Will It Run AI
CalculatorModelsHardwareCompare
Product
  • Calculator
  • Compare
  • Tier List
Browse
  • Models
  • Hardware
  • Docs
About
  • Why It Works
  • What's New
  • Legal Notice
  • Privacy Policy

All estimates are approximations based on mathematical models and public specifications. Actual performance may vary. Do not make purchasing decisions based solely on these estimates.

Data sourced from Hugging Face, Ollama, and official model documentation. Model names and logos are trademarks of their respective owners.

© 2026 Will It Run AI — Fase Consulting Ibiza, S.L. (NIF: B57969656)

Home/Hardware/GPUs/AMD Instinct MI210 64GB
AMD

AMD

AMD Instinct MI210 64GB

InstinctDatacenterCDNA 2PCIe 4ROCm
64GB
VRAM
1.6kGB/s
Bandwidth
181TFLOPS
FP16 Compute
362TOPS
INT8 Inference
$10,000 MSRP
VRAM64 GBBandwidth1.6k GB/sCompute181 TFInference362 TOPSValue1.81 TF/$k
AMD Instinct MI210 64GBCategory AvgNVIDIA A100 80GB

Specifications

Compute
FP16181 TFLOPS
INT8362 TOPS
ArchitectureCDNA 2
Memory
VRAM64 GB
Bandwidth1638 GB/s
General
FamilyInstinct
SegmentDatacenter
InterconnectPCIe 4
Compute PlatformROCM
MSRP$10,000

Architecture

CDNA 2

CDNA 2 powers the Instinct MI210 and MI250/MI250X accelerators. It introduced multi-die packaging with up to 128 GB HBM2e and Infinity Fabric for die-to-die communication.

AI Relevance

With up to 128 GB HBM2e memory and strong ROCm support, CDNA 2 GPUs can host large language models. The MI250X was used in the Frontier exascale supercomputer and supports major AI frameworks.

Process: TSMC 6nmPlatform: ROCMPrecisions: FP64, FP32, TF32, FP16, BF16, INT8

Recommendations by Workload

Agentic Coding

C

Qwen3-Coder-Next

This model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It should run, but memory headroom will be limited. Known channels: huggingface, ollama, lm-studio.

Decode 69.2 tok/s · 36K ctx · llama.cpp
57.0 GB / 64.0 GB VRAM

Chat

C

Qwen 3 32B

This model is a direct match for chat. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 57.1 tok/s · 17K ctx · llama.cpp
29.3 GB / 64.0 GB VRAM

Coding

C

Qwen3-Coder-Next

This model is a direct match for coding. It belongs to a current frontier family for local AI. It should run, but memory headroom will be limited. Known channels: huggingface, ollama, lm-studio.

Decode 69.2 tok/s · 18K ctx · llama.cpp
56.9 GB / 64.0 GB VRAM

RAG

C

Command R 35B

This model is a direct match for rag. It sits in the middle of the current model mix. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 52.2 tok/s · 52K ctx · llama.cpp
39.6 GB / 64.0 GB VRAM

Reasoning

C

Qwen 3 32B

This model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.

Decode 57.1 tok/s · 32K ctx · llama.cpp
31.8 GB / 64.0 GB VRAM

Full Model Compatibility

AlibabaQwen3-Coder-Next
C53
80B56.9 GB69 tok/s18K ctx
moe
UnslothQwen3.5 35B A3B
C52
35B34.1 GB52 tok/s30K ctx
dense
Lmstudio-communityLQwen3.5 35B A3B
C52
35B34.1 GB52 tok/s30K ctx
dense
AlibabaQwen 2.5 Coder 32B
C51
32B31.8 GB57 tok/s32K ctx
dense
AlibabaQwen3-Coder 30B A3B Instruct
C51
30.5B26.7 GB155 tok/s38K ctx
moe
AlibabaQwen3-VL 30B A3B Instruct
C51
30B26.4 GB160 tok/s39K ctx
moe
UnslothQwen3.5 27B
C51
27B28.0 GB68 tok/s37K ctx
dense
Unslothgemma 3 27b it
C51
27B28.0 GB68 tok/s37K ctx
dense
MistralDevstral Small 2 24B Instruct
C50
24B25.7 GB76 tok/s40K ctx
dense
MetaLlama 3.3 70B
C50
70B60.9 GB26 tok/s17K ctx
dense
MistralDevstral Small 1.1
C50
24B25.7 GB76 tok/s40K ctx
dense
AlibabaQwen 2.5 72B
C50
72B62.5 GB25 tok/s16K ctx
dense
AlibabaQwen 2.5 VL 72B
C50
72B62.5 GB25 tok/s16K ctx
dense
MistralCodestral 2 25.08
C50
22B24.2 GB83 tok/s42K ctx
dense
UnslothQwen3.5 9B
C47
9B14.2 GB203 tok/s72K ctx
dense
HauhauCSHQwen3.5 9B Uncensored HauhauCS Aggressive
C47
9B14.2 GB203 tok/s72K ctx
dense
Lmstudio-communityLQwen3.5 9B
C47
9B14.2 GB203 tok/s72K ctx
dense
BartowskiBMeta Llama 3.1 8B Instruct
C47
8B13.4 GB228 tok/s76K ctx
dense
XtunerXllava llama 3 8b v1 1
C47
8B13.4 GB228 tok/s76K ctx
dense
UnslothDeepSeek R1 0528 Qwen3 8B
C47
8B13.4 GB228 tok/s76K ctx
dense
MaziyarPanahiMMeta Llama 3 8B Instruct
C47
8B13.4 GB228 tok/s76K ctx
dense
TheBlokeTLlama 2 7B Chat
C46
7B12.7 GB261 tok/s81K ctx
dense
TheBlokeTMistral 7B Instruct v0.2
C46
7B12.7 GB261 tok/s81K ctx
dense
MaziyarPanahiMMistral 7B Instruct v0.3
C46
7B12.7 GB261 tok/s81K ctx
dense
UnslothQwen3.5 4B
C46
4B10.5 GB457 tok/s97K ctx
dense
BartowskiBLlama 3.2 3B Instruct
C46
3B10.3 GB526 tok/s100K ctx
dense
Lmstudio-communityLgemma 3 4b it
C46
4B10.5 GB457 tok/s97K ctx
dense
BartowskiBgemma 2 2b it
C45
2B9.7 GB713 tok/s105K ctx
dense
QwenQwen2.5 3B Instruct
C45
3B9.9 GB609 tok/s103K ctx
dense
Googlegemma 2b
C45
2B9.3 GB913 tok/s110K ctx
dense
TheDrummerTGemmasutra Mini 2B v1
C45
2B9.3 GB913 tok/s110K ctx
dense
QwenQwen2.5 1.5B Instruct
C45
1.5B9.0 GB1114 tok/s114K ctx
dense
Hugging-quantsHLlama 3.2 1B Instruct Q8 0
C45
1B8.9 GB1170 tok/s115K ctx
dense
TheBlokeTTinyLlama 1.1B Chat v1.0
C45
1.1B8.8 GB1114 tok/s117K ctx
dense
Ggml-orgGSmolVLM 500M Instruct
C45
0.5B8.5 GB1170 tok/s120K ctx
dense
Ggml-orgGembeddinggemma 300M
C45
0.3B8.3 GB1170 tok/s123K ctx
dense
DeepSeekDeepSeek R1 671B
F0
671B422.4 GB8 tok/s4K ctx
moe
MistralDevstral 2 123B Instruct
F0
123B101.5 GB15 tok/s10K ctx
dense
Z.aiGLM-5
F0
744B467.4 GB7 tok/s4K ctx
moe
Moonshot AIKimi K2.5
F0
1000B622.3 GB6 tok/s4K ctx
moe
MistralMistral Large 3
F0
675B425.5 GB8 tok/s4K ctx
+1moe
MistralMistral Small 4 119B
F0
119B80.9 GB45 tok/s13K ctx
moe
AlibabaQwen3-Coder 480B A35B Instruct
F0
480B305.6 GB11 tok/s4K ctx
moe
UnslothQwen3.5 122B A10B
F0
122B86.1 GB17 tok/s12K ctx
dense
DeepSeekDeepSeek V3 671B
F0
671B422.4 GB8 tok/s4K ctx
moe
MistralMixtral 8x22B
F0
141B99.4 GB25 tok/s10K ctx
moe
AlibabaQwen 3 235B A22B
F0
235B154.1 GB21 tok/s7K ctx
moe
UnslothQwen3.5 397B A17B
F0
397B311.5 GB5 tok/s4K ctx
dense
MetaLlama 4 Maverick 17B 128E
F0
400B254.0 GB14 tok/s4K ctx
moe
CohereCommand A 111B
F0
111B92.4 GB17 tok/s11K ctx
dense

Just out of reach

Models you could run with an upgrade

High-quality models that need a bit more memory

DeepSeekDeepSeek R1 671B
671BTier 5Needs ~428.2 GB
MistralDevstral 2 123B Instruct
123BTier 5Needs ~120.8 GB
Runs on Mac Studio M3 Ultra 256GB
Z.aiGLM-5
744BTier 5Needs ~473.6 GB
Moonshot AIKimi K2.5
1000BTier 5Needs ~627.3 GB
MistralMistral Large 3
675BTier 5Needs ~431.9 GB

Upgrade paths

Upgrade from AMD Instinct MI210 64GB

See what you unlock with more powerful hardware

Upgrade options

Upgrade options

NVIDIANVIDIA A100 80GBNext step up
80 GB VRAM (+16)2039 GB/s (+401)
A
Unlocks Mistral Small 4 119B, Llama 4 Scout 17B 16E+53% faster avg

~$15,000 MSRP

AMDAMD Instinct MI250X 128GBAMD upgrade
128 GB VRAM (+64)3200 GB/s (+1562)
A
Unlocks Devstral 2 123B Instruct, Mistral Small 4 119B, Qwen3.5 122B A10B+7 more · +117% faster avg

~$15,000 MSRP

AMDAMD Instinct MI325X 256GBBiggest leap
256 GB VRAM (+192)6000 GB/s (+4362)
A
Unlocks Devstral 2 123B Instruct, Mistral Small 4 119B, Qwen3.5 122B A10B+14 more · +274% faster avg

 

AMDAMD Instinct MI350X 288GBBest value
288 GB VRAM (+224)8000 GB/s (+6362)
A
Unlocks Devstral 2 123B Instruct, Mistral Small 4 119B, Qwen3-Coder 480B A35B Instruct+15 more · +397% faster avg

~$8,000 MSRP

Compare this GPU