▸ AMD· RDNA 3.0

AMD Radeon PRO W7900

Running LLMs on the Radeon PRO W7900 — the long read: which models fit at which quantisation, and the settings worth changing. · Or what else $3,999 buys

VRAM

48 GB

FLAGSHIP

BANDWIDTH

864

GB/S

MODELS Q4

345/449

77%

7B Q4 SPEED

~110

BLAZING

▸ MODEL COVERAGE @ Q477% OF ALL

▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

~256

TOK/S

~110

TOK/S

14B

~55

TOK/S

32B

~24

TOK/S

70B

~11

TOK/S

▸ 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

48 GB

BANDWIDTH

864 GB/s

FP16 COMPUTE

122.6 TFLOPS

TDP

295W

MEMORY

GDDR6

ARCHITECTURE

RDNA 3.0

STREAM PROCESSORS

6,144

COMPUTE UNITS

PCIE

Gen 4 x16

MSRP

$3,999

297

FAST MODELS · >30 TOK/S

Real-time chat speed

345

USABLE · >10 TOK/S

Comfortable for all tasks

345

TOTAL COMPATIBLE

Fit in VRAM at Q4

▸ RENT IT IN THE CLOUD

Buying Radeon PRO W7900 costs $15–$40k and isn’t practical for most teams. Spin one up by the hour instead:

Rent on Vast.ai →Or RunPod →

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· 345

Falcon-H1R Tiny 90M0.09B

FALCON·256K CTX· CHAT· REASONING

nomic-embed-text-v1.5 100M0.14B

EMBEDDING·8K CTX· CHAT

SMOLLM·8K CTX· CHAT· VISION

Snowflake Arctic Embed M v2.00.305B

EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL

bge-large-en-v1.5 335M0.335B

EMBEDDING·1K CTX· CHAT

mxbai-embed-large-v10.335B

EMBEDDING·1K CTX· EMBEDDING

Snowflake Arctic Embed L0.335B

EMBEDDING·1K CTX· EMBEDDING

LFM·125K CTX· CHAT· TOOL_USE

GPT-2 Medium 345M0.38B

SMOLLM·8K CTX· CHAT· VISION

FALCON·128K CTX· CHAT

EMBEDDING·8K CTX· EMBEDDING

Snowflake Arctic Embed L v2.00.568B

EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL

Qwen3-Embedding 0.6B0.6B

EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL

Falcon-H1R Tiny 0.6B0.6B

FALCON·32K CTX· CHAT· REASONING

Falcon Perception 0.6B0.6B

FALCON·4K CTX· VISION

LFM·125K CTX· CHAT· TOOL_USE

QWEN·32K CTX· CHAT· REASONING

GPT-2 Large 774M0.81B

QWEN·256K CTX· CHAT· CODING· MULTILINGUAL

OTHER·32K CTX· CHAT· VISION

MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL

LFM2.5-1.2B-Thinking1.2B

LFM·122K CTX· CHAT· REASONING· TOOL_USE

LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL

OTHER·256K CTX· CHAT· VISION

DeepSeek Coder 1.3B1.35B

DEEPSEEK·16K CTX· CODING

PHI·2K CTX· CHAT· CODING

LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL

▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Radeon PRO W7900.

▸Find models for Radeon PRO W7900 Compare with other GPUs

WHAT THIS CARD IS WORTH

Radeon PRO W7900 holds 345 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is DeepSeek R1 Distill Llama 70B at IQ4_XS.

TOKENS/SEC PER $100

2.3

8B at Q4_K_M, so cards compare like for like

VRAM PER $100

1.20 GB

what memory costs on this card

THE BIGGEST IT TAKES

DeepSeek R1 Distill Llama 70B70.6B · IQ4_XS43.1 GB14 tok/sEST

Llama 3.3 70B70.6B · IQ4_XS43.1 GB14 tok/sEST

Llama 3.1 70B70.6B · IQ4_XS43.1 GB14 tok/sEST

Llama 3 70B70.6B · IQ4_XS43.1 GB14 tok/sEST

Llama-3.1-Nemotron-70B70.6B · IQ4_XS43.1 GB14 tok/sEST

Cogito 70B70B · IQ4_XS42.8 GB14 tok/sEST

Llama 2 70B70B · IQ4_XS42.8 GB14 tok/sEST

CodeLlama 70B70B · IQ4_XS42.8 GB14 tok/sEST

AGAINST CARDS OF SIMILAR MEMORY

H100 SXM5 64 GB64 GB$25,0000.2 tok/s per $100

Radeon Instinct MI20064 GB$10,0001.8 tok/s per $100

Radeon Instinct MI21064 GB$8,0002.2 tok/s per $100

M1 Ultra (64GB)48 GB$2,4992.9 tok/s per $100

Radeon PRO W790048 GB$3,9992.3 tok/s per $100

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 W7900 — 48 GB VRAM.

▸ RADEON PRO W7900 SPEC

BRAND: AMD
VRAM: 48 GB GDDR6
BANDWIDTH: 864 GB/s
FP16 COMPUTE: 122.6 TFLOPS
FP32 COMPUTE: 61.3 TFLOPS
STREAM PROCESSORS: 6,144
TDP: 295 W
ARCHITECTURE: RDNA 3.0
MSRP: $3999

▸ AI CAPABILITY

345/ 449 models @ Q4

With 48 GB VRAM and 864 GB/s bandwidth, this GPU handles models up to 65.2B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~99 tok/s.

RENT IT — LIVE PRICES

Checking what the clouds are charging…

§ 01TOP MODELS FOR RADEON PRO W7900

345 FIT · SHOWING 20

MODEL	SIZE	VRAM Q4	TOK/S	AVG
LLaMA 1 65B	65.2B	40.3 GB	12	42.6
Jamba 2 Mini	52B	32.3 GB	64	9.2
Jamba 1.5 Mini 52B	51.6B	32.0 GB	64	24.2
Kimi-Linear-48B-A3B	48B	29.8 GB	256	26.6
Nemotron-H 47B	47B	29.2 GB	16	84.6
Mixtral-8x7B	46.7B	29.0 GB	59	18.8
Nous-Hermes-2-Mixtral-8x7B-DPO	46.7B	29.0 GB	59	27.4
Dolphin 2.6 Mixtral 8x7B	46.7B	29.0 GB	59	23.8
Phi-3.5 MoE 42B	41.9B	26.1 GB	116	56.7
Falcon 40B	40B	24.9 GB	19	20.9
InternVL3 38B	38B	23.7 GB	20	78.9
Seed-OSS 36B Instruct	36B	22.5 GB	21	54.4
c4ai-command-r-v01 35B	35B	21.9 GB	22	27.5
Qwen 3.5 35B A3B	35B	21.9 GB	256	53.3
Qwen 3.6 35B A3B	35B	21.9 GB	256	53.9
Nous Capybara 34B	34.4B	21.5 GB	22	42.0
Yi-1.5 34B	34.4B	21.5 GB	22	45.3
Falcon-H1 34B	34B	21.3 GB	23	66.1
CodeLlama 34B	34B	21.3 GB	23	25.4
Nous Hermes 2 34B	34B	21.3 GB	23	47.0