FitMyLLM
NVIDIA· Volta

NVIDIA Tesla PG500-216

Running LLMs on the Tesla PG500-216 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
32 GB
High-end
Bandwidth
1130
GB/s
Models Q4
362/498
73%
7B Q4 speed
~129
Blazing
Model coverage at Q473% of all
Estimated speed· By model size at Q4

Average speeds at Q4 quantization. Actual performance varies by model architecture and context length.

3B
~301
tok/s
7B
~129
tok/s
14B
~65
tok/s
32B
~28
tok/s
70B
—
39.4GB needed
Measured rig reports

We rent the machine and time every model on it: decode, VRAM peak, concurrency, watts and cost per million tokens. Yours may already be one of them — and two are free to read in full.

See the reports →
Specifications
VRAM
32 GB
Bandwidth
1130 GB/s
FP16 Compute
28.3 TFLOPS
TDP
250W
Memory
HBM2
Architecture
Volta
CUDA Cores
5,120
Tensor Cores
640
PCIe
Gen 3 x16
325
Fast models · >30 tok/s
Real-time chat speed
362
Usable · >10 tok/s
Comfortable for all tasks
362
Total compatible
Fit in VRAM at Q4
Rent it in the cloud

Buying Tesla PG500-216 costs $15–$40k and isn’t practical for most teams. Spin one up by the hour instead:

Spin up in ~60s. Pay by the second. Cancel anytime.

Some links are affiliate links — we may earn a small commission at no extra cost to you. This helps keep FitMyLLM free and independent.

Compatible models· 362
S
Falcon-H1R Tiny 90M0.09B
Falcon·256K ctx· chat· reasoning
10044
tok/s · 2% VRAM
›
S
SmolLM2 135M0.135B
Smollm·2K ctx· chat
6696
tok/s · 2% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
Embedding·8K ctx· chat
6457
tok/s · 2% VRAM
›
S
GPT-2 124M0.14B
Gpt2·1K ctx· chat
6457
tok/s · 2% VRAM
›
S
SmolVLM 256M0.256B
Smollm·8K ctx· chat· vision
3531
tok/s · 2% VRAM
›
S
Gemma 3 270M0.27B
Gemma·32K ctx· chat
3348
tok/s · 2% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
Embedding·8K ctx· embedding· multilingual
2964
tok/s · 2% VRAM
›
S
bge-large-en-v1.5 335M0.335B
Embedding·1K ctx· chat
2699
tok/s · 2% VRAM
›
S
mxbai-embed-large-v10.335B
Embedding·1K ctx· embedding
2699
tok/s · 2% VRAM
›
S
Snowflake Arctic Embed L0.335B
Embedding·1K ctx· embedding
2699
tok/s · 2% VRAM
›
S
LFM2 350M0.35B
Lfm·125K ctx· chat· tool_use
2583
tok/s · 2% VRAM
›
S
SmolLM2 360M0.36B
Smollm·8K ctx· chat
2511
tok/s · 2% VRAM
›
S
GPT-2 Medium 345M0.38B
Gpt2·1K ctx· chat
2379
tok/s · 2% VRAM
›
S
Qwen 2.5 0.5B0.5B
Qwen·32K ctx· chat
1808
tok/s · 2% VRAM
›
S
SmolVLM 500M0.5B
Smollm·8K ctx· chat· vision
1808
tok/s · 2% VRAM
›
S
Falcon-H1 0.5B0.52B
Falcon·128K ctx· chat
1738
tok/s · 3% VRAM
›
S
BGE-M30.568B
Embedding·8K ctx· embedding
1592
tok/s · 3% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
Embedding·8K ctx· embedding· multilingual
1592
tok/s · 3% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
Embedding·32K ctx· embedding· multilingual
1507
tok/s · 3% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
Falcon·32K ctx· chat· reasoning
1507
tok/s · 3% VRAM
›
S
Falcon Perception 0.6B0.6B
Falcon·4K ctx· vision
1507
tok/s · 3% VRAM
›
S
Qwen 1.5 0.5B0.62B
Qwen·32K ctx· chat
1458
tok/s · 3% VRAM
›
S
LFM2 700M0.74B
Lfm·125K ctx· chat· tool_use
1222
tok/s · 3% VRAM
›
S
Qwen3 0.6B0.75B
Qwen·32K ctx· chat· reasoning
1205
tok/s · 3% VRAM
›
S
GPT-2 Large 774M0.81B
Gpt2·1K ctx· chat
1116
tok/s · 3% VRAM
›
S
Qwen 3.5 0.8B0.87B
Qwen·256K ctx· chat· coding· multilingual
1039
tok/s · 3% VRAM
›
S
InternVL3 1B0.94B
Other·32K ctx· chat· vision
962
tok/s · 3% VRAM
›
S
MiniCPM5 1B1.08B
Minicpm·128K ctx· chat· reasoning· multilingual
837
tok/s · 4% VRAM
›
S
K2 Horizon 0.9B1.08B
Other·128K ctx· chat· coding· reasoning
837
tok/s · 4% VRAM
›
S
TinyLlama 1.1B1.1B
Llama·2K ctx· chat
822
tok/s · 4% VRAM
›
S
LFM2.5-1.2B-Thinking1.2B
Lfm·122K ctx· chat· reasoning· tool_use
753
tok/s · 4% VRAM
›
S
Llama-3.2-1B1.2B
Llama·4K ctx· chat
753
tok/s · 4% VRAM
›
S
LFM2 1.2B1.2B
Lfm·125K ctx· chat· tool_use· multilingual
753
tok/s · 4% VRAM
›
S
Zamba2 1.2B1.2B
Other·4K ctx· chat
753
tok/s · 4% VRAM
›
S
Ling 3.0 Tiny7.89BMoE
Other·128K ctx· chat· coding· reasoning
695
tok/s · 17% VRAM
›
S
EXAONE-4.0-1.2B1.3B
Exaone·64K ctx· chat
695
tok/s · 4% VRAM
›
S
OPT 1.3B1.3B
Opt·2K ctx· chat
695
tok/s · 4% VRAM
›
S
MiniCPM-V 4.61.3B
Other·256K ctx· chat· vision
695
tok/s · 4% VRAM
›
S
DeepSeek Coder 1.3B1.35B
Deepseek·16K ctx· coding
670
tok/s · 4% VRAM
›
S
Phi-1 1.3B1.42B
Phi·2K ctx· coding
637
tok/s · 4% VRAM
›
Next step

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Tesla PG500-216.

What this card is worth

Tesla PG500-216 holds 362 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is Nemotron-H 47B at IQ4_XS.

Tokens/sec per $100
—
8B at Q4_K_M, so cards compare like for like
VRAM per $100
—
what memory costs on this card
The biggest it takes
Nemotron-H 47B47B · IQ4_XS28.8 GB18 tok/sest
Phi-3.5 MoE 42B41.9B · Q4_K_M28.6 GB118 tok/sest
Falcon 40B40B · Q4_K_M27.4 GB19 tok/sest
Flint39.79B · Q4_K_M27.0 GB312 tok/sest
InternVL3 38B38B · Q4_K_M26.7 GB21 tok/sest
K2 Horizon MoVA 36B-A4B37.44B · Q4_K_M26.1 GB195 tok/sest
Seed-OSS 36B Instruct36B · Q5_K_S28.6 GB19 tok/sest
MiniCPM-V 4.7 35B-A3B35.21B · Q5_K_M27.7 GB223 tok/sest
Against cards of similar memory
M4 Max (48GB)36 GB$2,4992.2 tok/s per $100
M3 Max (48GB)36 GB$2,8991.6 tok/s per $100
M4 Pro (48GB)36 GB$1,7991.9 tok/s per $100
A100 SXM4 40 GB40 GB$10,0001.4 tok/s per $100
Tesla PG500-21632 GB——

Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.

Device under test

NVIDIA Tesla PG500-216 — 32 GB VRAM.

Tesla PG500-216 spec
Brand
NVIDIA
VRAM
32 GB HBM2
Bandwidth
1130 GB/s
FP16 Compute
28.3 TFLOPS
FP32 Compute
14.1 TFLOPS
CUDA Cores
5,120
Tensor Cores
640
TDP
250 W
Architecture
Volta
AI capability
362/ 498 models @ Q4

With 32 GB VRAM and 1130 GB/s bandwidth, this GPU handles models up to 41.9B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~129 tok/s.

Rent it — live prices
Checking what the clouds are charging…
Top models for Tesla PG500-216
362 fit · showing 20
ModelSizeVRAM Q4tok/sAvg
Phi-3.5 MoE 42B41.9B26.1 GB13756.7
Falcon 40B40B24.9 GB2320.9
Flint39.79B24.8 GB36277.3
InternVL3 38B38B23.7 GB2478.9
K2 Horizon MoVA 36B-A4B37.44B23.4 GB22645.7
Seed-OSS 36B Instruct36B22.5 GB2554.4
MiniCPM-V 4.7 35B-A3B35.21B22.0 GB30176.4
c4ai-command-r-v01 35B35B21.9 GB2627.5
Qwen 3.5 35B A3B35B21.9 GB30156.1
Qwen 3.6 35B A3B35B21.9 GB30153.3
K2 Horizon 32B34.78B21.7 GB2654.0
Nous Capybara 34B34.4B21.5 GB2642.0
Yi-1.5 34B34.4B21.5 GB2645.3
Falcon-H1 34B34B21.3 GB2758.6
CodeLlama 34B34B21.3 GB2725.4
Nous Hermes 2 34B34B21.3 GB2747.0
Phind CodeLlama 34B34B21.3 GB2768.1
LLaVA-1.6 Yi 34B34B21.3 GB2747.4
WizardCoder Python 34B34B21.3 GB2773.2
Yi 34B34B21.3 GB2733.4