Cloudzat GPU & Local AI Hardware Intelligence
What AI Models Can My GPU Run?
Owning a GPU does not guarantee that every model using the same parameter count will fit. Select the exact GPU, model size, quantization and context target to see a practical fit classification.
Check whether your GPU can run the model
Memory fit and software support are evaluated separately.
Comfortable means headroom remains
A comfortable result leaves room for context growth, runtime workspace and modest desktop activity. It is not a guaranteed performance benchmark.
Usable with limits may need compromises
The model may fit at the selected settings but leave little margin. Reducing context, batch size or background GPU use can improve reliability.
CPU offload can rescue a larger model
Partial offload may allow inference when the full model does not fit in VRAM. Expect lower speed and higher system RAM use.
Software support is separate from memory fit
A card can have enough VRAM but lack mature support in the chosen application. Check CUDA, ROCm, oneAPI or DirectML compatibility before buying.
Current GPU catalogue
Current Amazon offers identified by exact GPU model. Use the calculator above for workload-specific ranking.
| GPU offer | VRAM | Stack | Power | Condition | Price | Price/GB | |
|---|---|---|---|---|---|---|---|
 Intel Arc B580WEELIAO GUNNIR Intel Arc B580 Index 12GB GDDR6 2670MHz Dual Fan Graphics Card (192-bit, PCI-Express 4.0 x8, HDMI/DisplayPort, Supports 4k) |
12GB GDDR6 |
oneAPI / DirectML | 190W 600W PSU |
Used | $319.00 | $26.58 | View on Amazon |
 Intel Arc B580ASRock Intel Arc B580 Challenger 12GB OC & PRO-650G 650W 80+ Gold PSU | Efficient 1440p Gaming Bundle |
12GB GDDR6 |
oneAPI / DirectML | 190W 600W PSU |
New | $349.18 | $29.10 | View on Amazon |
 Intel Arc B580ASRock Intel Arc B580 Challenger 12GB OC & Challenger CL-650G 650W Gold PSU | Modern Gaming & Streaming Bundle |
12GB GDDR6 |
oneAPI / DirectML | 190W 600W PSU |
New | $358.88 | $29.91 | View on Amazon |
 Intel Arc A770 16GBWEELIAO GUNNIR Intel Arc A770 Photon 8GB OC GDDR6 2400MHz Triple Fan White Graphics Card (256-bit, PCIe 4.0, HDMI/DisplayPort, Supports 4k) |
16GB GDDR6 |
oneAPI / DirectML | 225W 600W PSU |
New | $369.00 | $23.06 | View on Amazon |
 Radeon RX 9070GIGABYTE Gaming Radeon RX 9070 GRE 12GB GDDR6 PCI Express 5.0 ATX Graphics Card GV-R907GREGAMING OC-12GD |
16GB GDDR6 |
ROCm / DirectML | 220W 650W PSU |
New | $534.99 | $33.44 | View on Amazon |
 Radeon RX 9070XFX Swift AMD Radeon RX 9070 GRE Triple Fan Gaming Edition with 12GB GDDR6 HDMI 3xDP, AMD RDNA 4 RX-97GRE29B7 |
16GB GDDR6 |
ROCm / DirectML | 220W 650W PSU |
New | $549.99 | $34.37 | View on Amazon |
 Radeon RX 9070PowerColor Reaper AMD Radeon RX 9070 GRE 12GB GDDR6 |
16GB GDDR6 |
ROCm / DirectML | 220W 650W PSU |
New | $549.99 | $34.37 | View on Amazon |
 GeForce RTX 5060 Ti 16GBZOTAC Gaming Geforce RTX 5060 Ti Amp Nvidia 16 Gb Gddr7, W129163508 (Amp Nvidia 16 Gb Gddr7) |
16GB GDDR7 |
CUDA | 180W 600W PSU |
New | $569.99 | $35.62 | View on Amazon |
Frequently asked questions
Can a 24GB GPU run a 70B model?
A heavily quantized 70B model may require more than 24GB once cache and overhead are included, so partial CPU offload or multiple GPUs may be necessary.
Can I use integrated graphics memory?
Some integrated systems share system memory, but performance and software support differ greatly from discrete GPUs.
Why does context length affect fit?
The runtime stores attention cache for the active context. More tokens and more simultaneous sequences require additional memory.