NV NVIDIA SME Discrete GPU

NVIDIA A100 PCIe 80GB

Runs downloadable LLMs up to ~123.5B at Q4, ~315.3-375.8 tok/s on an 8B model.

Other configs: 80GB
NVIDIA A100 PCIe 80GB
80GB
VRAM
1935
GB/s bandwidth
123.5B
Max model (Q4)
$12,000
MSRP

What you can run

At 8k context, FP16 KV. Full-speed tok/s holds up to "max ctx"; beyond it the context spills to system RAM (CPU/PCIe offload) and drops to the degraded figure.

Model sizeFP16Q8Q4
3B 235.4-280.6 t/s
≤1009k ctx · then ~4.9-5.8 (slow)
470.9-561.2 t/s
≤1054k ctx · then ~9.7-11.6 (slow)
840.8-1002.1 t/s
≤1075k ctx · then ~17.4-20.7 (slow)
7-8B 88.3-105.2 t/s
≤428k ctx · then ~1.8-2.2 (slow)
176.6-210.4 t/s
≤489k ctx · then ~3.7-4.4 (slow)
315.3-375.8 t/s
≤516k ctx · then ~6.5-7.8 (slow)
13B 54.3-64.7 t/s
≤281k ctx · then ~1.1-1.3 (slow)
108.7-129.5 t/s
≤361k ctx · then ~2.2-2.7 (slow)
194-231.2 t/s
≤396k ctx · then ~4-4.8 (slow)
32B 22.1-26.3 t/s
≤31k ctx · then ~0.5-0.5 (slow)
44.1-52.6 t/s
≤153k ctx · then ~0.9-1.1 (slow)
78.8-93.9 t/s
≤207k ctx · then ~1.6-1.9 (slow)
70B won't run 20.2-24 t/s
≤6k ctx · then ~0.4-0.5 (slow)
36-42.9 t/s
≤100k ctx · then ~0.7-0.9 (slow)
120B won't run won't run 21-25.1 t/s
≤15k ctx · then ~0.4-0.5 (slow)
400B won't run won't run won't run

Estimate: tok/s ≈ bandwidth ÷ (params × bytes/param) × efficiency (NVIDIA). Cyan = fits ~73.6 GB usable. Beyond "ctx", KV spills to system RAM (CPU/PCIe offload) (rough degraded speed shown).

What else you need for a full build

Minimum supporting components for a desktop built around the NVIDIA A100 PCIe 80GB. Guidance — exact needs depend on your case, other parts and how much you offload to system RAM.

Power supply
≥ 700W · 80+ Gold
Headroom over the 300W card plus the rest of the build.
System RAM
≥ 128 GB · 256 GB recommended
At least the 80 GB of VRAM; 2× lets you CPU-offload models larger than VRAM.
Motherboard
1× free PCIe x16 slot (4.0/5.0)
A full-length x16 electrical slot; PCIe 4.0 is ample, 5.0 future-proofs.
CPU
Any modern multi-core
The GPU runs inference; more cores and memory channels only matter when offloading to RAM.
Storage
≥ 1 TB NVMe SSD
Weights are large — a 70B model at Q4 is ~40 GB; budget for a few.
PSU connector
12V-2×6 / 12VHPWR (or 3× 8-pin)
High-power NVIDIA cards use the 16-pin 12V-2×6 power connector.

Run popular models on the NVIDIA A100 PCIe 80GB

Videos about the NVIDIA A100 PCIe 80GB

NVIDIA A100 80GB vLLM Benchmark: Testing Hugging Face's Top Models at 50 & 300 Concurrent Requests

Bechmarking LLMs on Ollama with Nvidia A100 40GB GPU

More hardware like the NVIDIA A100 PCIe 80GB

About the NVIDIA A100 PCIe 80GB

NVIDIA A100 PCIe 80GB is a discrete GPU from NVIDIA on Ampere, 2020. For local genAI the numbers that matter are its 80 GB (how big a model + context fits) and 1935 GB/s bandwidth (generation speed). Plan for a ~700W power supply.

Frequently asked questions

What size LLM can the NVIDIA A100 PCIe 80GB run?

With ~73.6 GB usable it runs up to ~123.5B at Q4 (~34.5B at FP16) at 8k context. Bigger spills into system RAM (CPU/PCIe offload) and slows sharply.

How many tokens/sec does the NVIDIA A100 PCIe 80GB generate?

Generation is bandwidth-bound. At 1935 GB/s an 8B model at Q4 runs ~315.3-375.8 tok/s (estimate).

Why does long context slow it down?

The KV cache for the context must also fit fast memory; once weights+KV exceed 73.6 GB it spills to system RAM (CPU/PCIe offload) and throughput collapses. Each row shows the max usable context.

What else do I need to run the NVIDIA A100 PCIe 80GB?

Besides the card: a ≥700W 80+ Gold power supply, at least 80GB of system RAM (2× the 80GB VRAM is ideal for CPU offload), a free PCIe x16 slot, any modern multi-core CPU, and a ≥1TB NVMe SSD for model weights.