AI Hardware Analysis

Select your existing hardware to see which models you can run and for how many concurrent users. Use workload assumptions to refine.

Your hardware

GPU / device

Number of GPUs

Total VRAM: 8 GB

Workload assumptions

Target Params (B)

13B

Target Quant (Bits)

Desired Speed (T/s)

Concurrent Users

Context Window

RAG Enabled?

Agent Complexity (M)

With your hardware

RTX 3050 8GB — 8 GB VRAM

Model	Max concurrent users	Est. speed (T/s)
1.5B / 4-bit	3	~179
1.5B / 8-bit	3	~90
3B / 4-bit	3	~90
3B / 8-bit	2	~45
4B / 4-bit	2	~67
4B / 8-bit	1	~34
7B / 4-bit	1	~38
8B / 4-bit	1	~34

Requirements

Req. VRAM (GB)

28.3

Req. Bandwidth

2357 GB/s

Req. System RAM

21 GB

Best value hardware

Est. GPU / device cost

$12,000

Best value hardware

AMD MI300X 192GBgpu

Qty Needed

Cluster VRAM

192 GB

Hardware Name

AMD MI300X 192GB

Meets desired speed (~495 T/s)

15%VRAM used

Model fits in GPU

28.3 GB required / 192 GB available

Get a certified hardware + license quote

VRAM_weights = (P×Q/8)×1.2; VRAM_context = U×C×M×0.0005; B_req = T_target×(VRAM_weights+VRAM_context)/0.6. Max users from available VRAM. Actual results may vary.