP16
1300 GB
NO
Llama 3.1 8B
128k ctx
Q4
5 GB
FITS
Q8
9 GB
FITS
FP16
16 GB
NO
Qwen3 14B
128k ctx
Q4
8 GB
FITS
Q8
15 GB
NO
FP16
28 GB
NO
Mistral 7B
32k ctx
Q4
4 GB
FITS
Q8
8 GB
FITS
FP16
14 GB
NO
Gemma 2 27B
8k ctx
Q4
16 GB
NO
Q8
30 GB
NO
FP16
54 GB
NO
Codestral 22B
32k ctx
Q4
13 GB
NO
Q8
24 GB
NO
FP16
44 GB
NO
+ STRENGTHS
  • ✓12GB VRAM is enough for 32B-class models at Q4
  • ✓672 GB/s memory bandwidth · top tier in its class
  • ✓Strong tooling: FP16, FP8, Q8, Q4 all officially supported
− TRADE-OFFS
  • −Draws 250W under load — plan PSU and thermals accordingly
  • −Limited to dual-slot chassis
  • −Driver lock-in to vendor stack
04  //  You may also be considering
Open compare
RP6
RTX PRO 6000 Blackwell
96GB · $8,499
vs
R5
GeForce RTX 5090
32GB · $1,999
vs
R4
GeForce RTX 4090
24GB · $1,799
vs