FitMyLLM
AMD· GCN 5.0

AMD Radeon Pro Vega 48

Running LLMs on the Radeon Pro Vega 48 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what a budget buys

VRAM
8 GB
ENTRY-LEVEL
BANDWIDTH
402
GB/S
MODELS Q4
201/449
45%
7B Q4 SPEED
~51
BLAZING
▸ MODEL COVERAGE @ Q445% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

3B
~119
TOK/S
7B
~51
TOK/S
14B
7.9GB NEEDED
32B
18.0GB NEEDED
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
8 GB
BANDWIDTH
402 GB/s
FP16 COMPUTE
14.8 TFLOPS
TDP
130W
MEMORY
HBM2
ARCHITECTURE
GCN 5.0
STREAM PROCESSORS
3,072
COMPUTE UNITS
48
PCIE
Gen 3 x16
201
FAST MODELS · >30 TOK/S
Real-time chat speed
201
USABLE · >10 TOK/S
Comfortable for all tasks
201
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ DON’T WANT TO BUY?

Test Radeon Pro Vega 48 (or anything bigger) without committing. Pay by the second, cancel anytime.

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· 201
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
3970
TOK/S · 7% VRAM
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
2647
TOK/S · 7% VRAM
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
2552
TOK/S · 7% VRAM
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
2552
TOK/S · 7% VRAM
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
1396
TOK/S · 8% VRAM
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
1323
TOK/S · 8% VRAM
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
1172
TOK/S · 8% VRAM
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
1067
TOK/S · 9% VRAM
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
1067
TOK/S · 9% VRAM
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
1067
TOK/S · 9% VRAM
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
1021
TOK/S · 9% VRAM
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
993
TOK/S · 9% VRAM
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
940
TOK/S · 9% VRAM
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
715
TOK/S · 10% VRAM
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
715
TOK/S · 10% VRAM
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
687
TOK/S · 10% VRAM
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
629
TOK/S · 10% VRAM
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
629
TOK/S · 10% VRAM
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
596
TOK/S · 11% VRAM
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
596
TOK/S · 11% VRAM
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
596
TOK/S · 11% VRAM
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
576
TOK/S · 11% VRAM
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
483
TOK/S · 12% VRAM
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
476
TOK/S · 12% VRAM
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
441
TOK/S · 12% VRAM
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
411
TOK/S · 13% VRAM
S
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
380
TOK/S · 13% VRAM
S
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
331
TOK/S · 14% VRAM
S
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
325
TOK/S · 15% VRAM
S
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
298
TOK/S · 15% VRAM
S
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
298
TOK/S · 15% VRAM
S
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
298
TOK/S · 15% VRAM
S
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
298
TOK/S · 15% VRAM
S
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
275
TOK/S · 16% VRAM
S
OPT 1.3B1.3B
OPT·2K CTX· CHAT
275
TOK/S · 16% VRAM
S
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
275
TOK/S · 16% VRAM
S
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
265
TOK/S · 16% VRAM
S
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
252
TOK/S · 17% VRAM
S
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
252
TOK/S · 17% VRAM
S
LFM2 8B A1B8.3BMoE
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
238
TOK/S · 70% VRAM
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Radeon Pro Vega 48.

WHAT THIS CARD IS WORTH

Radeon Pro Vega 48 holds 201 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is Falcon3-10B 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
Falcon3-10B10.3B · IQ4_XS7.1 GB37 tok/sEST
Bamba 9B v29.78B · Q4_K_M7.2 GB36 tok/sEST
RecurrentGemma 9B9.63B · Q4_K_M6.8 GB36 tok/sEST
glm-4-9b9.4B · Q4_K_M6.6 GB37 tok/sEST
Yi 1.5 9B9B · Q4_K_M6.5 GB38 tok/sEST
Yi Coder 9B9B · Q4_K_M6.5 GB38 tok/sEST
Ministral 3 8B8.92B · Q4_K_M6.6 GB38 tok/sEST
Ministral 3 8B Reasoning8.92B · Q4_K_M6.6 GB38 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
M3 Pro (18GB)12 GB$1,5991.5 tok/s per $100
M1 Pro (16GB)11 GB$9992.9 tok/s per $100
M2 Pro (16GB)11 GB$1,2992.2 tok/s per $100
M4 (16GB)11 GB$4994.0 tok/s per $100
Radeon Pro Vega 488 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

AMD Radeon Pro Vega 48 — 8 GB VRAM.

RADEON PRO VEGA 48 SPEC
BRAND
AMD
VRAM
8 GB HBM2
BANDWIDTH
402 GB/s
FP16 COMPUTE
14.8 TFLOPS
FP32 COMPUTE
7.4 TFLOPS
STREAM PROCESSORS
3,072
TDP
130 W
ARCHITECTURE
GCN 5.0
▸ AI CAPABILITY
201/ 449 models @ Q4

With 8 GB VRAM and 402 GB/s bandwidth, this GPU handles models up to 9.63B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~46 tok/s.

RENT IT — LIVE PRICES
Checking what the clouds are charging…
§ 01TOP MODELS FOR RADEON PRO VEGA 48
201 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
RecurrentGemma 9B9.63B6.4 GB3735.0
glm-4-9b9.4B6.2 GB3820.5
gemma-2-9b9.2B6.1 GB3930.2
Yi 1.5 9B9B6.0 GB4030.3
Yi Coder 9B9B6.0 GB4035.8
Ministral 3 8B8.92B5.9 GB4025.7
Ministral 3 8B Reasoning8.92B5.9 GB40
NVIDIA-Nemotron-Nano-9B-v28.9B5.9 GB4044.2
InternLM3 8B Instruct8.8B5.9 GB4138.7
Gemma 1 7B8.54B5.7 GB4224.7
CodeGemma 7B8.54B5.7 GB4240.2
LFM2 8B A1B8.3B5.6 GB23824.3
Seed-Coder 8B Instruct8.25B5.5 GB4334.1
Seed-Coder 8B Reasoning8.25B5.5 GB4332.9
DeepSeek R1-0528 Qwen3 8B8.2B5.5 GB4436.3
Qwen3-8B8.2B5.5 GB4443.3
Granite 3.0 8B8.17B5.5 GB4436.4
Granite 3.1 8B8.17B5.5 GB4438.6
Command-R7B8.03B5.4 GB4435.3
Aya Expanse 8B8B5.4 GB4527.8